diff --git "a/parakeet-tdt_ctc-110m/AudioEncoder.mlmodelc/model.mil" "b/parakeet-tdt_ctc-110m/AudioEncoder.mlmodelc/model.mil" new file mode 100644--- /dev/null +++ "b/parakeet-tdt_ctc-110m/AudioEncoder.mlmodelc/model.mil" @@ -0,0 +1,4887 @@ +program(1.0) +[buildInfo = dict, tensor>({{"coremlc-component-MIL", "3404.16.1"}, {"coremlc-version", "3404.23.1"}})] +{ + func main(tensor input_1, tensor melspectrogram_features) { + tensor pos_emb_to_fp16 = const()[name = tensor("pos_emb_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64)))]; + tensor var_42_cast_fp16 = sub(x = pos_emb_to_fp16, y = input_1)[name = tensor("op_42_cast_fp16")]; + tensor obj_3_cast_fp16 = mul(x = var_42_cast_fp16, y = input_1)[name = tensor("obj_3_cast_fp16")]; + tensor input_1_pad_type_0 = const()[name = tensor("input_1_pad_type_0"), val = tensor("custom")]; + tensor input_1_pad_0 = const()[name = tensor("input_1_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_1_strides_0 = const()[name = tensor("input_1_strides_0"), val = tensor([2, 2])]; + tensor input_1_dilations_0 = const()[name = tensor("input_1_dilations_0"), val = tensor([1, 1])]; + tensor input_1_groups_0 = const()[name = tensor("input_1_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_0_weight_to_fp16 = const()[name = tensor("pre_encode_conv_0_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(384128)))]; + tensor pre_encode_conv_0_bias_to_fp16 = const()[name = tensor("pre_encode_conv_0_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(388800)))]; + tensor input_1_cast_fp16 = conv(bias = pre_encode_conv_0_bias_to_fp16, dilations = input_1_dilations_0, groups = input_1_groups_0, pad = input_1_pad_0, pad_type = input_1_pad_type_0, strides = input_1_strides_0, weight = pre_encode_conv_0_weight_to_fp16, x = melspectrogram_features)[name = tensor("input_1_cast_fp16")]; + tensor input_3_cast_fp16 = relu(x = input_1_cast_fp16)[name = tensor("input_3_cast_fp16")]; + tensor input_5_pad_type_0 = const()[name = tensor("input_5_pad_type_0"), val = tensor("custom")]; + tensor input_5_pad_0 = const()[name = tensor("input_5_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_5_strides_0 = const()[name = tensor("input_5_strides_0"), val = tensor([2, 2])]; + tensor input_5_groups_0 = const()[name = tensor("input_5_groups_0"), val = tensor(256)]; + tensor input_5_dilations_0 = const()[name = tensor("input_5_dilations_0"), val = tensor([1, 1])]; + tensor pre_encode_conv_2_weight_to_fp16 = const()[name = tensor("pre_encode_conv_2_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(389376)))]; + tensor pre_encode_conv_2_bias_to_fp16 = const()[name = tensor("pre_encode_conv_2_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394048)))]; + tensor input_5_cast_fp16 = conv(bias = pre_encode_conv_2_bias_to_fp16, dilations = input_5_dilations_0, groups = input_5_groups_0, pad = input_5_pad_0, pad_type = input_5_pad_type_0, strides = input_5_strides_0, weight = pre_encode_conv_2_weight_to_fp16, x = input_3_cast_fp16)[name = tensor("input_5_cast_fp16")]; + tensor input_7_pad_type_0 = const()[name = tensor("input_7_pad_type_0"), val = tensor("valid")]; + tensor input_7_strides_0 = const()[name = tensor("input_7_strides_0"), val = tensor([1, 1])]; + tensor input_7_pad_0 = const()[name = tensor("input_7_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_7_dilations_0 = const()[name = tensor("input_7_dilations_0"), val = tensor([1, 1])]; + tensor input_7_groups_0 = const()[name = tensor("input_7_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_3_weight_to_fp16 = const()[name = tensor("pre_encode_conv_3_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(394624)))]; + tensor pre_encode_conv_3_bias_to_fp16 = const()[name = tensor("pre_encode_conv_3_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(525760)))]; + tensor input_7_cast_fp16 = conv(bias = pre_encode_conv_3_bias_to_fp16, dilations = input_7_dilations_0, groups = input_7_groups_0, pad = input_7_pad_0, pad_type = input_7_pad_type_0, strides = input_7_strides_0, weight = pre_encode_conv_3_weight_to_fp16, x = input_5_cast_fp16)[name = tensor("input_7_cast_fp16")]; + tensor input_9_cast_fp16 = relu(x = input_7_cast_fp16)[name = tensor("input_9_cast_fp16")]; + tensor input_11_pad_type_0 = const()[name = tensor("input_11_pad_type_0"), val = tensor("custom")]; + tensor input_11_pad_0 = const()[name = tensor("input_11_pad_0"), val = tensor([1, 1, 1, 1])]; + tensor input_11_strides_0 = const()[name = tensor("input_11_strides_0"), val = tensor([2, 2])]; + tensor input_11_groups_0 = const()[name = tensor("input_11_groups_0"), val = tensor(256)]; + tensor input_11_dilations_0 = const()[name = tensor("input_11_dilations_0"), val = tensor([1, 1])]; + tensor pre_encode_conv_5_weight_to_fp16 = const()[name = tensor("pre_encode_conv_5_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(526336)))]; + tensor pre_encode_conv_5_bias_to_fp16 = const()[name = tensor("pre_encode_conv_5_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(531008)))]; + tensor input_11_cast_fp16 = conv(bias = pre_encode_conv_5_bias_to_fp16, dilations = input_11_dilations_0, groups = input_11_groups_0, pad = input_11_pad_0, pad_type = input_11_pad_type_0, strides = input_11_strides_0, weight = pre_encode_conv_5_weight_to_fp16, x = input_9_cast_fp16)[name = tensor("input_11_cast_fp16")]; + tensor input_13_pad_type_0 = const()[name = tensor("input_13_pad_type_0"), val = tensor("valid")]; + tensor input_13_strides_0 = const()[name = tensor("input_13_strides_0"), val = tensor([1, 1])]; + tensor input_13_pad_0 = const()[name = tensor("input_13_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor input_13_dilations_0 = const()[name = tensor("input_13_dilations_0"), val = tensor([1, 1])]; + tensor input_13_groups_0 = const()[name = tensor("input_13_groups_0"), val = tensor(1)]; + tensor pre_encode_conv_6_weight_to_fp16 = const()[name = tensor("pre_encode_conv_6_weight_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(531584)))]; + tensor pre_encode_conv_6_bias_to_fp16 = const()[name = tensor("pre_encode_conv_6_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(662720)))]; + tensor input_13_cast_fp16 = conv(bias = pre_encode_conv_6_bias_to_fp16, dilations = input_13_dilations_0, groups = input_13_groups_0, pad = input_13_pad_0, pad_type = input_13_pad_type_0, strides = input_13_strides_0, weight = pre_encode_conv_6_weight_to_fp16, x = input_11_cast_fp16)[name = tensor("input_13_cast_fp16")]; + tensor x_1_cast_fp16 = relu(x = input_13_cast_fp16)[name = tensor("x_1_cast_fp16")]; + tensor var_103_perm_0 = const()[name = tensor("op_103_perm_0"), val = tensor([0, 1, 3, 2])]; + tensor var_106 = const()[name = tensor("op_106"), val = tensor([1, 2560, 1, 188])]; + tensor var_103_cast_fp16 = transpose(perm = var_103_perm_0, x = x_1_cast_fp16)[name = tensor("transpose_0")]; + tensor input_15_cast_fp16 = reshape(shape = var_106, x = var_103_cast_fp16)[name = tensor("input_15_cast_fp16")]; + tensor var_116_pad_type_0 = const()[name = tensor("op_116_pad_type_0"), val = tensor("valid")]; + tensor var_116_strides_0 = const()[name = tensor("op_116_strides_0"), val = tensor([1, 1])]; + tensor var_116_pad_0 = const()[name = tensor("op_116_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_116_dilations_0 = const()[name = tensor("op_116_dilations_0"), val = tensor([1, 1])]; + tensor var_116_groups_0 = const()[name = tensor("op_116_groups_0"), val = tensor(1)]; + tensor pre_encode_out_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(663296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1646400))), name = tensor("pre_encode_out_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2560, 1, 1])]; + tensor pre_encode_out_inlier_module_bias_to_fp16 = const()[name = tensor("pre_encode_out_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1646592)))]; + tensor var_116_cast_fp16 = conv(bias = pre_encode_out_inlier_module_bias_to_fp16, dilations = var_116_dilations_0, groups = var_116_groups_0, pad = var_116_pad_0, pad_type = var_116_pad_type_0, strides = var_116_strides_0, weight = pre_encode_out_inlier_module_weight_to_fp16_palettized, x = input_15_cast_fp16)[name = tensor("op_116_cast_fp16")]; + tensor var_122_pad_type_0 = const()[name = tensor("op_122_pad_type_0"), val = tensor("valid")]; + tensor var_122_strides_0 = const()[name = tensor("op_122_strides_0"), val = tensor([1, 1])]; + tensor var_122_pad_0 = const()[name = tensor("op_122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_122_dilations_0 = const()[name = tensor("op_122_dilations_0"), val = tensor([1, 1])]; + tensor var_122_groups_0 = const()[name = tensor("op_122_groups_0"), val = tensor(1)]; + tensor pre_encode_out_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1696256))), name = tensor("pre_encode_out_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1647680))), shape = tensor([512, 2560, 1, 1])]; + tensor var_122_cast_fp16 = conv(dilations = var_122_dilations_0, groups = var_122_groups_0, pad = var_122_pad_0, pad_type = var_122_pad_type_0, strides = var_122_strides_0, weight = pre_encode_out_outlier_module_weight_to_fp16_sparsified, x = input_15_cast_fp16)[name = tensor("op_122_cast_fp16")]; + tensor inputs_1_cast_fp16 = add(x = var_116_cast_fp16, y = var_122_cast_fp16)[name = tensor("inputs_1_cast_fp16")]; + tensor var_128 = const()[name = tensor("op_128"), val = tensor(3)]; + tensor out_1_axes_0 = const()[name = tensor("out_1_axes_0"), val = tensor([1])]; + tensor var_159_to_fp16 = const()[name = tensor("op_159_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_1_cast_fp16 = layer_norm(axes = out_1_axes_0, epsilon = var_159_to_fp16, x = inputs_1_cast_fp16)[name = tensor("out_1_cast_fp16")]; + tensor input_17_mean_0_to_fp16 = const()[name = tensor("input_17_mean_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1860160)))]; + tensor input_17_variance_0_to_fp16 = const()[name = tensor("input_17_variance_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1861248)))]; + tensor input_17_gamma_0_to_fp16 = const()[name = tensor("input_17_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1862336)))]; + tensor input_17_beta_0_to_fp16 = const()[name = tensor("input_17_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1863424)))]; + tensor input_17_epsilon_0_to_fp16 = const()[name = tensor("input_17_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_17_cast_fp16 = batch_norm(beta = input_17_beta_0_to_fp16, epsilon = input_17_epsilon_0_to_fp16, gamma = input_17_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_1_cast_fp16)[name = tensor("input_17_cast_fp16")]; + tensor var_179_pad_type_0 = const()[name = tensor("op_179_pad_type_0"), val = tensor("valid")]; + tensor var_179_strides_0 = const()[name = tensor("op_179_strides_0"), val = tensor([1, 1])]; + tensor var_179_pad_0 = const()[name = tensor("op_179_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_179_dilations_0 = const()[name = tensor("op_179_dilations_0"), val = tensor([1, 1])]; + tensor var_179_groups_0 = const()[name = tensor("op_179_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(1864512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2651008))), name = tensor("layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2651200)))]; + tensor var_179_cast_fp16 = conv(bias = layers_0_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_179_dilations_0, groups = var_179_groups_0, pad = var_179_pad_0, pad_type = var_179_pad_type_0, strides = var_179_strides_0, weight = layers_0_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_17_cast_fp16)[name = tensor("op_179_cast_fp16")]; + tensor var_185_pad_type_0 = const()[name = tensor("op_185_pad_type_0"), val = tensor("valid")]; + tensor var_185_strides_0 = const()[name = tensor("op_185_strides_0"), val = tensor([1, 1])]; + tensor var_185_pad_0 = const()[name = tensor("op_185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_185_dilations_0 = const()[name = tensor("op_185_dilations_0"), val = tensor([1, 1])]; + tensor var_185_groups_0 = const()[name = tensor("op_185_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2688192))), name = tensor("layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2655360))), shape = tensor([2048, 512, 1, 1])]; + tensor var_185_cast_fp16 = conv(dilations = var_185_dilations_0, groups = var_185_groups_0, pad = var_185_pad_0, pad_type = var_185_pad_type_0, strides = var_185_strides_0, weight = layers_0_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_17_cast_fp16)[name = tensor("op_185_cast_fp16")]; + tensor input_19_cast_fp16 = add(x = var_179_cast_fp16, y = var_185_cast_fp16)[name = tensor("input_19_cast_fp16")]; + tensor input_21_cast_fp16 = silu(x = input_19_cast_fp16)[name = tensor("input_21_cast_fp16")]; + tensor var_196_pad_type_0 = const()[name = tensor("op_196_pad_type_0"), val = tensor("valid")]; + tensor var_196_strides_0 = const()[name = tensor("op_196_strides_0"), val = tensor([1, 1])]; + tensor var_196_pad_0 = const()[name = tensor("op_196_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_196_dilations_0 = const()[name = tensor("op_196_dilations_0"), val = tensor([1, 1])]; + tensor var_196_groups_0 = const()[name = tensor("op_196_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(2819328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3605824))), name = tensor("layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_0_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3606016)))]; + tensor var_196_cast_fp16 = conv(bias = layers_0_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_196_dilations_0, groups = var_196_groups_0, pad = var_196_pad_0, pad_type = var_196_pad_type_0, strides = var_196_strides_0, weight = layers_0_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_21_cast_fp16)[name = tensor("op_196_cast_fp16")]; + tensor var_202_pad_type_0 = const()[name = tensor("op_202_pad_type_0"), val = tensor("valid")]; + tensor var_202_strides_0 = const()[name = tensor("op_202_strides_0"), val = tensor([1, 1])]; + tensor var_202_pad_0 = const()[name = tensor("op_202_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_202_dilations_0 = const()[name = tensor("op_202_dilations_0"), val = tensor([1, 1])]; + tensor var_202_groups_0 = const()[name = tensor("op_202_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3635200))), name = tensor("layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3607104))), shape = tensor([512, 2048, 1, 1])]; + tensor var_202_cast_fp16 = conv(dilations = var_202_dilations_0, groups = var_202_groups_0, pad = var_202_pad_0, pad_type = var_202_pad_type_0, strides = var_202_strides_0, weight = layers_0_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_21_cast_fp16)[name = tensor("op_202_cast_fp16")]; + tensor x_3_cast_fp16 = add(x = var_196_cast_fp16, y = var_202_cast_fp16)[name = tensor("x_3_cast_fp16")]; + tensor var_204_to_fp16 = const()[name = tensor("op_204_to_fp16"), val = tensor(0x1p-1)]; + tensor var_205_cast_fp16 = mul(x = x_3_cast_fp16, y = var_204_to_fp16)[name = tensor("op_205_cast_fp16")]; + tensor inputs_3_cast_fp16 = add(x = inputs_1_cast_fp16, y = var_205_cast_fp16)[name = tensor("inputs_3_cast_fp16")]; + tensor out_3_axes_0 = const()[name = tensor("out_3_axes_0"), val = tensor([1])]; + tensor var_215_to_fp16 = const()[name = tensor("op_215_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_3_cast_fp16 = layer_norm(axes = out_3_axes_0, epsilon = var_215_to_fp16, x = inputs_3_cast_fp16)[name = tensor("out_3_cast_fp16")]; + tensor obj_1_gamma_0_to_fp16 = const()[name = tensor("obj_1_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3766336)))]; + tensor obj_1_beta_0_to_fp16 = const()[name = tensor("obj_1_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3767424)))]; + tensor obj_1_epsilon_0_to_fp16 = const()[name = tensor("obj_1_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_1_cast_fp16 = batch_norm(beta = obj_1_beta_0_to_fp16, epsilon = obj_1_epsilon_0_to_fp16, gamma = obj_1_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_3_cast_fp16)[name = tensor("obj_1_cast_fp16")]; + tensor var_240_pad_type_0 = const()[name = tensor("op_240_pad_type_0"), val = tensor("valid")]; + tensor var_240_strides_0 = const()[name = tensor("op_240_strides_0"), val = tensor([1, 1])]; + tensor var_240_pad_0 = const()[name = tensor("op_240_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_240_dilations_0 = const()[name = tensor("op_240_dilations_0"), val = tensor([1, 1])]; + tensor var_240_groups_0 = const()[name = tensor("op_240_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3768512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3965184))), name = tensor("layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_0_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3965376)))]; + tensor var_240_cast_fp16 = conv(bias = layers_0_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_240_dilations_0, groups = var_240_groups_0, pad = var_240_pad_0, pad_type = var_240_pad_type_0, strides = var_240_strides_0, weight = layers_0_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_240_cast_fp16")]; + tensor var_246_pad_type_0 = const()[name = tensor("op_246_pad_type_0"), val = tensor("valid")]; + tensor var_246_strides_0 = const()[name = tensor("op_246_strides_0"), val = tensor([1, 1])]; + tensor var_246_pad_0 = const()[name = tensor("op_246_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_246_dilations_0 = const()[name = tensor("op_246_dilations_0"), val = tensor([1, 1])]; + tensor var_246_groups_0 = const()[name = tensor("op_246_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3977984))), name = tensor("layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(3966464))), shape = tensor([512, 512, 1, 1])]; + tensor var_246_cast_fp16 = conv(dilations = var_246_dilations_0, groups = var_246_groups_0, pad = var_246_pad_0, pad_type = var_246_pad_type_0, strides = var_246_strides_0, weight = layers_0_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_246_cast_fp16")]; + tensor query_1_cast_fp16 = add(x = var_240_cast_fp16, y = var_246_cast_fp16)[name = tensor("query_1_cast_fp16")]; + tensor var_255_pad_type_0 = const()[name = tensor("op_255_pad_type_0"), val = tensor("valid")]; + tensor var_255_strides_0 = const()[name = tensor("op_255_strides_0"), val = tensor([1, 1])]; + tensor var_255_pad_0 = const()[name = tensor("op_255_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_255_dilations_0 = const()[name = tensor("op_255_dilations_0"), val = tensor([1, 1])]; + tensor var_255_groups_0 = const()[name = tensor("op_255_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4010816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4207488))), name = tensor("layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_255_cast_fp16 = conv(dilations = var_255_dilations_0, groups = var_255_groups_0, pad = var_255_pad_0, pad_type = var_255_pad_type_0, strides = var_255_strides_0, weight = layers_0_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_255_cast_fp16")]; + tensor var_261_pad_type_0 = const()[name = tensor("op_261_pad_type_0"), val = tensor("valid")]; + tensor var_261_strides_0 = const()[name = tensor("op_261_strides_0"), val = tensor([1, 1])]; + tensor var_261_pad_0 = const()[name = tensor("op_261_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_261_dilations_0 = const()[name = tensor("op_261_dilations_0"), val = tensor([1, 1])]; + tensor var_261_groups_0 = const()[name = tensor("op_261_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4220160))), name = tensor("layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4207680))), shape = tensor([512, 512, 1, 1])]; + tensor var_261_cast_fp16 = conv(dilations = var_261_dilations_0, groups = var_261_groups_0, pad = var_261_pad_0, pad_type = var_261_pad_type_0, strides = var_261_strides_0, weight = layers_0_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_261_cast_fp16")]; + tensor key_1_cast_fp16 = add(x = var_255_cast_fp16, y = var_261_cast_fp16)[name = tensor("key_1_cast_fp16")]; + tensor var_271_pad_type_0 = const()[name = tensor("op_271_pad_type_0"), val = tensor("valid")]; + tensor var_271_strides_0 = const()[name = tensor("op_271_strides_0"), val = tensor([1, 1])]; + tensor var_271_pad_0 = const()[name = tensor("op_271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_271_dilations_0 = const()[name = tensor("op_271_dilations_0"), val = tensor([1, 1])]; + tensor var_271_groups_0 = const()[name = tensor("op_271_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4252992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4449664))), name = tensor("layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_0_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4449856)))]; + tensor var_271_cast_fp16 = conv(bias = layers_0_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_271_dilations_0, groups = var_271_groups_0, pad = var_271_pad_0, pad_type = var_271_pad_type_0, strides = var_271_strides_0, weight = layers_0_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_1_cast_fp16)[name = tensor("op_271_cast_fp16")]; + tensor var_277_pad_type_0 = const()[name = tensor("op_277_pad_type_0"), val = tensor("valid")]; + tensor var_277_strides_0 = const()[name = tensor("op_277_strides_0"), val = tensor([1, 1])]; + tensor var_277_pad_0 = const()[name = tensor("op_277_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_277_dilations_0 = const()[name = tensor("op_277_dilations_0"), val = tensor([1, 1])]; + tensor var_277_groups_0 = const()[name = tensor("op_277_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4461312))), name = tensor("layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4450944))), shape = tensor([512, 512, 1, 1])]; + tensor var_277_cast_fp16 = conv(dilations = var_277_dilations_0, groups = var_277_groups_0, pad = var_277_pad_0, pad_type = var_277_pad_type_0, strides = var_277_strides_0, weight = layers_0_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_1_cast_fp16)[name = tensor("op_277_cast_fp16")]; + tensor value_1_cast_fp16 = add(x = var_271_cast_fp16, y = var_277_cast_fp16)[name = tensor("value_1_cast_fp16")]; + tensor var_280_to_fp16 = const()[name = tensor("op_280_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4494144)))]; + tensor query_3_cast_fp16 = add(x = query_1_cast_fp16, y = var_280_to_fp16)[name = tensor("query_3_cast_fp16")]; + tensor var_283_to_fp16 = const()[name = tensor("op_283_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4495232)))]; + tensor q_with_bias_v_1_cast_fp16 = add(x = query_1_cast_fp16, y = var_283_to_fp16)[name = tensor("q_with_bias_v_1_cast_fp16")]; + tensor var_293_pad_type_0 = const()[name = tensor("op_293_pad_type_0"), val = tensor("valid")]; + tensor var_293_strides_0 = const()[name = tensor("op_293_strides_0"), val = tensor([1, 1])]; + tensor var_293_pad_0 = const()[name = tensor("op_293_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_293_dilations_0 = const()[name = tensor("op_293_dilations_0"), val = tensor([1, 1])]; + tensor var_293_groups_0 = const()[name = tensor("op_293_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4496320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4692992))), name = tensor("layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_293_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_293_dilations_0, groups = var_293_groups_0, pad = var_293_pad_0, pad_type = var_293_pad_type_0, strides = var_293_strides_0, weight = layers_0_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_293_cast_fp16")]; + tensor var_299_pad_type_0 = const()[name = tensor("op_299_pad_type_0"), val = tensor("valid")]; + tensor var_299_strides_0 = const()[name = tensor("op_299_strides_0"), val = tensor([1, 1])]; + tensor var_299_pad_0 = const()[name = tensor("op_299_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_299_dilations_0 = const()[name = tensor("op_299_dilations_0"), val = tensor([1, 1])]; + tensor var_299_groups_0 = const()[name = tensor("op_299_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4712704))), name = tensor("layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4693184))), shape = tensor([512, 512, 1, 1])]; + tensor var_299_cast_fp16 = conv(dilations = var_299_dilations_0, groups = var_299_groups_0, pad = var_299_pad_0, pad_type = var_299_pad_type_0, strides = var_299_strides_0, weight = layers_0_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_299_cast_fp16")]; + tensor p_1_cast_fp16 = add(x = var_293_cast_fp16, y = var_299_cast_fp16)[name = tensor("p_1_cast_fp16")]; + tensor var_303 = const()[name = tensor("op_303"), val = tensor([1, 8, 64, 188])]; + tensor var_304_cast_fp16 = reshape(shape = var_303, x = q_with_bias_v_1_cast_fp16)[name = tensor("op_304_cast_fp16")]; + tensor var_305 = const()[name = tensor("op_305"), val = tensor([1, 8, 64, -1])]; + tensor var_306_cast_fp16 = reshape(shape = var_305, x = p_1_cast_fp16)[name = tensor("op_306_cast_fp16")]; + tensor matrix_bd_1_transpose_x_0 = const()[name = tensor("matrix_bd_1_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_1_transpose_y_0 = const()[name = tensor("matrix_bd_1_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_1_cast_fp16 = matmul(transpose_x = matrix_bd_1_transpose_x_0, transpose_y = matrix_bd_1_transpose_y_0, x = var_304_cast_fp16, y = var_306_cast_fp16)[name = tensor("matrix_bd_1_cast_fp16")]; + tensor matrix_bd_3_pad_0 = const()[name = tensor("matrix_bd_3_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_3_mode_0 = const()[name = tensor("matrix_bd_3_mode_0"), val = tensor("constant")]; + tensor const_10_to_fp16 = const()[name = tensor("const_10_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_3_cast_fp16 = pad(constant_val = const_10_to_fp16, mode = matrix_bd_3_mode_0, pad = matrix_bd_3_pad_0, x = matrix_bd_1_cast_fp16)[name = tensor("matrix_bd_3_cast_fp16")]; + tensor var_315 = const()[name = tensor("op_315"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_5_cast_fp16 = reshape(shape = var_315, x = matrix_bd_3_cast_fp16)[name = tensor("matrix_bd_5_cast_fp16")]; + tensor var_319_begin_0 = const()[name = tensor("op_319_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_319_end_0 = const()[name = tensor("op_319_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_319_end_mask_0 = const()[name = tensor("op_319_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_319_cast_fp16 = slice_by_index(begin = var_319_begin_0, end = var_319_end_0, end_mask = var_319_end_mask_0, x = matrix_bd_5_cast_fp16)[name = tensor("op_319_cast_fp16")]; + tensor var_320 = const()[name = tensor("op_320"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_7_cast_fp16 = reshape(shape = var_320, x = var_319_cast_fp16)[name = tensor("matrix_bd_7_cast_fp16")]; + tensor var_325_begin_0 = const()[name = tensor("op_325_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_325_end_0 = const()[name = tensor("op_325_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_325_end_mask_0 = const()[name = tensor("op_325_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_325_cast_fp16 = slice_by_index(begin = var_325_begin_0, end = var_325_end_0, end_mask = var_325_end_mask_0, x = matrix_bd_7_cast_fp16)[name = tensor("op_325_cast_fp16")]; + tensor var_326_to_fp16 = const()[name = tensor("op_326_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_1_cast_fp16 = mul(x = var_325_cast_fp16, y = var_326_to_fp16)[name = tensor("qk_mask_1_cast_fp16")]; + tensor var_330 = const()[name = tensor("op_330"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_1_cast_fp16 = reshape(shape = var_330, x = query_3_cast_fp16)[name = tensor("mh_q_1_cast_fp16")]; + tensor var_332_to_fp16 = const()[name = tensor("op_332_to_fp16"), val = tensor(0x1p-3)]; + tensor var_333_cast_fp16 = mul(x = mh_q_1_cast_fp16, y = var_332_to_fp16)[name = tensor("op_333_cast_fp16")]; + tensor var_336 = const()[name = tensor("op_336"), val = tensor([1, 8, 64, 188])]; + tensor var_337_cast_fp16 = reshape(shape = var_336, x = key_1_cast_fp16)[name = tensor("op_337_cast_fp16")]; + tensor mh_w_1_transpose_x_0 = const()[name = tensor("mh_w_1_transpose_x_0"), val = tensor(true)]; + tensor mh_w_1_transpose_y_0 = const()[name = tensor("mh_w_1_transpose_y_0"), val = tensor(false)]; + tensor mh_w_1_cast_fp16 = matmul(transpose_x = mh_w_1_transpose_x_0, transpose_y = mh_w_1_transpose_y_0, x = var_333_cast_fp16, y = var_337_cast_fp16)[name = tensor("mh_w_1_cast_fp16")]; + tensor mh_w_3_cast_fp16 = add(x = mh_w_1_cast_fp16, y = qk_mask_1_cast_fp16)[name = tensor("mh_w_3_cast_fp16")]; + tensor var_341_cast_fp16 = softmax(axis = var_128, x = mh_w_3_cast_fp16)[name = tensor("op_341_cast_fp16")]; + tensor var_342 = const()[name = tensor("op_342"), val = tensor([1, 8, 64, 188])]; + tensor var_343_cast_fp16 = reshape(shape = var_342, x = value_1_cast_fp16)[name = tensor("op_343_cast_fp16")]; + tensor attn_1_transpose_x_0 = const()[name = tensor("attn_1_transpose_x_0"), val = tensor(false)]; + tensor attn_1_transpose_y_0 = const()[name = tensor("attn_1_transpose_y_0"), val = tensor(true)]; + tensor attn_1_cast_fp16 = matmul(transpose_x = attn_1_transpose_x_0, transpose_y = attn_1_transpose_y_0, x = var_343_cast_fp16, y = var_341_cast_fp16)[name = tensor("attn_1_cast_fp16")]; + tensor var_346 = const()[name = tensor("op_346"), val = tensor([1, 512, 1, 188])]; + tensor input_23_cast_fp16 = reshape(shape = var_346, x = attn_1_cast_fp16)[name = tensor("input_23_cast_fp16")]; + tensor var_356_pad_type_0 = const()[name = tensor("op_356_pad_type_0"), val = tensor("valid")]; + tensor var_356_strides_0 = const()[name = tensor("op_356_strides_0"), val = tensor([1, 1])]; + tensor var_356_pad_0 = const()[name = tensor("op_356_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_356_dilations_0 = const()[name = tensor("op_356_dilations_0"), val = tensor([1, 1])]; + tensor var_356_groups_0 = const()[name = tensor("op_356_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4745536))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4942208))), name = tensor("layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_0_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4942400)))]; + tensor var_356_cast_fp16 = conv(bias = layers_0_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_356_dilations_0, groups = var_356_groups_0, pad = var_356_pad_0, pad_type = var_356_pad_type_0, strides = var_356_strides_0, weight = layers_0_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_23_cast_fp16)[name = tensor("op_356_cast_fp16")]; + tensor var_362_pad_type_0 = const()[name = tensor("op_362_pad_type_0"), val = tensor("valid")]; + tensor var_362_strides_0 = const()[name = tensor("op_362_strides_0"), val = tensor([1, 1])]; + tensor var_362_pad_0 = const()[name = tensor("op_362_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_362_dilations_0 = const()[name = tensor("op_362_dilations_0"), val = tensor([1, 1])]; + tensor var_362_groups_0 = const()[name = tensor("op_362_groups_0"), val = tensor(1)]; + tensor layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4952448))), name = tensor("layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4943488))), shape = tensor([512, 512, 1, 1])]; + tensor var_362_cast_fp16 = conv(dilations = var_362_dilations_0, groups = var_362_groups_0, pad = var_362_pad_0, pad_type = var_362_pad_type_0, strides = var_362_strides_0, weight = layers_0_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_23_cast_fp16)[name = tensor("op_362_cast_fp16")]; + tensor obj_5_cast_fp16 = add(x = var_356_cast_fp16, y = var_362_cast_fp16)[name = tensor("obj_5_cast_fp16")]; + tensor inputs_5_cast_fp16 = add(x = inputs_3_cast_fp16, y = obj_5_cast_fp16)[name = tensor("inputs_5_cast_fp16")]; + tensor out_5_axes_0 = const()[name = tensor("out_5_axes_0"), val = tensor([1])]; + tensor var_373_to_fp16 = const()[name = tensor("op_373_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_5_cast_fp16 = layer_norm(axes = out_5_axes_0, epsilon = var_373_to_fp16, x = inputs_5_cast_fp16)[name = tensor("out_5_cast_fp16")]; + tensor input_25_gamma_0_to_fp16 = const()[name = tensor("input_25_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4985280)))]; + tensor input_25_beta_0_to_fp16 = const()[name = tensor("input_25_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4986368)))]; + tensor input_25_epsilon_0_to_fp16 = const()[name = tensor("input_25_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_25_cast_fp16 = batch_norm(beta = input_25_beta_0_to_fp16, epsilon = input_25_epsilon_0_to_fp16, gamma = input_25_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_5_cast_fp16)[name = tensor("input_25_cast_fp16")]; + tensor var_395_pad_type_0 = const()[name = tensor("op_395_pad_type_0"), val = tensor("valid")]; + tensor var_395_strides_0 = const()[name = tensor("op_395_strides_0"), val = tensor([1, 1])]; + tensor var_395_pad_0 = const()[name = tensor("op_395_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_395_dilations_0 = const()[name = tensor("op_395_dilations_0"), val = tensor([1, 1])]; + tensor var_395_groups_0 = const()[name = tensor("op_395_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(4987456))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5380736))), name = tensor("layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_0_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5380928)))]; + tensor var_395_cast_fp16 = conv(bias = layers_0_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_395_dilations_0, groups = var_395_groups_0, pad = var_395_pad_0, pad_type = var_395_pad_type_0, strides = var_395_strides_0, weight = layers_0_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_25_cast_fp16)[name = tensor("op_395_cast_fp16")]; + tensor var_401_pad_type_0 = const()[name = tensor("op_401_pad_type_0"), val = tensor("valid")]; + tensor var_401_strides_0 = const()[name = tensor("op_401_strides_0"), val = tensor([1, 1])]; + tensor var_401_pad_0 = const()[name = tensor("op_401_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_401_dilations_0 = const()[name = tensor("op_401_dilations_0"), val = tensor([1, 1])]; + tensor var_401_groups_0 = const()[name = tensor("op_401_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5401344))), name = tensor("layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5383040))), shape = tensor([1024, 512, 1, 1])]; + tensor var_401_cast_fp16 = conv(dilations = var_401_dilations_0, groups = var_401_groups_0, pad = var_401_pad_0, pad_type = var_401_pad_type_0, strides = var_401_strides_0, weight = layers_0_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_25_cast_fp16)[name = tensor("op_401_cast_fp16")]; + tensor input_27_cast_fp16 = add(x = var_395_cast_fp16, y = var_401_cast_fp16)[name = tensor("input_27_cast_fp16")]; + tensor input_29_split_num_splits_0 = const()[name = tensor("input_29_split_num_splits_0"), val = tensor(2)]; + tensor input_29_split_axis_0 = const()[name = tensor("input_29_split_axis_0"), val = tensor(1)]; + tensor input_29_split_cast_fp16_0, tensor input_29_split_cast_fp16_1 = split(axis = input_29_split_axis_0, num_splits = input_29_split_num_splits_0, x = input_27_cast_fp16)[name = tensor("input_29_split_cast_fp16")]; + tensor input_29_split_1_sigmoid_cast_fp16 = sigmoid(x = input_29_split_cast_fp16_1)[name = tensor("input_29_split_1_sigmoid_cast_fp16")]; + tensor input_29_cast_fp16 = mul(x = input_29_split_cast_fp16_0, y = input_29_split_1_sigmoid_cast_fp16)[name = tensor("input_29_cast_fp16")]; + tensor input_31_pad_type_0 = const()[name = tensor("input_31_pad_type_0"), val = tensor("custom")]; + tensor input_31_pad_0 = const()[name = tensor("input_31_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_31_groups_0 = const()[name = tensor("input_31_groups_0"), val = tensor(512)]; + tensor input_31_strides_0 = const()[name = tensor("input_31_strides_0"), val = tensor([1, 1])]; + tensor input_31_dilations_0 = const()[name = tensor("input_31_dilations_0"), val = tensor([1, 1])]; + tensor const_191_to_fp16 = const()[name = tensor("const_191_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5466944)))]; + tensor const_192_to_fp16 = const()[name = tensor("const_192_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5476224)))]; + tensor input_33_cast_fp16 = conv(bias = const_192_to_fp16, dilations = input_31_dilations_0, groups = input_31_groups_0, pad = input_31_pad_0, pad_type = input_31_pad_type_0, strides = input_31_strides_0, weight = const_191_to_fp16, x = input_29_cast_fp16)[name = tensor("input_33_cast_fp16")]; + tensor input_35_cast_fp16 = silu(x = input_33_cast_fp16)[name = tensor("input_35_cast_fp16")]; + tensor var_425_pad_type_0 = const()[name = tensor("op_425_pad_type_0"), val = tensor("valid")]; + tensor var_425_strides_0 = const()[name = tensor("op_425_strides_0"), val = tensor([1, 1])]; + tensor var_425_pad_0 = const()[name = tensor("op_425_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_425_dilations_0 = const()[name = tensor("op_425_dilations_0"), val = tensor([1, 1])]; + tensor var_425_groups_0 = const()[name = tensor("op_425_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5477312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5673984))), name = tensor("layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_0_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5674176)))]; + tensor var_425_cast_fp16 = conv(bias = layers_0_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_425_dilations_0, groups = var_425_groups_0, pad = var_425_pad_0, pad_type = var_425_pad_type_0, strides = var_425_strides_0, weight = layers_0_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_35_cast_fp16)[name = tensor("op_425_cast_fp16")]; + tensor var_431_pad_type_0 = const()[name = tensor("op_431_pad_type_0"), val = tensor("valid")]; + tensor var_431_strides_0 = const()[name = tensor("op_431_strides_0"), val = tensor([1, 1])]; + tensor var_431_pad_0 = const()[name = tensor("op_431_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_431_dilations_0 = const()[name = tensor("op_431_dilations_0"), val = tensor([1, 1])]; + tensor var_431_groups_0 = const()[name = tensor("op_431_groups_0"), val = tensor(1)]; + tensor layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5683840))), name = tensor("layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5675264))), shape = tensor([512, 512, 1, 1])]; + tensor var_431_cast_fp16 = conv(dilations = var_431_dilations_0, groups = var_431_groups_0, pad = var_431_pad_0, pad_type = var_431_pad_type_0, strides = var_431_strides_0, weight = layers_0_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_35_cast_fp16)[name = tensor("op_431_cast_fp16")]; + tensor x_5_cast_fp16 = add(x = var_425_cast_fp16, y = var_431_cast_fp16)[name = tensor("x_5_cast_fp16")]; + tensor inputs_7_cast_fp16 = add(x = inputs_5_cast_fp16, y = x_5_cast_fp16)[name = tensor("inputs_7_cast_fp16")]; + tensor out_7_axes_0 = const()[name = tensor("out_7_axes_0"), val = tensor([1])]; + tensor var_442_to_fp16 = const()[name = tensor("op_442_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_7_cast_fp16 = layer_norm(axes = out_7_axes_0, epsilon = var_442_to_fp16, x = inputs_7_cast_fp16)[name = tensor("out_7_cast_fp16")]; + tensor input_37_gamma_0_to_fp16 = const()[name = tensor("input_37_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5716672)))]; + tensor input_37_beta_0_to_fp16 = const()[name = tensor("input_37_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5717760)))]; + tensor input_37_epsilon_0_to_fp16 = const()[name = tensor("input_37_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_37_cast_fp16 = batch_norm(beta = input_37_beta_0_to_fp16, epsilon = input_37_epsilon_0_to_fp16, gamma = input_37_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_7_cast_fp16)[name = tensor("input_37_cast_fp16")]; + tensor var_462_pad_type_0 = const()[name = tensor("op_462_pad_type_0"), val = tensor("valid")]; + tensor var_462_strides_0 = const()[name = tensor("op_462_strides_0"), val = tensor([1, 1])]; + tensor var_462_pad_0 = const()[name = tensor("op_462_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_462_dilations_0 = const()[name = tensor("op_462_dilations_0"), val = tensor([1, 1])]; + tensor var_462_groups_0 = const()[name = tensor("op_462_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(5718848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6505344))), name = tensor("layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_0_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6505536)))]; + tensor var_462_cast_fp16 = conv(bias = layers_0_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_462_dilations_0, groups = var_462_groups_0, pad = var_462_pad_0, pad_type = var_462_pad_type_0, strides = var_462_strides_0, weight = layers_0_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_37_cast_fp16)[name = tensor("op_462_cast_fp16")]; + tensor var_468_pad_type_0 = const()[name = tensor("op_468_pad_type_0"), val = tensor("valid")]; + tensor var_468_strides_0 = const()[name = tensor("op_468_strides_0"), val = tensor([1, 1])]; + tensor var_468_pad_0 = const()[name = tensor("op_468_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_468_dilations_0 = const()[name = tensor("op_468_dilations_0"), val = tensor([1, 1])]; + tensor var_468_groups_0 = const()[name = tensor("op_468_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6541376))), name = tensor("layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6509696))), shape = tensor([2048, 512, 1, 1])]; + tensor var_468_cast_fp16 = conv(dilations = var_468_dilations_0, groups = var_468_groups_0, pad = var_468_pad_0, pad_type = var_468_pad_type_0, strides = var_468_strides_0, weight = layers_0_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_37_cast_fp16)[name = tensor("op_468_cast_fp16")]; + tensor input_39_cast_fp16 = add(x = var_462_cast_fp16, y = var_468_cast_fp16)[name = tensor("input_39_cast_fp16")]; + tensor input_41_cast_fp16 = silu(x = input_39_cast_fp16)[name = tensor("input_41_cast_fp16")]; + tensor var_479_pad_type_0 = const()[name = tensor("op_479_pad_type_0"), val = tensor("valid")]; + tensor var_479_strides_0 = const()[name = tensor("op_479_strides_0"), val = tensor([1, 1])]; + tensor var_479_pad_0 = const()[name = tensor("op_479_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_479_dilations_0 = const()[name = tensor("op_479_dilations_0"), val = tensor([1, 1])]; + tensor var_479_groups_0 = const()[name = tensor("op_479_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(6672512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7459008))), name = tensor("layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_0_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_0_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7459200)))]; + tensor var_479_cast_fp16 = conv(bias = layers_0_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_479_dilations_0, groups = var_479_groups_0, pad = var_479_pad_0, pad_type = var_479_pad_type_0, strides = var_479_strides_0, weight = layers_0_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_41_cast_fp16)[name = tensor("op_479_cast_fp16")]; + tensor var_485_pad_type_0 = const()[name = tensor("op_485_pad_type_0"), val = tensor("valid")]; + tensor var_485_strides_0 = const()[name = tensor("op_485_strides_0"), val = tensor([1, 1])]; + tensor var_485_pad_0 = const()[name = tensor("op_485_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_485_dilations_0 = const()[name = tensor("op_485_dilations_0"), val = tensor([1, 1])]; + tensor var_485_groups_0 = const()[name = tensor("op_485_groups_0"), val = tensor(1)]; + tensor layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7490688))), name = tensor("layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7460288))), shape = tensor([512, 2048, 1, 1])]; + tensor var_485_cast_fp16 = conv(dilations = var_485_dilations_0, groups = var_485_groups_0, pad = var_485_pad_0, pad_type = var_485_pad_type_0, strides = var_485_strides_0, weight = layers_0_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_41_cast_fp16)[name = tensor("op_485_cast_fp16")]; + tensor x_7_cast_fp16 = add(x = var_479_cast_fp16, y = var_485_cast_fp16)[name = tensor("x_7_cast_fp16")]; + tensor var_487_to_fp16 = const()[name = tensor("op_487_to_fp16"), val = tensor(0x1p-1)]; + tensor var_488_cast_fp16 = mul(x = x_7_cast_fp16, y = var_487_to_fp16)[name = tensor("op_488_cast_fp16")]; + tensor inputs_9_cast_fp16 = add(x = inputs_7_cast_fp16, y = var_488_cast_fp16)[name = tensor("inputs_9_cast_fp16")]; + tensor out_9_axes_0 = const()[name = tensor("out_9_axes_0"), val = tensor([1])]; + tensor var_498_to_fp16 = const()[name = tensor("op_498_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_9_cast_fp16 = layer_norm(axes = out_9_axes_0, epsilon = var_498_to_fp16, x = inputs_9_cast_fp16)[name = tensor("out_9_cast_fp16")]; + tensor inputs_11_gamma_0_to_fp16 = const()[name = tensor("inputs_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7621824)))]; + tensor inputs_11_beta_0_to_fp16 = const()[name = tensor("inputs_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7622912)))]; + tensor inputs_11_epsilon_0_to_fp16 = const()[name = tensor("inputs_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_11_cast_fp16 = batch_norm(beta = inputs_11_beta_0_to_fp16, epsilon = inputs_11_epsilon_0_to_fp16, gamma = inputs_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_9_cast_fp16)[name = tensor("inputs_11_cast_fp16")]; + tensor var_512 = const()[name = tensor("op_512"), val = tensor(3)]; + tensor out_11_axes_0 = const()[name = tensor("out_11_axes_0"), val = tensor([1])]; + tensor var_543_to_fp16 = const()[name = tensor("op_543_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_11_cast_fp16 = layer_norm(axes = out_11_axes_0, epsilon = var_543_to_fp16, x = inputs_11_cast_fp16)[name = tensor("out_11_cast_fp16")]; + tensor input_43_gamma_0_to_fp16 = const()[name = tensor("input_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7624000)))]; + tensor input_43_beta_0_to_fp16 = const()[name = tensor("input_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7625088)))]; + tensor input_43_epsilon_0_to_fp16 = const()[name = tensor("input_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_43_cast_fp16 = batch_norm(beta = input_43_beta_0_to_fp16, epsilon = input_43_epsilon_0_to_fp16, gamma = input_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_11_cast_fp16)[name = tensor("input_43_cast_fp16")]; + tensor var_563_pad_type_0 = const()[name = tensor("op_563_pad_type_0"), val = tensor("valid")]; + tensor var_563_strides_0 = const()[name = tensor("op_563_strides_0"), val = tensor([1, 1])]; + tensor var_563_pad_0 = const()[name = tensor("op_563_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_563_dilations_0 = const()[name = tensor("op_563_dilations_0"), val = tensor([1, 1])]; + tensor var_563_groups_0 = const()[name = tensor("op_563_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(7626176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8412672))), name = tensor("layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_1_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8412864)))]; + tensor var_563_cast_fp16 = conv(bias = layers_1_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_563_dilations_0, groups = var_563_groups_0, pad = var_563_pad_0, pad_type = var_563_pad_type_0, strides = var_563_strides_0, weight = layers_1_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_43_cast_fp16)[name = tensor("op_563_cast_fp16")]; + tensor var_569_pad_type_0 = const()[name = tensor("op_569_pad_type_0"), val = tensor("valid")]; + tensor var_569_strides_0 = const()[name = tensor("op_569_strides_0"), val = tensor([1, 1])]; + tensor var_569_pad_0 = const()[name = tensor("op_569_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_569_dilations_0 = const()[name = tensor("op_569_dilations_0"), val = tensor([1, 1])]; + tensor var_569_groups_0 = const()[name = tensor("op_569_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8451136))), name = tensor("layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8417024))), shape = tensor([2048, 512, 1, 1])]; + tensor var_569_cast_fp16 = conv(dilations = var_569_dilations_0, groups = var_569_groups_0, pad = var_569_pad_0, pad_type = var_569_pad_type_0, strides = var_569_strides_0, weight = layers_1_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_43_cast_fp16)[name = tensor("op_569_cast_fp16")]; + tensor input_45_cast_fp16 = add(x = var_563_cast_fp16, y = var_569_cast_fp16)[name = tensor("input_45_cast_fp16")]; + tensor input_47_cast_fp16 = silu(x = input_45_cast_fp16)[name = tensor("input_47_cast_fp16")]; + tensor var_580_pad_type_0 = const()[name = tensor("op_580_pad_type_0"), val = tensor("valid")]; + tensor var_580_strides_0 = const()[name = tensor("op_580_strides_0"), val = tensor([1, 1])]; + tensor var_580_pad_0 = const()[name = tensor("op_580_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_580_dilations_0 = const()[name = tensor("op_580_dilations_0"), val = tensor([1, 1])]; + tensor var_580_groups_0 = const()[name = tensor("op_580_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(8582272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9368768))), name = tensor("layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_1_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9368960)))]; + tensor var_580_cast_fp16 = conv(bias = layers_1_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_580_dilations_0, groups = var_580_groups_0, pad = var_580_pad_0, pad_type = var_580_pad_type_0, strides = var_580_strides_0, weight = layers_1_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_47_cast_fp16)[name = tensor("op_580_cast_fp16")]; + tensor var_586_pad_type_0 = const()[name = tensor("op_586_pad_type_0"), val = tensor("valid")]; + tensor var_586_strides_0 = const()[name = tensor("op_586_strides_0"), val = tensor([1, 1])]; + tensor var_586_pad_0 = const()[name = tensor("op_586_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_586_dilations_0 = const()[name = tensor("op_586_dilations_0"), val = tensor([1, 1])]; + tensor var_586_groups_0 = const()[name = tensor("op_586_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9406336))), name = tensor("layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9370048))), shape = tensor([512, 2048, 1, 1])]; + tensor var_586_cast_fp16 = conv(dilations = var_586_dilations_0, groups = var_586_groups_0, pad = var_586_pad_0, pad_type = var_586_pad_type_0, strides = var_586_strides_0, weight = layers_1_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_47_cast_fp16)[name = tensor("op_586_cast_fp16")]; + tensor x_9_cast_fp16 = add(x = var_580_cast_fp16, y = var_586_cast_fp16)[name = tensor("x_9_cast_fp16")]; + tensor var_588_to_fp16 = const()[name = tensor("op_588_to_fp16"), val = tensor(0x1p-1)]; + tensor var_589_cast_fp16 = mul(x = x_9_cast_fp16, y = var_588_to_fp16)[name = tensor("op_589_cast_fp16")]; + tensor inputs_13_cast_fp16 = add(x = inputs_11_cast_fp16, y = var_589_cast_fp16)[name = tensor("inputs_13_cast_fp16")]; + tensor out_13_axes_0 = const()[name = tensor("out_13_axes_0"), val = tensor([1])]; + tensor var_599_to_fp16 = const()[name = tensor("op_599_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_13_cast_fp16 = layer_norm(axes = out_13_axes_0, epsilon = var_599_to_fp16, x = inputs_13_cast_fp16)[name = tensor("out_13_cast_fp16")]; + tensor obj_7_gamma_0_to_fp16 = const()[name = tensor("obj_7_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9537472)))]; + tensor obj_7_beta_0_to_fp16 = const()[name = tensor("obj_7_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9538560)))]; + tensor obj_7_epsilon_0_to_fp16 = const()[name = tensor("obj_7_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_7_cast_fp16 = batch_norm(beta = obj_7_beta_0_to_fp16, epsilon = obj_7_epsilon_0_to_fp16, gamma = obj_7_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_13_cast_fp16)[name = tensor("obj_7_cast_fp16")]; + tensor var_624_pad_type_0 = const()[name = tensor("op_624_pad_type_0"), val = tensor("valid")]; + tensor var_624_strides_0 = const()[name = tensor("op_624_strides_0"), val = tensor([1, 1])]; + tensor var_624_pad_0 = const()[name = tensor("op_624_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_624_dilations_0 = const()[name = tensor("op_624_dilations_0"), val = tensor([1, 1])]; + tensor var_624_groups_0 = const()[name = tensor("op_624_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9539648))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9736320))), name = tensor("layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_1_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9736512)))]; + tensor var_624_cast_fp16 = conv(bias = layers_1_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_624_dilations_0, groups = var_624_groups_0, pad = var_624_pad_0, pad_type = var_624_pad_type_0, strides = var_624_strides_0, weight = layers_1_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_624_cast_fp16")]; + tensor var_630_pad_type_0 = const()[name = tensor("op_630_pad_type_0"), val = tensor("valid")]; + tensor var_630_strides_0 = const()[name = tensor("op_630_strides_0"), val = tensor([1, 1])]; + tensor var_630_pad_0 = const()[name = tensor("op_630_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_630_dilations_0 = const()[name = tensor("op_630_dilations_0"), val = tensor([1, 1])]; + tensor var_630_groups_0 = const()[name = tensor("op_630_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9746432))), name = tensor("layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9737600))), shape = tensor([512, 512, 1, 1])]; + tensor var_630_cast_fp16 = conv(dilations = var_630_dilations_0, groups = var_630_groups_0, pad = var_630_pad_0, pad_type = var_630_pad_type_0, strides = var_630_strides_0, weight = layers_1_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_630_cast_fp16")]; + tensor query_5_cast_fp16 = add(x = var_624_cast_fp16, y = var_630_cast_fp16)[name = tensor("query_5_cast_fp16")]; + tensor var_639_pad_type_0 = const()[name = tensor("op_639_pad_type_0"), val = tensor("valid")]; + tensor var_639_strides_0 = const()[name = tensor("op_639_strides_0"), val = tensor([1, 1])]; + tensor var_639_pad_0 = const()[name = tensor("op_639_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_639_dilations_0 = const()[name = tensor("op_639_dilations_0"), val = tensor([1, 1])]; + tensor var_639_groups_0 = const()[name = tensor("op_639_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9779264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9975936))), name = tensor("layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_639_cast_fp16 = conv(dilations = var_639_dilations_0, groups = var_639_groups_0, pad = var_639_pad_0, pad_type = var_639_pad_type_0, strides = var_639_strides_0, weight = layers_1_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_639_cast_fp16")]; + tensor var_645_pad_type_0 = const()[name = tensor("op_645_pad_type_0"), val = tensor("valid")]; + tensor var_645_strides_0 = const()[name = tensor("op_645_strides_0"), val = tensor([1, 1])]; + tensor var_645_pad_0 = const()[name = tensor("op_645_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_645_dilations_0 = const()[name = tensor("op_645_dilations_0"), val = tensor([1, 1])]; + tensor var_645_groups_0 = const()[name = tensor("op_645_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9985344))), name = tensor("layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(9976128))), shape = tensor([512, 512, 1, 1])]; + tensor var_645_cast_fp16 = conv(dilations = var_645_dilations_0, groups = var_645_groups_0, pad = var_645_pad_0, pad_type = var_645_pad_type_0, strides = var_645_strides_0, weight = layers_1_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_645_cast_fp16")]; + tensor key_3_cast_fp16 = add(x = var_639_cast_fp16, y = var_645_cast_fp16)[name = tensor("key_3_cast_fp16")]; + tensor var_655_pad_type_0 = const()[name = tensor("op_655_pad_type_0"), val = tensor("valid")]; + tensor var_655_strides_0 = const()[name = tensor("op_655_strides_0"), val = tensor([1, 1])]; + tensor var_655_pad_0 = const()[name = tensor("op_655_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_655_dilations_0 = const()[name = tensor("op_655_dilations_0"), val = tensor([1, 1])]; + tensor var_655_groups_0 = const()[name = tensor("op_655_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10018176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10214848))), name = tensor("layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_1_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10215040)))]; + tensor var_655_cast_fp16 = conv(bias = layers_1_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_655_dilations_0, groups = var_655_groups_0, pad = var_655_pad_0, pad_type = var_655_pad_type_0, strides = var_655_strides_0, weight = layers_1_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_7_cast_fp16)[name = tensor("op_655_cast_fp16")]; + tensor var_661_pad_type_0 = const()[name = tensor("op_661_pad_type_0"), val = tensor("valid")]; + tensor var_661_strides_0 = const()[name = tensor("op_661_strides_0"), val = tensor([1, 1])]; + tensor var_661_pad_0 = const()[name = tensor("op_661_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_661_dilations_0 = const()[name = tensor("op_661_dilations_0"), val = tensor([1, 1])]; + tensor var_661_groups_0 = const()[name = tensor("op_661_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10223872))), name = tensor("layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10216128))), shape = tensor([512, 512, 1, 1])]; + tensor var_661_cast_fp16 = conv(dilations = var_661_dilations_0, groups = var_661_groups_0, pad = var_661_pad_0, pad_type = var_661_pad_type_0, strides = var_661_strides_0, weight = layers_1_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_7_cast_fp16)[name = tensor("op_661_cast_fp16")]; + tensor value_3_cast_fp16 = add(x = var_655_cast_fp16, y = var_661_cast_fp16)[name = tensor("value_3_cast_fp16")]; + tensor var_664_to_fp16 = const()[name = tensor("op_664_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10256704)))]; + tensor query_7_cast_fp16 = add(x = query_5_cast_fp16, y = var_664_to_fp16)[name = tensor("query_7_cast_fp16")]; + tensor var_667_to_fp16 = const()[name = tensor("op_667_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10257792)))]; + tensor q_with_bias_v_3_cast_fp16 = add(x = query_5_cast_fp16, y = var_667_to_fp16)[name = tensor("q_with_bias_v_3_cast_fp16")]; + tensor var_677_pad_type_0 = const()[name = tensor("op_677_pad_type_0"), val = tensor("valid")]; + tensor var_677_strides_0 = const()[name = tensor("op_677_strides_0"), val = tensor([1, 1])]; + tensor var_677_pad_0 = const()[name = tensor("op_677_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_677_dilations_0 = const()[name = tensor("op_677_dilations_0"), val = tensor([1, 1])]; + tensor var_677_groups_0 = const()[name = tensor("op_677_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10258880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10455552))), name = tensor("layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_677_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_677_dilations_0, groups = var_677_groups_0, pad = var_677_pad_0, pad_type = var_677_pad_type_0, strides = var_677_strides_0, weight = layers_1_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_677_cast_fp16")]; + tensor var_683_pad_type_0 = const()[name = tensor("op_683_pad_type_0"), val = tensor("valid")]; + tensor var_683_strides_0 = const()[name = tensor("op_683_strides_0"), val = tensor([1, 1])]; + tensor var_683_pad_0 = const()[name = tensor("op_683_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_683_dilations_0 = const()[name = tensor("op_683_dilations_0"), val = tensor([1, 1])]; + tensor var_683_groups_0 = const()[name = tensor("op_683_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10477632))), name = tensor("layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10455744))), shape = tensor([512, 512, 1, 1])]; + tensor var_683_cast_fp16 = conv(dilations = var_683_dilations_0, groups = var_683_groups_0, pad = var_683_pad_0, pad_type = var_683_pad_type_0, strides = var_683_strides_0, weight = layers_1_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_683_cast_fp16")]; + tensor p_3_cast_fp16 = add(x = var_677_cast_fp16, y = var_683_cast_fp16)[name = tensor("p_3_cast_fp16")]; + tensor var_687 = const()[name = tensor("op_687"), val = tensor([1, 8, 64, 188])]; + tensor var_688_cast_fp16 = reshape(shape = var_687, x = q_with_bias_v_3_cast_fp16)[name = tensor("op_688_cast_fp16")]; + tensor var_689 = const()[name = tensor("op_689"), val = tensor([1, 8, 64, -1])]; + tensor var_690_cast_fp16 = reshape(shape = var_689, x = p_3_cast_fp16)[name = tensor("op_690_cast_fp16")]; + tensor matrix_bd_9_transpose_x_0 = const()[name = tensor("matrix_bd_9_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_9_transpose_y_0 = const()[name = tensor("matrix_bd_9_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_9_cast_fp16 = matmul(transpose_x = matrix_bd_9_transpose_x_0, transpose_y = matrix_bd_9_transpose_y_0, x = var_688_cast_fp16, y = var_690_cast_fp16)[name = tensor("matrix_bd_9_cast_fp16")]; + tensor matrix_bd_11_pad_0 = const()[name = tensor("matrix_bd_11_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_11_mode_0 = const()[name = tensor("matrix_bd_11_mode_0"), val = tensor("constant")]; + tensor const_21_to_fp16 = const()[name = tensor("const_21_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_11_cast_fp16 = pad(constant_val = const_21_to_fp16, mode = matrix_bd_11_mode_0, pad = matrix_bd_11_pad_0, x = matrix_bd_9_cast_fp16)[name = tensor("matrix_bd_11_cast_fp16")]; + tensor var_699 = const()[name = tensor("op_699"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_13_cast_fp16 = reshape(shape = var_699, x = matrix_bd_11_cast_fp16)[name = tensor("matrix_bd_13_cast_fp16")]; + tensor var_703_begin_0 = const()[name = tensor("op_703_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_703_end_0 = const()[name = tensor("op_703_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_703_end_mask_0 = const()[name = tensor("op_703_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_703_cast_fp16 = slice_by_index(begin = var_703_begin_0, end = var_703_end_0, end_mask = var_703_end_mask_0, x = matrix_bd_13_cast_fp16)[name = tensor("op_703_cast_fp16")]; + tensor var_704 = const()[name = tensor("op_704"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_15_cast_fp16 = reshape(shape = var_704, x = var_703_cast_fp16)[name = tensor("matrix_bd_15_cast_fp16")]; + tensor var_709_begin_0 = const()[name = tensor("op_709_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_709_end_0 = const()[name = tensor("op_709_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_709_end_mask_0 = const()[name = tensor("op_709_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_709_cast_fp16 = slice_by_index(begin = var_709_begin_0, end = var_709_end_0, end_mask = var_709_end_mask_0, x = matrix_bd_15_cast_fp16)[name = tensor("op_709_cast_fp16")]; + tensor var_710_to_fp16 = const()[name = tensor("op_710_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_3_cast_fp16 = mul(x = var_709_cast_fp16, y = var_710_to_fp16)[name = tensor("qk_mask_3_cast_fp16")]; + tensor var_714 = const()[name = tensor("op_714"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_3_cast_fp16 = reshape(shape = var_714, x = query_7_cast_fp16)[name = tensor("mh_q_3_cast_fp16")]; + tensor var_716_to_fp16 = const()[name = tensor("op_716_to_fp16"), val = tensor(0x1p-3)]; + tensor var_717_cast_fp16 = mul(x = mh_q_3_cast_fp16, y = var_716_to_fp16)[name = tensor("op_717_cast_fp16")]; + tensor var_720 = const()[name = tensor("op_720"), val = tensor([1, 8, 64, 188])]; + tensor var_721_cast_fp16 = reshape(shape = var_720, x = key_3_cast_fp16)[name = tensor("op_721_cast_fp16")]; + tensor mh_w_5_transpose_x_0 = const()[name = tensor("mh_w_5_transpose_x_0"), val = tensor(true)]; + tensor mh_w_5_transpose_y_0 = const()[name = tensor("mh_w_5_transpose_y_0"), val = tensor(false)]; + tensor mh_w_5_cast_fp16 = matmul(transpose_x = mh_w_5_transpose_x_0, transpose_y = mh_w_5_transpose_y_0, x = var_717_cast_fp16, y = var_721_cast_fp16)[name = tensor("mh_w_5_cast_fp16")]; + tensor mh_w_7_cast_fp16 = add(x = mh_w_5_cast_fp16, y = qk_mask_3_cast_fp16)[name = tensor("mh_w_7_cast_fp16")]; + tensor var_725_cast_fp16 = softmax(axis = var_512, x = mh_w_7_cast_fp16)[name = tensor("op_725_cast_fp16")]; + tensor var_726 = const()[name = tensor("op_726"), val = tensor([1, 8, 64, 188])]; + tensor var_727_cast_fp16 = reshape(shape = var_726, x = value_3_cast_fp16)[name = tensor("op_727_cast_fp16")]; + tensor attn_3_transpose_x_0 = const()[name = tensor("attn_3_transpose_x_0"), val = tensor(false)]; + tensor attn_3_transpose_y_0 = const()[name = tensor("attn_3_transpose_y_0"), val = tensor(true)]; + tensor attn_3_cast_fp16 = matmul(transpose_x = attn_3_transpose_x_0, transpose_y = attn_3_transpose_y_0, x = var_727_cast_fp16, y = var_725_cast_fp16)[name = tensor("attn_3_cast_fp16")]; + tensor var_730 = const()[name = tensor("op_730"), val = tensor([1, 512, 1, 188])]; + tensor input_49_cast_fp16 = reshape(shape = var_730, x = attn_3_cast_fp16)[name = tensor("input_49_cast_fp16")]; + tensor var_740_pad_type_0 = const()[name = tensor("op_740_pad_type_0"), val = tensor("valid")]; + tensor var_740_strides_0 = const()[name = tensor("op_740_strides_0"), val = tensor([1, 1])]; + tensor var_740_pad_0 = const()[name = tensor("op_740_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_740_dilations_0 = const()[name = tensor("op_740_dilations_0"), val = tensor([1, 1])]; + tensor var_740_groups_0 = const()[name = tensor("op_740_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10510464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10707136))), name = tensor("layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_1_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10707328)))]; + tensor var_740_cast_fp16 = conv(bias = layers_1_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_740_dilations_0, groups = var_740_groups_0, pad = var_740_pad_0, pad_type = var_740_pad_type_0, strides = var_740_strides_0, weight = layers_1_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_49_cast_fp16)[name = tensor("op_740_cast_fp16")]; + tensor var_746_pad_type_0 = const()[name = tensor("op_746_pad_type_0"), val = tensor("valid")]; + tensor var_746_strides_0 = const()[name = tensor("op_746_strides_0"), val = tensor([1, 1])]; + tensor var_746_pad_0 = const()[name = tensor("op_746_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_746_dilations_0 = const()[name = tensor("op_746_dilations_0"), val = tensor([1, 1])]; + tensor var_746_groups_0 = const()[name = tensor("op_746_groups_0"), val = tensor(1)]; + tensor layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10716288))), name = tensor("layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10708416))), shape = tensor([512, 512, 1, 1])]; + tensor var_746_cast_fp16 = conv(dilations = var_746_dilations_0, groups = var_746_groups_0, pad = var_746_pad_0, pad_type = var_746_pad_type_0, strides = var_746_strides_0, weight = layers_1_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_49_cast_fp16)[name = tensor("op_746_cast_fp16")]; + tensor obj_9_cast_fp16 = add(x = var_740_cast_fp16, y = var_746_cast_fp16)[name = tensor("obj_9_cast_fp16")]; + tensor inputs_15_cast_fp16 = add(x = inputs_13_cast_fp16, y = obj_9_cast_fp16)[name = tensor("inputs_15_cast_fp16")]; + tensor out_15_axes_0 = const()[name = tensor("out_15_axes_0"), val = tensor([1])]; + tensor var_757_to_fp16 = const()[name = tensor("op_757_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_15_cast_fp16 = layer_norm(axes = out_15_axes_0, epsilon = var_757_to_fp16, x = inputs_15_cast_fp16)[name = tensor("out_15_cast_fp16")]; + tensor input_51_gamma_0_to_fp16 = const()[name = tensor("input_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10749120)))]; + tensor input_51_beta_0_to_fp16 = const()[name = tensor("input_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10750208)))]; + tensor input_51_epsilon_0_to_fp16 = const()[name = tensor("input_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_51_cast_fp16 = batch_norm(beta = input_51_beta_0_to_fp16, epsilon = input_51_epsilon_0_to_fp16, gamma = input_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_15_cast_fp16)[name = tensor("input_51_cast_fp16")]; + tensor var_779_pad_type_0 = const()[name = tensor("op_779_pad_type_0"), val = tensor("valid")]; + tensor var_779_strides_0 = const()[name = tensor("op_779_strides_0"), val = tensor([1, 1])]; + tensor var_779_pad_0 = const()[name = tensor("op_779_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_779_dilations_0 = const()[name = tensor("op_779_dilations_0"), val = tensor([1, 1])]; + tensor var_779_groups_0 = const()[name = tensor("op_779_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(10751296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11144576))), name = tensor("layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_1_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11144768)))]; + tensor var_779_cast_fp16 = conv(bias = layers_1_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_779_dilations_0, groups = var_779_groups_0, pad = var_779_pad_0, pad_type = var_779_pad_type_0, strides = var_779_strides_0, weight = layers_1_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_51_cast_fp16)[name = tensor("op_779_cast_fp16")]; + tensor var_785_pad_type_0 = const()[name = tensor("op_785_pad_type_0"), val = tensor("valid")]; + tensor var_785_strides_0 = const()[name = tensor("op_785_strides_0"), val = tensor([1, 1])]; + tensor var_785_pad_0 = const()[name = tensor("op_785_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_785_dilations_0 = const()[name = tensor("op_785_dilations_0"), val = tensor([1, 1])]; + tensor var_785_groups_0 = const()[name = tensor("op_785_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11165248))), name = tensor("layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11146880))), shape = tensor([1024, 512, 1, 1])]; + tensor var_785_cast_fp16 = conv(dilations = var_785_dilations_0, groups = var_785_groups_0, pad = var_785_pad_0, pad_type = var_785_pad_type_0, strides = var_785_strides_0, weight = layers_1_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_51_cast_fp16)[name = tensor("op_785_cast_fp16")]; + tensor input_53_cast_fp16 = add(x = var_779_cast_fp16, y = var_785_cast_fp16)[name = tensor("input_53_cast_fp16")]; + tensor input_55_split_num_splits_0 = const()[name = tensor("input_55_split_num_splits_0"), val = tensor(2)]; + tensor input_55_split_axis_0 = const()[name = tensor("input_55_split_axis_0"), val = tensor(1)]; + tensor input_55_split_cast_fp16_0, tensor input_55_split_cast_fp16_1 = split(axis = input_55_split_axis_0, num_splits = input_55_split_num_splits_0, x = input_53_cast_fp16)[name = tensor("input_55_split_cast_fp16")]; + tensor input_55_split_1_sigmoid_cast_fp16 = sigmoid(x = input_55_split_cast_fp16_1)[name = tensor("input_55_split_1_sigmoid_cast_fp16")]; + tensor input_55_cast_fp16 = mul(x = input_55_split_cast_fp16_0, y = input_55_split_1_sigmoid_cast_fp16)[name = tensor("input_55_cast_fp16")]; + tensor input_57_pad_type_0 = const()[name = tensor("input_57_pad_type_0"), val = tensor("custom")]; + tensor input_57_pad_0 = const()[name = tensor("input_57_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_57_groups_0 = const()[name = tensor("input_57_groups_0"), val = tensor(512)]; + tensor input_57_strides_0 = const()[name = tensor("input_57_strides_0"), val = tensor([1, 1])]; + tensor input_57_dilations_0 = const()[name = tensor("input_57_dilations_0"), val = tensor([1, 1])]; + tensor const_193_to_fp16 = const()[name = tensor("const_193_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11230848)))]; + tensor const_194_to_fp16 = const()[name = tensor("const_194_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11240128)))]; + tensor input_59_cast_fp16 = conv(bias = const_194_to_fp16, dilations = input_57_dilations_0, groups = input_57_groups_0, pad = input_57_pad_0, pad_type = input_57_pad_type_0, strides = input_57_strides_0, weight = const_193_to_fp16, x = input_55_cast_fp16)[name = tensor("input_59_cast_fp16")]; + tensor input_61_cast_fp16 = silu(x = input_59_cast_fp16)[name = tensor("input_61_cast_fp16")]; + tensor var_809_pad_type_0 = const()[name = tensor("op_809_pad_type_0"), val = tensor("valid")]; + tensor var_809_strides_0 = const()[name = tensor("op_809_strides_0"), val = tensor([1, 1])]; + tensor var_809_pad_0 = const()[name = tensor("op_809_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_809_dilations_0 = const()[name = tensor("op_809_dilations_0"), val = tensor([1, 1])]; + tensor var_809_groups_0 = const()[name = tensor("op_809_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11241216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11437888))), name = tensor("layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_1_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11438080)))]; + tensor var_809_cast_fp16 = conv(bias = layers_1_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_809_dilations_0, groups = var_809_groups_0, pad = var_809_pad_0, pad_type = var_809_pad_type_0, strides = var_809_strides_0, weight = layers_1_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_61_cast_fp16)[name = tensor("op_809_cast_fp16")]; + tensor var_815_pad_type_0 = const()[name = tensor("op_815_pad_type_0"), val = tensor("valid")]; + tensor var_815_strides_0 = const()[name = tensor("op_815_strides_0"), val = tensor([1, 1])]; + tensor var_815_pad_0 = const()[name = tensor("op_815_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_815_dilations_0 = const()[name = tensor("op_815_dilations_0"), val = tensor([1, 1])]; + tensor var_815_groups_0 = const()[name = tensor("op_815_groups_0"), val = tensor(1)]; + tensor layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11447872))), name = tensor("layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11439168))), shape = tensor([512, 512, 1, 1])]; + tensor var_815_cast_fp16 = conv(dilations = var_815_dilations_0, groups = var_815_groups_0, pad = var_815_pad_0, pad_type = var_815_pad_type_0, strides = var_815_strides_0, weight = layers_1_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_61_cast_fp16)[name = tensor("op_815_cast_fp16")]; + tensor x_11_cast_fp16 = add(x = var_809_cast_fp16, y = var_815_cast_fp16)[name = tensor("x_11_cast_fp16")]; + tensor inputs_17_cast_fp16 = add(x = inputs_15_cast_fp16, y = x_11_cast_fp16)[name = tensor("inputs_17_cast_fp16")]; + tensor out_17_axes_0 = const()[name = tensor("out_17_axes_0"), val = tensor([1])]; + tensor var_826_to_fp16 = const()[name = tensor("op_826_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_17_cast_fp16 = layer_norm(axes = out_17_axes_0, epsilon = var_826_to_fp16, x = inputs_17_cast_fp16)[name = tensor("out_17_cast_fp16")]; + tensor input_63_gamma_0_to_fp16 = const()[name = tensor("input_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11480704)))]; + tensor input_63_beta_0_to_fp16 = const()[name = tensor("input_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11481792)))]; + tensor input_63_epsilon_0_to_fp16 = const()[name = tensor("input_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_63_cast_fp16 = batch_norm(beta = input_63_beta_0_to_fp16, epsilon = input_63_epsilon_0_to_fp16, gamma = input_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_17_cast_fp16)[name = tensor("input_63_cast_fp16")]; + tensor var_846_pad_type_0 = const()[name = tensor("op_846_pad_type_0"), val = tensor("valid")]; + tensor var_846_strides_0 = const()[name = tensor("op_846_strides_0"), val = tensor([1, 1])]; + tensor var_846_pad_0 = const()[name = tensor("op_846_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_846_dilations_0 = const()[name = tensor("op_846_dilations_0"), val = tensor([1, 1])]; + tensor var_846_groups_0 = const()[name = tensor("op_846_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(11482880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12269376))), name = tensor("layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_1_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12269568)))]; + tensor var_846_cast_fp16 = conv(bias = layers_1_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_846_dilations_0, groups = var_846_groups_0, pad = var_846_pad_0, pad_type = var_846_pad_type_0, strides = var_846_strides_0, weight = layers_1_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_63_cast_fp16)[name = tensor("op_846_cast_fp16")]; + tensor var_852_pad_type_0 = const()[name = tensor("op_852_pad_type_0"), val = tensor("valid")]; + tensor var_852_strides_0 = const()[name = tensor("op_852_strides_0"), val = tensor([1, 1])]; + tensor var_852_pad_0 = const()[name = tensor("op_852_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_852_dilations_0 = const()[name = tensor("op_852_dilations_0"), val = tensor([1, 1])]; + tensor var_852_groups_0 = const()[name = tensor("op_852_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12306112))), name = tensor("layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12273728))), shape = tensor([2048, 512, 1, 1])]; + tensor var_852_cast_fp16 = conv(dilations = var_852_dilations_0, groups = var_852_groups_0, pad = var_852_pad_0, pad_type = var_852_pad_type_0, strides = var_852_strides_0, weight = layers_1_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_63_cast_fp16)[name = tensor("op_852_cast_fp16")]; + tensor input_65_cast_fp16 = add(x = var_846_cast_fp16, y = var_852_cast_fp16)[name = tensor("input_65_cast_fp16")]; + tensor input_67_cast_fp16 = silu(x = input_65_cast_fp16)[name = tensor("input_67_cast_fp16")]; + tensor var_863_pad_type_0 = const()[name = tensor("op_863_pad_type_0"), val = tensor("valid")]; + tensor var_863_strides_0 = const()[name = tensor("op_863_strides_0"), val = tensor([1, 1])]; + tensor var_863_pad_0 = const()[name = tensor("op_863_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_863_dilations_0 = const()[name = tensor("op_863_dilations_0"), val = tensor([1, 1])]; + tensor var_863_groups_0 = const()[name = tensor("op_863_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(12437248))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13223744))), name = tensor("layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_1_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_1_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13223936)))]; + tensor var_863_cast_fp16 = conv(bias = layers_1_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_863_dilations_0, groups = var_863_groups_0, pad = var_863_pad_0, pad_type = var_863_pad_type_0, strides = var_863_strides_0, weight = layers_1_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_67_cast_fp16)[name = tensor("op_863_cast_fp16")]; + tensor var_869_pad_type_0 = const()[name = tensor("op_869_pad_type_0"), val = tensor("valid")]; + tensor var_869_strides_0 = const()[name = tensor("op_869_strides_0"), val = tensor([1, 1])]; + tensor var_869_pad_0 = const()[name = tensor("op_869_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_869_dilations_0 = const()[name = tensor("op_869_dilations_0"), val = tensor([1, 1])]; + tensor var_869_groups_0 = const()[name = tensor("op_869_groups_0"), val = tensor(1)]; + tensor layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13260736))), name = tensor("layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13225024))), shape = tensor([512, 2048, 1, 1])]; + tensor var_869_cast_fp16 = conv(dilations = var_869_dilations_0, groups = var_869_groups_0, pad = var_869_pad_0, pad_type = var_869_pad_type_0, strides = var_869_strides_0, weight = layers_1_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_67_cast_fp16)[name = tensor("op_869_cast_fp16")]; + tensor x_13_cast_fp16 = add(x = var_863_cast_fp16, y = var_869_cast_fp16)[name = tensor("x_13_cast_fp16")]; + tensor var_871_to_fp16 = const()[name = tensor("op_871_to_fp16"), val = tensor(0x1p-1)]; + tensor var_872_cast_fp16 = mul(x = x_13_cast_fp16, y = var_871_to_fp16)[name = tensor("op_872_cast_fp16")]; + tensor inputs_19_cast_fp16 = add(x = inputs_17_cast_fp16, y = var_872_cast_fp16)[name = tensor("inputs_19_cast_fp16")]; + tensor out_19_axes_0 = const()[name = tensor("out_19_axes_0"), val = tensor([1])]; + tensor var_882_to_fp16 = const()[name = tensor("op_882_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_19_cast_fp16 = layer_norm(axes = out_19_axes_0, epsilon = var_882_to_fp16, x = inputs_19_cast_fp16)[name = tensor("out_19_cast_fp16")]; + tensor inputs_21_gamma_0_to_fp16 = const()[name = tensor("inputs_21_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13391872)))]; + tensor inputs_21_beta_0_to_fp16 = const()[name = tensor("inputs_21_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13392960)))]; + tensor inputs_21_epsilon_0_to_fp16 = const()[name = tensor("inputs_21_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_21_cast_fp16 = batch_norm(beta = inputs_21_beta_0_to_fp16, epsilon = inputs_21_epsilon_0_to_fp16, gamma = inputs_21_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_19_cast_fp16)[name = tensor("inputs_21_cast_fp16")]; + tensor var_896 = const()[name = tensor("op_896"), val = tensor(3)]; + tensor out_21_axes_0 = const()[name = tensor("out_21_axes_0"), val = tensor([1])]; + tensor var_927_to_fp16 = const()[name = tensor("op_927_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_21_cast_fp16 = layer_norm(axes = out_21_axes_0, epsilon = var_927_to_fp16, x = inputs_21_cast_fp16)[name = tensor("out_21_cast_fp16")]; + tensor input_69_gamma_0_to_fp16 = const()[name = tensor("input_69_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13394048)))]; + tensor input_69_beta_0_to_fp16 = const()[name = tensor("input_69_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13395136)))]; + tensor input_69_epsilon_0_to_fp16 = const()[name = tensor("input_69_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_69_cast_fp16 = batch_norm(beta = input_69_beta_0_to_fp16, epsilon = input_69_epsilon_0_to_fp16, gamma = input_69_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_21_cast_fp16)[name = tensor("input_69_cast_fp16")]; + tensor var_947_pad_type_0 = const()[name = tensor("op_947_pad_type_0"), val = tensor("valid")]; + tensor var_947_strides_0 = const()[name = tensor("op_947_strides_0"), val = tensor([1, 1])]; + tensor var_947_pad_0 = const()[name = tensor("op_947_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_947_dilations_0 = const()[name = tensor("op_947_dilations_0"), val = tensor([1, 1])]; + tensor var_947_groups_0 = const()[name = tensor("op_947_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(13396224))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14182720))), name = tensor("layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_2_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14182912)))]; + tensor var_947_cast_fp16 = conv(bias = layers_2_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_947_dilations_0, groups = var_947_groups_0, pad = var_947_pad_0, pad_type = var_947_pad_type_0, strides = var_947_strides_0, weight = layers_2_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_69_cast_fp16)[name = tensor("op_947_cast_fp16")]; + tensor var_953_pad_type_0 = const()[name = tensor("op_953_pad_type_0"), val = tensor("valid")]; + tensor var_953_strides_0 = const()[name = tensor("op_953_strides_0"), val = tensor([1, 1])]; + tensor var_953_pad_0 = const()[name = tensor("op_953_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_953_dilations_0 = const()[name = tensor("op_953_dilations_0"), val = tensor([1, 1])]; + tensor var_953_groups_0 = const()[name = tensor("op_953_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14219904))), name = tensor("layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14187072))), shape = tensor([2048, 512, 1, 1])]; + tensor var_953_cast_fp16 = conv(dilations = var_953_dilations_0, groups = var_953_groups_0, pad = var_953_pad_0, pad_type = var_953_pad_type_0, strides = var_953_strides_0, weight = layers_2_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_69_cast_fp16)[name = tensor("op_953_cast_fp16")]; + tensor input_71_cast_fp16 = add(x = var_947_cast_fp16, y = var_953_cast_fp16)[name = tensor("input_71_cast_fp16")]; + tensor input_73_cast_fp16 = silu(x = input_71_cast_fp16)[name = tensor("input_73_cast_fp16")]; + tensor var_964_pad_type_0 = const()[name = tensor("op_964_pad_type_0"), val = tensor("valid")]; + tensor var_964_strides_0 = const()[name = tensor("op_964_strides_0"), val = tensor([1, 1])]; + tensor var_964_pad_0 = const()[name = tensor("op_964_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_964_dilations_0 = const()[name = tensor("op_964_dilations_0"), val = tensor([1, 1])]; + tensor var_964_groups_0 = const()[name = tensor("op_964_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(14351040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15137536))), name = tensor("layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_2_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15137728)))]; + tensor var_964_cast_fp16 = conv(bias = layers_2_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_964_dilations_0, groups = var_964_groups_0, pad = var_964_pad_0, pad_type = var_964_pad_type_0, strides = var_964_strides_0, weight = layers_2_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_73_cast_fp16)[name = tensor("op_964_cast_fp16")]; + tensor var_970_pad_type_0 = const()[name = tensor("op_970_pad_type_0"), val = tensor("valid")]; + tensor var_970_strides_0 = const()[name = tensor("op_970_strides_0"), val = tensor([1, 1])]; + tensor var_970_pad_0 = const()[name = tensor("op_970_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_970_dilations_0 = const()[name = tensor("op_970_dilations_0"), val = tensor([1, 1])]; + tensor var_970_groups_0 = const()[name = tensor("op_970_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15173824))), name = tensor("layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15138816))), shape = tensor([512, 2048, 1, 1])]; + tensor var_970_cast_fp16 = conv(dilations = var_970_dilations_0, groups = var_970_groups_0, pad = var_970_pad_0, pad_type = var_970_pad_type_0, strides = var_970_strides_0, weight = layers_2_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_73_cast_fp16)[name = tensor("op_970_cast_fp16")]; + tensor x_15_cast_fp16 = add(x = var_964_cast_fp16, y = var_970_cast_fp16)[name = tensor("x_15_cast_fp16")]; + tensor var_972_to_fp16 = const()[name = tensor("op_972_to_fp16"), val = tensor(0x1p-1)]; + tensor var_973_cast_fp16 = mul(x = x_15_cast_fp16, y = var_972_to_fp16)[name = tensor("op_973_cast_fp16")]; + tensor inputs_23_cast_fp16 = add(x = inputs_21_cast_fp16, y = var_973_cast_fp16)[name = tensor("inputs_23_cast_fp16")]; + tensor out_23_axes_0 = const()[name = tensor("out_23_axes_0"), val = tensor([1])]; + tensor var_983_to_fp16 = const()[name = tensor("op_983_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_23_cast_fp16 = layer_norm(axes = out_23_axes_0, epsilon = var_983_to_fp16, x = inputs_23_cast_fp16)[name = tensor("out_23_cast_fp16")]; + tensor obj_11_gamma_0_to_fp16 = const()[name = tensor("obj_11_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15304960)))]; + tensor obj_11_beta_0_to_fp16 = const()[name = tensor("obj_11_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15306048)))]; + tensor obj_11_epsilon_0_to_fp16 = const()[name = tensor("obj_11_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_11_cast_fp16 = batch_norm(beta = obj_11_beta_0_to_fp16, epsilon = obj_11_epsilon_0_to_fp16, gamma = obj_11_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_23_cast_fp16)[name = tensor("obj_11_cast_fp16")]; + tensor var_1008_pad_type_0 = const()[name = tensor("op_1008_pad_type_0"), val = tensor("valid")]; + tensor var_1008_strides_0 = const()[name = tensor("op_1008_strides_0"), val = tensor([1, 1])]; + tensor var_1008_pad_0 = const()[name = tensor("op_1008_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1008_dilations_0 = const()[name = tensor("op_1008_dilations_0"), val = tensor([1, 1])]; + tensor var_1008_groups_0 = const()[name = tensor("op_1008_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15307136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15503808))), name = tensor("layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_2_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15504000)))]; + tensor var_1008_cast_fp16 = conv(bias = layers_2_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1008_dilations_0, groups = var_1008_groups_0, pad = var_1008_pad_0, pad_type = var_1008_pad_type_0, strides = var_1008_strides_0, weight = layers_2_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1008_cast_fp16")]; + tensor var_1014_pad_type_0 = const()[name = tensor("op_1014_pad_type_0"), val = tensor("valid")]; + tensor var_1014_strides_0 = const()[name = tensor("op_1014_strides_0"), val = tensor([1, 1])]; + tensor var_1014_pad_0 = const()[name = tensor("op_1014_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1014_dilations_0 = const()[name = tensor("op_1014_dilations_0"), val = tensor([1, 1])]; + tensor var_1014_groups_0 = const()[name = tensor("op_1014_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15513472))), name = tensor("layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15505088))), shape = tensor([512, 512, 1, 1])]; + tensor var_1014_cast_fp16 = conv(dilations = var_1014_dilations_0, groups = var_1014_groups_0, pad = var_1014_pad_0, pad_type = var_1014_pad_type_0, strides = var_1014_strides_0, weight = layers_2_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1014_cast_fp16")]; + tensor query_9_cast_fp16 = add(x = var_1008_cast_fp16, y = var_1014_cast_fp16)[name = tensor("query_9_cast_fp16")]; + tensor var_1023_pad_type_0 = const()[name = tensor("op_1023_pad_type_0"), val = tensor("valid")]; + tensor var_1023_strides_0 = const()[name = tensor("op_1023_strides_0"), val = tensor([1, 1])]; + tensor var_1023_pad_0 = const()[name = tensor("op_1023_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1023_dilations_0 = const()[name = tensor("op_1023_dilations_0"), val = tensor([1, 1])]; + tensor var_1023_groups_0 = const()[name = tensor("op_1023_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15546304))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15742976))), name = tensor("layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1023_cast_fp16 = conv(dilations = var_1023_dilations_0, groups = var_1023_groups_0, pad = var_1023_pad_0, pad_type = var_1023_pad_type_0, strides = var_1023_strides_0, weight = layers_2_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1023_cast_fp16")]; + tensor var_1029_pad_type_0 = const()[name = tensor("op_1029_pad_type_0"), val = tensor("valid")]; + tensor var_1029_strides_0 = const()[name = tensor("op_1029_strides_0"), val = tensor([1, 1])]; + tensor var_1029_pad_0 = const()[name = tensor("op_1029_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1029_dilations_0 = const()[name = tensor("op_1029_dilations_0"), val = tensor([1, 1])]; + tensor var_1029_groups_0 = const()[name = tensor("op_1029_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15751936))), name = tensor("layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15743168))), shape = tensor([512, 512, 1, 1])]; + tensor var_1029_cast_fp16 = conv(dilations = var_1029_dilations_0, groups = var_1029_groups_0, pad = var_1029_pad_0, pad_type = var_1029_pad_type_0, strides = var_1029_strides_0, weight = layers_2_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1029_cast_fp16")]; + tensor key_5_cast_fp16 = add(x = var_1023_cast_fp16, y = var_1029_cast_fp16)[name = tensor("key_5_cast_fp16")]; + tensor var_1039_pad_type_0 = const()[name = tensor("op_1039_pad_type_0"), val = tensor("valid")]; + tensor var_1039_strides_0 = const()[name = tensor("op_1039_strides_0"), val = tensor([1, 1])]; + tensor var_1039_pad_0 = const()[name = tensor("op_1039_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1039_dilations_0 = const()[name = tensor("op_1039_dilations_0"), val = tensor([1, 1])]; + tensor var_1039_groups_0 = const()[name = tensor("op_1039_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15784768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15981440))), name = tensor("layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_2_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15981632)))]; + tensor var_1039_cast_fp16 = conv(bias = layers_2_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1039_dilations_0, groups = var_1039_groups_0, pad = var_1039_pad_0, pad_type = var_1039_pad_type_0, strides = var_1039_strides_0, weight = layers_2_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_11_cast_fp16)[name = tensor("op_1039_cast_fp16")]; + tensor var_1045_pad_type_0 = const()[name = tensor("op_1045_pad_type_0"), val = tensor("valid")]; + tensor var_1045_strides_0 = const()[name = tensor("op_1045_strides_0"), val = tensor([1, 1])]; + tensor var_1045_pad_0 = const()[name = tensor("op_1045_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1045_dilations_0 = const()[name = tensor("op_1045_dilations_0"), val = tensor([1, 1])]; + tensor var_1045_groups_0 = const()[name = tensor("op_1045_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15990208))), name = tensor("layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(15982720))), shape = tensor([512, 512, 1, 1])]; + tensor var_1045_cast_fp16 = conv(dilations = var_1045_dilations_0, groups = var_1045_groups_0, pad = var_1045_pad_0, pad_type = var_1045_pad_type_0, strides = var_1045_strides_0, weight = layers_2_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_11_cast_fp16)[name = tensor("op_1045_cast_fp16")]; + tensor value_5_cast_fp16 = add(x = var_1039_cast_fp16, y = var_1045_cast_fp16)[name = tensor("value_5_cast_fp16")]; + tensor var_1048_to_fp16 = const()[name = tensor("op_1048_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16023040)))]; + tensor query_11_cast_fp16 = add(x = query_9_cast_fp16, y = var_1048_to_fp16)[name = tensor("query_11_cast_fp16")]; + tensor var_1051_to_fp16 = const()[name = tensor("op_1051_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16024128)))]; + tensor q_with_bias_v_5_cast_fp16 = add(x = query_9_cast_fp16, y = var_1051_to_fp16)[name = tensor("q_with_bias_v_5_cast_fp16")]; + tensor var_1061_pad_type_0 = const()[name = tensor("op_1061_pad_type_0"), val = tensor("valid")]; + tensor var_1061_strides_0 = const()[name = tensor("op_1061_strides_0"), val = tensor([1, 1])]; + tensor var_1061_pad_0 = const()[name = tensor("op_1061_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1061_dilations_0 = const()[name = tensor("op_1061_dilations_0"), val = tensor([1, 1])]; + tensor var_1061_groups_0 = const()[name = tensor("op_1061_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16025216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16221888))), name = tensor("layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1061_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1061_dilations_0, groups = var_1061_groups_0, pad = var_1061_pad_0, pad_type = var_1061_pad_type_0, strides = var_1061_strides_0, weight = layers_2_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1061_cast_fp16")]; + tensor var_1067_pad_type_0 = const()[name = tensor("op_1067_pad_type_0"), val = tensor("valid")]; + tensor var_1067_strides_0 = const()[name = tensor("op_1067_strides_0"), val = tensor([1, 1])]; + tensor var_1067_pad_0 = const()[name = tensor("op_1067_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1067_dilations_0 = const()[name = tensor("op_1067_dilations_0"), val = tensor([1, 1])]; + tensor var_1067_groups_0 = const()[name = tensor("op_1067_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16243968))), name = tensor("layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16222080))), shape = tensor([512, 512, 1, 1])]; + tensor var_1067_cast_fp16 = conv(dilations = var_1067_dilations_0, groups = var_1067_groups_0, pad = var_1067_pad_0, pad_type = var_1067_pad_type_0, strides = var_1067_strides_0, weight = layers_2_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1067_cast_fp16")]; + tensor p_5_cast_fp16 = add(x = var_1061_cast_fp16, y = var_1067_cast_fp16)[name = tensor("p_5_cast_fp16")]; + tensor var_1071 = const()[name = tensor("op_1071"), val = tensor([1, 8, 64, 188])]; + tensor var_1072_cast_fp16 = reshape(shape = var_1071, x = q_with_bias_v_5_cast_fp16)[name = tensor("op_1072_cast_fp16")]; + tensor var_1073 = const()[name = tensor("op_1073"), val = tensor([1, 8, 64, -1])]; + tensor var_1074_cast_fp16 = reshape(shape = var_1073, x = p_5_cast_fp16)[name = tensor("op_1074_cast_fp16")]; + tensor matrix_bd_17_transpose_x_0 = const()[name = tensor("matrix_bd_17_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_17_transpose_y_0 = const()[name = tensor("matrix_bd_17_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_17_cast_fp16 = matmul(transpose_x = matrix_bd_17_transpose_x_0, transpose_y = matrix_bd_17_transpose_y_0, x = var_1072_cast_fp16, y = var_1074_cast_fp16)[name = tensor("matrix_bd_17_cast_fp16")]; + tensor matrix_bd_19_pad_0 = const()[name = tensor("matrix_bd_19_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_19_mode_0 = const()[name = tensor("matrix_bd_19_mode_0"), val = tensor("constant")]; + tensor const_32_to_fp16 = const()[name = tensor("const_32_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_19_cast_fp16 = pad(constant_val = const_32_to_fp16, mode = matrix_bd_19_mode_0, pad = matrix_bd_19_pad_0, x = matrix_bd_17_cast_fp16)[name = tensor("matrix_bd_19_cast_fp16")]; + tensor var_1083 = const()[name = tensor("op_1083"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_21_cast_fp16 = reshape(shape = var_1083, x = matrix_bd_19_cast_fp16)[name = tensor("matrix_bd_21_cast_fp16")]; + tensor var_1087_begin_0 = const()[name = tensor("op_1087_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1087_end_0 = const()[name = tensor("op_1087_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1087_end_mask_0 = const()[name = tensor("op_1087_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1087_cast_fp16 = slice_by_index(begin = var_1087_begin_0, end = var_1087_end_0, end_mask = var_1087_end_mask_0, x = matrix_bd_21_cast_fp16)[name = tensor("op_1087_cast_fp16")]; + tensor var_1088 = const()[name = tensor("op_1088"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_23_cast_fp16 = reshape(shape = var_1088, x = var_1087_cast_fp16)[name = tensor("matrix_bd_23_cast_fp16")]; + tensor var_1093_begin_0 = const()[name = tensor("op_1093_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1093_end_0 = const()[name = tensor("op_1093_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1093_end_mask_0 = const()[name = tensor("op_1093_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1093_cast_fp16 = slice_by_index(begin = var_1093_begin_0, end = var_1093_end_0, end_mask = var_1093_end_mask_0, x = matrix_bd_23_cast_fp16)[name = tensor("op_1093_cast_fp16")]; + tensor var_1094_to_fp16 = const()[name = tensor("op_1094_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_5_cast_fp16 = mul(x = var_1093_cast_fp16, y = var_1094_to_fp16)[name = tensor("qk_mask_5_cast_fp16")]; + tensor var_1098 = const()[name = tensor("op_1098"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_5_cast_fp16 = reshape(shape = var_1098, x = query_11_cast_fp16)[name = tensor("mh_q_5_cast_fp16")]; + tensor var_1100_to_fp16 = const()[name = tensor("op_1100_to_fp16"), val = tensor(0x1p-3)]; + tensor var_1101_cast_fp16 = mul(x = mh_q_5_cast_fp16, y = var_1100_to_fp16)[name = tensor("op_1101_cast_fp16")]; + tensor var_1104 = const()[name = tensor("op_1104"), val = tensor([1, 8, 64, 188])]; + tensor var_1105_cast_fp16 = reshape(shape = var_1104, x = key_5_cast_fp16)[name = tensor("op_1105_cast_fp16")]; + tensor mh_w_9_transpose_x_0 = const()[name = tensor("mh_w_9_transpose_x_0"), val = tensor(true)]; + tensor mh_w_9_transpose_y_0 = const()[name = tensor("mh_w_9_transpose_y_0"), val = tensor(false)]; + tensor mh_w_9_cast_fp16 = matmul(transpose_x = mh_w_9_transpose_x_0, transpose_y = mh_w_9_transpose_y_0, x = var_1101_cast_fp16, y = var_1105_cast_fp16)[name = tensor("mh_w_9_cast_fp16")]; + tensor mh_w_11_cast_fp16 = add(x = mh_w_9_cast_fp16, y = qk_mask_5_cast_fp16)[name = tensor("mh_w_11_cast_fp16")]; + tensor var_1109_cast_fp16 = softmax(axis = var_896, x = mh_w_11_cast_fp16)[name = tensor("op_1109_cast_fp16")]; + tensor var_1110 = const()[name = tensor("op_1110"), val = tensor([1, 8, 64, 188])]; + tensor var_1111_cast_fp16 = reshape(shape = var_1110, x = value_5_cast_fp16)[name = tensor("op_1111_cast_fp16")]; + tensor attn_5_transpose_x_0 = const()[name = tensor("attn_5_transpose_x_0"), val = tensor(false)]; + tensor attn_5_transpose_y_0 = const()[name = tensor("attn_5_transpose_y_0"), val = tensor(true)]; + tensor attn_5_cast_fp16 = matmul(transpose_x = attn_5_transpose_x_0, transpose_y = attn_5_transpose_y_0, x = var_1111_cast_fp16, y = var_1109_cast_fp16)[name = tensor("attn_5_cast_fp16")]; + tensor var_1114 = const()[name = tensor("op_1114"), val = tensor([1, 512, 1, 188])]; + tensor input_75_cast_fp16 = reshape(shape = var_1114, x = attn_5_cast_fp16)[name = tensor("input_75_cast_fp16")]; + tensor var_1124_pad_type_0 = const()[name = tensor("op_1124_pad_type_0"), val = tensor("valid")]; + tensor var_1124_strides_0 = const()[name = tensor("op_1124_strides_0"), val = tensor([1, 1])]; + tensor var_1124_pad_0 = const()[name = tensor("op_1124_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1124_dilations_0 = const()[name = tensor("op_1124_dilations_0"), val = tensor([1, 1])]; + tensor var_1124_groups_0 = const()[name = tensor("op_1124_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16276800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16473472))), name = tensor("layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_2_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16473664)))]; + tensor var_1124_cast_fp16 = conv(bias = layers_2_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1124_dilations_0, groups = var_1124_groups_0, pad = var_1124_pad_0, pad_type = var_1124_pad_type_0, strides = var_1124_strides_0, weight = layers_2_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_75_cast_fp16)[name = tensor("op_1124_cast_fp16")]; + tensor var_1130_pad_type_0 = const()[name = tensor("op_1130_pad_type_0"), val = tensor("valid")]; + tensor var_1130_strides_0 = const()[name = tensor("op_1130_strides_0"), val = tensor([1, 1])]; + tensor var_1130_pad_0 = const()[name = tensor("op_1130_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1130_dilations_0 = const()[name = tensor("op_1130_dilations_0"), val = tensor([1, 1])]; + tensor var_1130_groups_0 = const()[name = tensor("op_1130_groups_0"), val = tensor(1)]; + tensor layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16482560))), name = tensor("layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16474752))), shape = tensor([512, 512, 1, 1])]; + tensor var_1130_cast_fp16 = conv(dilations = var_1130_dilations_0, groups = var_1130_groups_0, pad = var_1130_pad_0, pad_type = var_1130_pad_type_0, strides = var_1130_strides_0, weight = layers_2_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_75_cast_fp16)[name = tensor("op_1130_cast_fp16")]; + tensor obj_13_cast_fp16 = add(x = var_1124_cast_fp16, y = var_1130_cast_fp16)[name = tensor("obj_13_cast_fp16")]; + tensor inputs_25_cast_fp16 = add(x = inputs_23_cast_fp16, y = obj_13_cast_fp16)[name = tensor("inputs_25_cast_fp16")]; + tensor out_25_axes_0 = const()[name = tensor("out_25_axes_0"), val = tensor([1])]; + tensor var_1141_to_fp16 = const()[name = tensor("op_1141_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_25_cast_fp16 = layer_norm(axes = out_25_axes_0, epsilon = var_1141_to_fp16, x = inputs_25_cast_fp16)[name = tensor("out_25_cast_fp16")]; + tensor input_77_gamma_0_to_fp16 = const()[name = tensor("input_77_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16515392)))]; + tensor input_77_beta_0_to_fp16 = const()[name = tensor("input_77_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16516480)))]; + tensor input_77_epsilon_0_to_fp16 = const()[name = tensor("input_77_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_77_cast_fp16 = batch_norm(beta = input_77_beta_0_to_fp16, epsilon = input_77_epsilon_0_to_fp16, gamma = input_77_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_25_cast_fp16)[name = tensor("input_77_cast_fp16")]; + tensor var_1163_pad_type_0 = const()[name = tensor("op_1163_pad_type_0"), val = tensor("valid")]; + tensor var_1163_strides_0 = const()[name = tensor("op_1163_strides_0"), val = tensor([1, 1])]; + tensor var_1163_pad_0 = const()[name = tensor("op_1163_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1163_dilations_0 = const()[name = tensor("op_1163_dilations_0"), val = tensor([1, 1])]; + tensor var_1163_groups_0 = const()[name = tensor("op_1163_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16517568))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16910848))), name = tensor("layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_2_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16911040)))]; + tensor var_1163_cast_fp16 = conv(bias = layers_2_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_1163_dilations_0, groups = var_1163_groups_0, pad = var_1163_pad_0, pad_type = var_1163_pad_type_0, strides = var_1163_strides_0, weight = layers_2_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_77_cast_fp16)[name = tensor("op_1163_cast_fp16")]; + tensor var_1169_pad_type_0 = const()[name = tensor("op_1169_pad_type_0"), val = tensor("valid")]; + tensor var_1169_strides_0 = const()[name = tensor("op_1169_strides_0"), val = tensor([1, 1])]; + tensor var_1169_pad_0 = const()[name = tensor("op_1169_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1169_dilations_0 = const()[name = tensor("op_1169_dilations_0"), val = tensor([1, 1])]; + tensor var_1169_groups_0 = const()[name = tensor("op_1169_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16930752))), name = tensor("layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16913152))), shape = tensor([1024, 512, 1, 1])]; + tensor var_1169_cast_fp16 = conv(dilations = var_1169_dilations_0, groups = var_1169_groups_0, pad = var_1169_pad_0, pad_type = var_1169_pad_type_0, strides = var_1169_strides_0, weight = layers_2_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_77_cast_fp16)[name = tensor("op_1169_cast_fp16")]; + tensor input_79_cast_fp16 = add(x = var_1163_cast_fp16, y = var_1169_cast_fp16)[name = tensor("input_79_cast_fp16")]; + tensor input_81_split_num_splits_0 = const()[name = tensor("input_81_split_num_splits_0"), val = tensor(2)]; + tensor input_81_split_axis_0 = const()[name = tensor("input_81_split_axis_0"), val = tensor(1)]; + tensor input_81_split_cast_fp16_0, tensor input_81_split_cast_fp16_1 = split(axis = input_81_split_axis_0, num_splits = input_81_split_num_splits_0, x = input_79_cast_fp16)[name = tensor("input_81_split_cast_fp16")]; + tensor input_81_split_1_sigmoid_cast_fp16 = sigmoid(x = input_81_split_cast_fp16_1)[name = tensor("input_81_split_1_sigmoid_cast_fp16")]; + tensor input_81_cast_fp16 = mul(x = input_81_split_cast_fp16_0, y = input_81_split_1_sigmoid_cast_fp16)[name = tensor("input_81_cast_fp16")]; + tensor input_83_pad_type_0 = const()[name = tensor("input_83_pad_type_0"), val = tensor("custom")]; + tensor input_83_pad_0 = const()[name = tensor("input_83_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_83_groups_0 = const()[name = tensor("input_83_groups_0"), val = tensor(512)]; + tensor input_83_strides_0 = const()[name = tensor("input_83_strides_0"), val = tensor([1, 1])]; + tensor input_83_dilations_0 = const()[name = tensor("input_83_dilations_0"), val = tensor([1, 1])]; + tensor const_195_to_fp16 = const()[name = tensor("const_195_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(16996352)))]; + tensor const_196_to_fp16 = const()[name = tensor("const_196_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17005632)))]; + tensor input_85_cast_fp16 = conv(bias = const_196_to_fp16, dilations = input_83_dilations_0, groups = input_83_groups_0, pad = input_83_pad_0, pad_type = input_83_pad_type_0, strides = input_83_strides_0, weight = const_195_to_fp16, x = input_81_cast_fp16)[name = tensor("input_85_cast_fp16")]; + tensor input_87_cast_fp16 = silu(x = input_85_cast_fp16)[name = tensor("input_87_cast_fp16")]; + tensor var_1193_pad_type_0 = const()[name = tensor("op_1193_pad_type_0"), val = tensor("valid")]; + tensor var_1193_strides_0 = const()[name = tensor("op_1193_strides_0"), val = tensor([1, 1])]; + tensor var_1193_pad_0 = const()[name = tensor("op_1193_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1193_dilations_0 = const()[name = tensor("op_1193_dilations_0"), val = tensor([1, 1])]; + tensor var_1193_groups_0 = const()[name = tensor("op_1193_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17006720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17203392))), name = tensor("layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_2_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17203584)))]; + tensor var_1193_cast_fp16 = conv(bias = layers_2_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_1193_dilations_0, groups = var_1193_groups_0, pad = var_1193_pad_0, pad_type = var_1193_pad_type_0, strides = var_1193_strides_0, weight = layers_2_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_87_cast_fp16)[name = tensor("op_1193_cast_fp16")]; + tensor var_1199_pad_type_0 = const()[name = tensor("op_1199_pad_type_0"), val = tensor("valid")]; + tensor var_1199_strides_0 = const()[name = tensor("op_1199_strides_0"), val = tensor([1, 1])]; + tensor var_1199_pad_0 = const()[name = tensor("op_1199_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1199_dilations_0 = const()[name = tensor("op_1199_dilations_0"), val = tensor([1, 1])]; + tensor var_1199_groups_0 = const()[name = tensor("op_1199_groups_0"), val = tensor(1)]; + tensor layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17213248))), name = tensor("layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17204672))), shape = tensor([512, 512, 1, 1])]; + tensor var_1199_cast_fp16 = conv(dilations = var_1199_dilations_0, groups = var_1199_groups_0, pad = var_1199_pad_0, pad_type = var_1199_pad_type_0, strides = var_1199_strides_0, weight = layers_2_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_87_cast_fp16)[name = tensor("op_1199_cast_fp16")]; + tensor x_17_cast_fp16 = add(x = var_1193_cast_fp16, y = var_1199_cast_fp16)[name = tensor("x_17_cast_fp16")]; + tensor inputs_27_cast_fp16 = add(x = inputs_25_cast_fp16, y = x_17_cast_fp16)[name = tensor("inputs_27_cast_fp16")]; + tensor out_27_axes_0 = const()[name = tensor("out_27_axes_0"), val = tensor([1])]; + tensor var_1210_to_fp16 = const()[name = tensor("op_1210_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_27_cast_fp16 = layer_norm(axes = out_27_axes_0, epsilon = var_1210_to_fp16, x = inputs_27_cast_fp16)[name = tensor("out_27_cast_fp16")]; + tensor input_89_gamma_0_to_fp16 = const()[name = tensor("input_89_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17246080)))]; + tensor input_89_beta_0_to_fp16 = const()[name = tensor("input_89_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17247168)))]; + tensor input_89_epsilon_0_to_fp16 = const()[name = tensor("input_89_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_89_cast_fp16 = batch_norm(beta = input_89_beta_0_to_fp16, epsilon = input_89_epsilon_0_to_fp16, gamma = input_89_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_27_cast_fp16)[name = tensor("input_89_cast_fp16")]; + tensor var_1230_pad_type_0 = const()[name = tensor("op_1230_pad_type_0"), val = tensor("valid")]; + tensor var_1230_strides_0 = const()[name = tensor("op_1230_strides_0"), val = tensor([1, 1])]; + tensor var_1230_pad_0 = const()[name = tensor("op_1230_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1230_dilations_0 = const()[name = tensor("op_1230_dilations_0"), val = tensor([1, 1])]; + tensor var_1230_groups_0 = const()[name = tensor("op_1230_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(17248256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18034752))), name = tensor("layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_2_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18034944)))]; + tensor var_1230_cast_fp16 = conv(bias = layers_2_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_1230_dilations_0, groups = var_1230_groups_0, pad = var_1230_pad_0, pad_type = var_1230_pad_type_0, strides = var_1230_strides_0, weight = layers_2_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_89_cast_fp16)[name = tensor("op_1230_cast_fp16")]; + tensor var_1236_pad_type_0 = const()[name = tensor("op_1236_pad_type_0"), val = tensor("valid")]; + tensor var_1236_strides_0 = const()[name = tensor("op_1236_strides_0"), val = tensor([1, 1])]; + tensor var_1236_pad_0 = const()[name = tensor("op_1236_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1236_dilations_0 = const()[name = tensor("op_1236_dilations_0"), val = tensor([1, 1])]; + tensor var_1236_groups_0 = const()[name = tensor("op_1236_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18070784))), name = tensor("layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18039104))), shape = tensor([2048, 512, 1, 1])]; + tensor var_1236_cast_fp16 = conv(dilations = var_1236_dilations_0, groups = var_1236_groups_0, pad = var_1236_pad_0, pad_type = var_1236_pad_type_0, strides = var_1236_strides_0, weight = layers_2_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_89_cast_fp16)[name = tensor("op_1236_cast_fp16")]; + tensor input_91_cast_fp16 = add(x = var_1230_cast_fp16, y = var_1236_cast_fp16)[name = tensor("input_91_cast_fp16")]; + tensor input_93_cast_fp16 = silu(x = input_91_cast_fp16)[name = tensor("input_93_cast_fp16")]; + tensor var_1247_pad_type_0 = const()[name = tensor("op_1247_pad_type_0"), val = tensor("valid")]; + tensor var_1247_strides_0 = const()[name = tensor("op_1247_strides_0"), val = tensor([1, 1])]; + tensor var_1247_pad_0 = const()[name = tensor("op_1247_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1247_dilations_0 = const()[name = tensor("op_1247_dilations_0"), val = tensor([1, 1])]; + tensor var_1247_groups_0 = const()[name = tensor("op_1247_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18201920))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18988416))), name = tensor("layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_2_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_2_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18988608)))]; + tensor var_1247_cast_fp16 = conv(bias = layers_2_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_1247_dilations_0, groups = var_1247_groups_0, pad = var_1247_pad_0, pad_type = var_1247_pad_type_0, strides = var_1247_strides_0, weight = layers_2_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_93_cast_fp16)[name = tensor("op_1247_cast_fp16")]; + tensor var_1253_pad_type_0 = const()[name = tensor("op_1253_pad_type_0"), val = tensor("valid")]; + tensor var_1253_strides_0 = const()[name = tensor("op_1253_strides_0"), val = tensor([1, 1])]; + tensor var_1253_pad_0 = const()[name = tensor("op_1253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1253_dilations_0 = const()[name = tensor("op_1253_dilations_0"), val = tensor([1, 1])]; + tensor var_1253_groups_0 = const()[name = tensor("op_1253_groups_0"), val = tensor(1)]; + tensor layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19025600))), name = tensor("layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(18989696))), shape = tensor([512, 2048, 1, 1])]; + tensor var_1253_cast_fp16 = conv(dilations = var_1253_dilations_0, groups = var_1253_groups_0, pad = var_1253_pad_0, pad_type = var_1253_pad_type_0, strides = var_1253_strides_0, weight = layers_2_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_93_cast_fp16)[name = tensor("op_1253_cast_fp16")]; + tensor x_19_cast_fp16 = add(x = var_1247_cast_fp16, y = var_1253_cast_fp16)[name = tensor("x_19_cast_fp16")]; + tensor var_1255_to_fp16 = const()[name = tensor("op_1255_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1256_cast_fp16 = mul(x = x_19_cast_fp16, y = var_1255_to_fp16)[name = tensor("op_1256_cast_fp16")]; + tensor inputs_29_cast_fp16 = add(x = inputs_27_cast_fp16, y = var_1256_cast_fp16)[name = tensor("inputs_29_cast_fp16")]; + tensor out_29_axes_0 = const()[name = tensor("out_29_axes_0"), val = tensor([1])]; + tensor var_1266_to_fp16 = const()[name = tensor("op_1266_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_29_cast_fp16 = layer_norm(axes = out_29_axes_0, epsilon = var_1266_to_fp16, x = inputs_29_cast_fp16)[name = tensor("out_29_cast_fp16")]; + tensor inputs_31_gamma_0_to_fp16 = const()[name = tensor("inputs_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19156736)))]; + tensor inputs_31_beta_0_to_fp16 = const()[name = tensor("inputs_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19157824)))]; + tensor inputs_31_epsilon_0_to_fp16 = const()[name = tensor("inputs_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_31_cast_fp16 = batch_norm(beta = inputs_31_beta_0_to_fp16, epsilon = inputs_31_epsilon_0_to_fp16, gamma = inputs_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_29_cast_fp16)[name = tensor("inputs_31_cast_fp16")]; + tensor var_1280 = const()[name = tensor("op_1280"), val = tensor(3)]; + tensor out_31_axes_0 = const()[name = tensor("out_31_axes_0"), val = tensor([1])]; + tensor var_1311_to_fp16 = const()[name = tensor("op_1311_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_31_cast_fp16 = layer_norm(axes = out_31_axes_0, epsilon = var_1311_to_fp16, x = inputs_31_cast_fp16)[name = tensor("out_31_cast_fp16")]; + tensor input_95_gamma_0_to_fp16 = const()[name = tensor("input_95_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19158912)))]; + tensor input_95_beta_0_to_fp16 = const()[name = tensor("input_95_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19160000)))]; + tensor input_95_epsilon_0_to_fp16 = const()[name = tensor("input_95_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_95_cast_fp16 = batch_norm(beta = input_95_beta_0_to_fp16, epsilon = input_95_epsilon_0_to_fp16, gamma = input_95_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_31_cast_fp16)[name = tensor("input_95_cast_fp16")]; + tensor var_1331_pad_type_0 = const()[name = tensor("op_1331_pad_type_0"), val = tensor("valid")]; + tensor var_1331_strides_0 = const()[name = tensor("op_1331_strides_0"), val = tensor([1, 1])]; + tensor var_1331_pad_0 = const()[name = tensor("op_1331_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1331_dilations_0 = const()[name = tensor("op_1331_dilations_0"), val = tensor([1, 1])]; + tensor var_1331_groups_0 = const()[name = tensor("op_1331_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19161088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19947584))), name = tensor("layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_3_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19947776)))]; + tensor var_1331_cast_fp16 = conv(bias = layers_3_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1331_dilations_0, groups = var_1331_groups_0, pad = var_1331_pad_0, pad_type = var_1331_pad_type_0, strides = var_1331_strides_0, weight = layers_3_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_95_cast_fp16)[name = tensor("op_1331_cast_fp16")]; + tensor var_1337_pad_type_0 = const()[name = tensor("op_1337_pad_type_0"), val = tensor("valid")]; + tensor var_1337_strides_0 = const()[name = tensor("op_1337_strides_0"), val = tensor([1, 1])]; + tensor var_1337_pad_0 = const()[name = tensor("op_1337_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1337_dilations_0 = const()[name = tensor("op_1337_dilations_0"), val = tensor([1, 1])]; + tensor var_1337_groups_0 = const()[name = tensor("op_1337_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19984128))), name = tensor("layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(19951936))), shape = tensor([2048, 512, 1, 1])]; + tensor var_1337_cast_fp16 = conv(dilations = var_1337_dilations_0, groups = var_1337_groups_0, pad = var_1337_pad_0, pad_type = var_1337_pad_type_0, strides = var_1337_strides_0, weight = layers_3_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_95_cast_fp16)[name = tensor("op_1337_cast_fp16")]; + tensor input_97_cast_fp16 = add(x = var_1331_cast_fp16, y = var_1337_cast_fp16)[name = tensor("input_97_cast_fp16")]; + tensor input_99_cast_fp16 = silu(x = input_97_cast_fp16)[name = tensor("input_99_cast_fp16")]; + tensor var_1348_pad_type_0 = const()[name = tensor("op_1348_pad_type_0"), val = tensor("valid")]; + tensor var_1348_strides_0 = const()[name = tensor("op_1348_strides_0"), val = tensor([1, 1])]; + tensor var_1348_pad_0 = const()[name = tensor("op_1348_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1348_dilations_0 = const()[name = tensor("op_1348_dilations_0"), val = tensor([1, 1])]; + tensor var_1348_groups_0 = const()[name = tensor("op_1348_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20115264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20901760))), name = tensor("layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_3_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20901952)))]; + tensor var_1348_cast_fp16 = conv(bias = layers_3_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_1348_dilations_0, groups = var_1348_groups_0, pad = var_1348_pad_0, pad_type = var_1348_pad_type_0, strides = var_1348_strides_0, weight = layers_3_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_99_cast_fp16)[name = tensor("op_1348_cast_fp16")]; + tensor var_1354_pad_type_0 = const()[name = tensor("op_1354_pad_type_0"), val = tensor("valid")]; + tensor var_1354_strides_0 = const()[name = tensor("op_1354_strides_0"), val = tensor([1, 1])]; + tensor var_1354_pad_0 = const()[name = tensor("op_1354_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1354_dilations_0 = const()[name = tensor("op_1354_dilations_0"), val = tensor([1, 1])]; + tensor var_1354_groups_0 = const()[name = tensor("op_1354_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20938176))), name = tensor("layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(20903040))), shape = tensor([512, 2048, 1, 1])]; + tensor var_1354_cast_fp16 = conv(dilations = var_1354_dilations_0, groups = var_1354_groups_0, pad = var_1354_pad_0, pad_type = var_1354_pad_type_0, strides = var_1354_strides_0, weight = layers_3_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_99_cast_fp16)[name = tensor("op_1354_cast_fp16")]; + tensor x_21_cast_fp16 = add(x = var_1348_cast_fp16, y = var_1354_cast_fp16)[name = tensor("x_21_cast_fp16")]; + tensor var_1356_to_fp16 = const()[name = tensor("op_1356_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1357_cast_fp16 = mul(x = x_21_cast_fp16, y = var_1356_to_fp16)[name = tensor("op_1357_cast_fp16")]; + tensor inputs_33_cast_fp16 = add(x = inputs_31_cast_fp16, y = var_1357_cast_fp16)[name = tensor("inputs_33_cast_fp16")]; + tensor out_33_axes_0 = const()[name = tensor("out_33_axes_0"), val = tensor([1])]; + tensor var_1367_to_fp16 = const()[name = tensor("op_1367_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_33_cast_fp16 = layer_norm(axes = out_33_axes_0, epsilon = var_1367_to_fp16, x = inputs_33_cast_fp16)[name = tensor("out_33_cast_fp16")]; + tensor obj_15_gamma_0_to_fp16 = const()[name = tensor("obj_15_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21069312)))]; + tensor obj_15_beta_0_to_fp16 = const()[name = tensor("obj_15_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21070400)))]; + tensor obj_15_epsilon_0_to_fp16 = const()[name = tensor("obj_15_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_15_cast_fp16 = batch_norm(beta = obj_15_beta_0_to_fp16, epsilon = obj_15_epsilon_0_to_fp16, gamma = obj_15_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_33_cast_fp16)[name = tensor("obj_15_cast_fp16")]; + tensor var_1392_pad_type_0 = const()[name = tensor("op_1392_pad_type_0"), val = tensor("valid")]; + tensor var_1392_strides_0 = const()[name = tensor("op_1392_strides_0"), val = tensor([1, 1])]; + tensor var_1392_pad_0 = const()[name = tensor("op_1392_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1392_dilations_0 = const()[name = tensor("op_1392_dilations_0"), val = tensor([1, 1])]; + tensor var_1392_groups_0 = const()[name = tensor("op_1392_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21071488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21268160))), name = tensor("layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_3_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21268352)))]; + tensor var_1392_cast_fp16 = conv(bias = layers_3_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1392_dilations_0, groups = var_1392_groups_0, pad = var_1392_pad_0, pad_type = var_1392_pad_type_0, strides = var_1392_strides_0, weight = layers_3_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1392_cast_fp16")]; + tensor var_1398_pad_type_0 = const()[name = tensor("op_1398_pad_type_0"), val = tensor("valid")]; + tensor var_1398_strides_0 = const()[name = tensor("op_1398_strides_0"), val = tensor([1, 1])]; + tensor var_1398_pad_0 = const()[name = tensor("op_1398_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1398_dilations_0 = const()[name = tensor("op_1398_dilations_0"), val = tensor([1, 1])]; + tensor var_1398_groups_0 = const()[name = tensor("op_1398_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21278208))), name = tensor("layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21269440))), shape = tensor([512, 512, 1, 1])]; + tensor var_1398_cast_fp16 = conv(dilations = var_1398_dilations_0, groups = var_1398_groups_0, pad = var_1398_pad_0, pad_type = var_1398_pad_type_0, strides = var_1398_strides_0, weight = layers_3_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1398_cast_fp16")]; + tensor query_13_cast_fp16 = add(x = var_1392_cast_fp16, y = var_1398_cast_fp16)[name = tensor("query_13_cast_fp16")]; + tensor var_1407_pad_type_0 = const()[name = tensor("op_1407_pad_type_0"), val = tensor("valid")]; + tensor var_1407_strides_0 = const()[name = tensor("op_1407_strides_0"), val = tensor([1, 1])]; + tensor var_1407_pad_0 = const()[name = tensor("op_1407_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1407_dilations_0 = const()[name = tensor("op_1407_dilations_0"), val = tensor([1, 1])]; + tensor var_1407_groups_0 = const()[name = tensor("op_1407_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21311040))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21507712))), name = tensor("layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1407_cast_fp16 = conv(dilations = var_1407_dilations_0, groups = var_1407_groups_0, pad = var_1407_pad_0, pad_type = var_1407_pad_type_0, strides = var_1407_strides_0, weight = layers_3_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1407_cast_fp16")]; + tensor var_1413_pad_type_0 = const()[name = tensor("op_1413_pad_type_0"), val = tensor("valid")]; + tensor var_1413_strides_0 = const()[name = tensor("op_1413_strides_0"), val = tensor([1, 1])]; + tensor var_1413_pad_0 = const()[name = tensor("op_1413_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1413_dilations_0 = const()[name = tensor("op_1413_dilations_0"), val = tensor([1, 1])]; + tensor var_1413_groups_0 = const()[name = tensor("op_1413_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21517824))), name = tensor("layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21507904))), shape = tensor([512, 512, 1, 1])]; + tensor var_1413_cast_fp16 = conv(dilations = var_1413_dilations_0, groups = var_1413_groups_0, pad = var_1413_pad_0, pad_type = var_1413_pad_type_0, strides = var_1413_strides_0, weight = layers_3_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1413_cast_fp16")]; + tensor key_7_cast_fp16 = add(x = var_1407_cast_fp16, y = var_1413_cast_fp16)[name = tensor("key_7_cast_fp16")]; + tensor var_1423_pad_type_0 = const()[name = tensor("op_1423_pad_type_0"), val = tensor("valid")]; + tensor var_1423_strides_0 = const()[name = tensor("op_1423_strides_0"), val = tensor([1, 1])]; + tensor var_1423_pad_0 = const()[name = tensor("op_1423_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1423_dilations_0 = const()[name = tensor("op_1423_dilations_0"), val = tensor([1, 1])]; + tensor var_1423_groups_0 = const()[name = tensor("op_1423_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21550656))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21747328))), name = tensor("layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_3_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21747520)))]; + tensor var_1423_cast_fp16 = conv(bias = layers_3_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1423_dilations_0, groups = var_1423_groups_0, pad = var_1423_pad_0, pad_type = var_1423_pad_type_0, strides = var_1423_strides_0, weight = layers_3_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_15_cast_fp16)[name = tensor("op_1423_cast_fp16")]; + tensor var_1429_pad_type_0 = const()[name = tensor("op_1429_pad_type_0"), val = tensor("valid")]; + tensor var_1429_strides_0 = const()[name = tensor("op_1429_strides_0"), val = tensor([1, 1])]; + tensor var_1429_pad_0 = const()[name = tensor("op_1429_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1429_dilations_0 = const()[name = tensor("op_1429_dilations_0"), val = tensor([1, 1])]; + tensor var_1429_groups_0 = const()[name = tensor("op_1429_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21756352))), name = tensor("layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21748608))), shape = tensor([512, 512, 1, 1])]; + tensor var_1429_cast_fp16 = conv(dilations = var_1429_dilations_0, groups = var_1429_groups_0, pad = var_1429_pad_0, pad_type = var_1429_pad_type_0, strides = var_1429_strides_0, weight = layers_3_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_15_cast_fp16)[name = tensor("op_1429_cast_fp16")]; + tensor value_7_cast_fp16 = add(x = var_1423_cast_fp16, y = var_1429_cast_fp16)[name = tensor("value_7_cast_fp16")]; + tensor var_1432_to_fp16 = const()[name = tensor("op_1432_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21789184)))]; + tensor query_15_cast_fp16 = add(x = query_13_cast_fp16, y = var_1432_to_fp16)[name = tensor("query_15_cast_fp16")]; + tensor var_1435_to_fp16 = const()[name = tensor("op_1435_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21790272)))]; + tensor q_with_bias_v_7_cast_fp16 = add(x = query_13_cast_fp16, y = var_1435_to_fp16)[name = tensor("q_with_bias_v_7_cast_fp16")]; + tensor var_1445_pad_type_0 = const()[name = tensor("op_1445_pad_type_0"), val = tensor("valid")]; + tensor var_1445_strides_0 = const()[name = tensor("op_1445_strides_0"), val = tensor([1, 1])]; + tensor var_1445_pad_0 = const()[name = tensor("op_1445_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1445_dilations_0 = const()[name = tensor("op_1445_dilations_0"), val = tensor([1, 1])]; + tensor var_1445_groups_0 = const()[name = tensor("op_1445_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21791360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21988032))), name = tensor("layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1445_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1445_dilations_0, groups = var_1445_groups_0, pad = var_1445_pad_0, pad_type = var_1445_pad_type_0, strides = var_1445_strides_0, weight = layers_3_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1445_cast_fp16")]; + tensor var_1451_pad_type_0 = const()[name = tensor("op_1451_pad_type_0"), val = tensor("valid")]; + tensor var_1451_strides_0 = const()[name = tensor("op_1451_strides_0"), val = tensor([1, 1])]; + tensor var_1451_pad_0 = const()[name = tensor("op_1451_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1451_dilations_0 = const()[name = tensor("op_1451_dilations_0"), val = tensor([1, 1])]; + tensor var_1451_groups_0 = const()[name = tensor("op_1451_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22007936))), name = tensor("layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(21988224))), shape = tensor([512, 512, 1, 1])]; + tensor var_1451_cast_fp16 = conv(dilations = var_1451_dilations_0, groups = var_1451_groups_0, pad = var_1451_pad_0, pad_type = var_1451_pad_type_0, strides = var_1451_strides_0, weight = layers_3_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1451_cast_fp16")]; + tensor p_7_cast_fp16 = add(x = var_1445_cast_fp16, y = var_1451_cast_fp16)[name = tensor("p_7_cast_fp16")]; + tensor var_1455 = const()[name = tensor("op_1455"), val = tensor([1, 8, 64, 188])]; + tensor var_1456_cast_fp16 = reshape(shape = var_1455, x = q_with_bias_v_7_cast_fp16)[name = tensor("op_1456_cast_fp16")]; + tensor var_1457 = const()[name = tensor("op_1457"), val = tensor([1, 8, 64, -1])]; + tensor var_1458_cast_fp16 = reshape(shape = var_1457, x = p_7_cast_fp16)[name = tensor("op_1458_cast_fp16")]; + tensor matrix_bd_25_transpose_x_0 = const()[name = tensor("matrix_bd_25_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_25_transpose_y_0 = const()[name = tensor("matrix_bd_25_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_25_cast_fp16 = matmul(transpose_x = matrix_bd_25_transpose_x_0, transpose_y = matrix_bd_25_transpose_y_0, x = var_1456_cast_fp16, y = var_1458_cast_fp16)[name = tensor("matrix_bd_25_cast_fp16")]; + tensor matrix_bd_27_pad_0 = const()[name = tensor("matrix_bd_27_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_27_mode_0 = const()[name = tensor("matrix_bd_27_mode_0"), val = tensor("constant")]; + tensor const_43_to_fp16 = const()[name = tensor("const_43_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_27_cast_fp16 = pad(constant_val = const_43_to_fp16, mode = matrix_bd_27_mode_0, pad = matrix_bd_27_pad_0, x = matrix_bd_25_cast_fp16)[name = tensor("matrix_bd_27_cast_fp16")]; + tensor var_1467 = const()[name = tensor("op_1467"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_29_cast_fp16 = reshape(shape = var_1467, x = matrix_bd_27_cast_fp16)[name = tensor("matrix_bd_29_cast_fp16")]; + tensor var_1471_begin_0 = const()[name = tensor("op_1471_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1471_end_0 = const()[name = tensor("op_1471_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1471_end_mask_0 = const()[name = tensor("op_1471_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1471_cast_fp16 = slice_by_index(begin = var_1471_begin_0, end = var_1471_end_0, end_mask = var_1471_end_mask_0, x = matrix_bd_29_cast_fp16)[name = tensor("op_1471_cast_fp16")]; + tensor var_1472 = const()[name = tensor("op_1472"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_31_cast_fp16 = reshape(shape = var_1472, x = var_1471_cast_fp16)[name = tensor("matrix_bd_31_cast_fp16")]; + tensor var_1477_begin_0 = const()[name = tensor("op_1477_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1477_end_0 = const()[name = tensor("op_1477_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1477_end_mask_0 = const()[name = tensor("op_1477_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1477_cast_fp16 = slice_by_index(begin = var_1477_begin_0, end = var_1477_end_0, end_mask = var_1477_end_mask_0, x = matrix_bd_31_cast_fp16)[name = tensor("op_1477_cast_fp16")]; + tensor var_1478_to_fp16 = const()[name = tensor("op_1478_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_7_cast_fp16 = mul(x = var_1477_cast_fp16, y = var_1478_to_fp16)[name = tensor("qk_mask_7_cast_fp16")]; + tensor var_1482 = const()[name = tensor("op_1482"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_7_cast_fp16 = reshape(shape = var_1482, x = query_15_cast_fp16)[name = tensor("mh_q_7_cast_fp16")]; + tensor var_1484_to_fp16 = const()[name = tensor("op_1484_to_fp16"), val = tensor(0x1p-3)]; + tensor var_1485_cast_fp16 = mul(x = mh_q_7_cast_fp16, y = var_1484_to_fp16)[name = tensor("op_1485_cast_fp16")]; + tensor var_1488 = const()[name = tensor("op_1488"), val = tensor([1, 8, 64, 188])]; + tensor var_1489_cast_fp16 = reshape(shape = var_1488, x = key_7_cast_fp16)[name = tensor("op_1489_cast_fp16")]; + tensor mh_w_13_transpose_x_0 = const()[name = tensor("mh_w_13_transpose_x_0"), val = tensor(true)]; + tensor mh_w_13_transpose_y_0 = const()[name = tensor("mh_w_13_transpose_y_0"), val = tensor(false)]; + tensor mh_w_13_cast_fp16 = matmul(transpose_x = mh_w_13_transpose_x_0, transpose_y = mh_w_13_transpose_y_0, x = var_1485_cast_fp16, y = var_1489_cast_fp16)[name = tensor("mh_w_13_cast_fp16")]; + tensor mh_w_15_cast_fp16 = add(x = mh_w_13_cast_fp16, y = qk_mask_7_cast_fp16)[name = tensor("mh_w_15_cast_fp16")]; + tensor var_1493_cast_fp16 = softmax(axis = var_1280, x = mh_w_15_cast_fp16)[name = tensor("op_1493_cast_fp16")]; + tensor var_1494 = const()[name = tensor("op_1494"), val = tensor([1, 8, 64, 188])]; + tensor var_1495_cast_fp16 = reshape(shape = var_1494, x = value_7_cast_fp16)[name = tensor("op_1495_cast_fp16")]; + tensor attn_7_transpose_x_0 = const()[name = tensor("attn_7_transpose_x_0"), val = tensor(false)]; + tensor attn_7_transpose_y_0 = const()[name = tensor("attn_7_transpose_y_0"), val = tensor(true)]; + tensor attn_7_cast_fp16 = matmul(transpose_x = attn_7_transpose_x_0, transpose_y = attn_7_transpose_y_0, x = var_1495_cast_fp16, y = var_1493_cast_fp16)[name = tensor("attn_7_cast_fp16")]; + tensor var_1498 = const()[name = tensor("op_1498"), val = tensor([1, 512, 1, 188])]; + tensor input_101_cast_fp16 = reshape(shape = var_1498, x = attn_7_cast_fp16)[name = tensor("input_101_cast_fp16")]; + tensor var_1508_pad_type_0 = const()[name = tensor("op_1508_pad_type_0"), val = tensor("valid")]; + tensor var_1508_strides_0 = const()[name = tensor("op_1508_strides_0"), val = tensor([1, 1])]; + tensor var_1508_pad_0 = const()[name = tensor("op_1508_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1508_dilations_0 = const()[name = tensor("op_1508_dilations_0"), val = tensor([1, 1])]; + tensor var_1508_groups_0 = const()[name = tensor("op_1508_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22040768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22237440))), name = tensor("layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_3_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22237632)))]; + tensor var_1508_cast_fp16 = conv(bias = layers_3_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1508_dilations_0, groups = var_1508_groups_0, pad = var_1508_pad_0, pad_type = var_1508_pad_type_0, strides = var_1508_strides_0, weight = layers_3_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_101_cast_fp16)[name = tensor("op_1508_cast_fp16")]; + tensor var_1514_pad_type_0 = const()[name = tensor("op_1514_pad_type_0"), val = tensor("valid")]; + tensor var_1514_strides_0 = const()[name = tensor("op_1514_strides_0"), val = tensor([1, 1])]; + tensor var_1514_pad_0 = const()[name = tensor("op_1514_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1514_dilations_0 = const()[name = tensor("op_1514_dilations_0"), val = tensor([1, 1])]; + tensor var_1514_groups_0 = const()[name = tensor("op_1514_groups_0"), val = tensor(1)]; + tensor layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22247104))), name = tensor("layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22238720))), shape = tensor([512, 512, 1, 1])]; + tensor var_1514_cast_fp16 = conv(dilations = var_1514_dilations_0, groups = var_1514_groups_0, pad = var_1514_pad_0, pad_type = var_1514_pad_type_0, strides = var_1514_strides_0, weight = layers_3_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_101_cast_fp16)[name = tensor("op_1514_cast_fp16")]; + tensor obj_17_cast_fp16 = add(x = var_1508_cast_fp16, y = var_1514_cast_fp16)[name = tensor("obj_17_cast_fp16")]; + tensor inputs_35_cast_fp16 = add(x = inputs_33_cast_fp16, y = obj_17_cast_fp16)[name = tensor("inputs_35_cast_fp16")]; + tensor out_35_axes_0 = const()[name = tensor("out_35_axes_0"), val = tensor([1])]; + tensor var_1525_to_fp16 = const()[name = tensor("op_1525_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_35_cast_fp16 = layer_norm(axes = out_35_axes_0, epsilon = var_1525_to_fp16, x = inputs_35_cast_fp16)[name = tensor("out_35_cast_fp16")]; + tensor input_103_gamma_0_to_fp16 = const()[name = tensor("input_103_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22279936)))]; + tensor input_103_beta_0_to_fp16 = const()[name = tensor("input_103_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22281024)))]; + tensor input_103_epsilon_0_to_fp16 = const()[name = tensor("input_103_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_103_cast_fp16 = batch_norm(beta = input_103_beta_0_to_fp16, epsilon = input_103_epsilon_0_to_fp16, gamma = input_103_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_35_cast_fp16)[name = tensor("input_103_cast_fp16")]; + tensor var_1547_pad_type_0 = const()[name = tensor("op_1547_pad_type_0"), val = tensor("valid")]; + tensor var_1547_strides_0 = const()[name = tensor("op_1547_strides_0"), val = tensor([1, 1])]; + tensor var_1547_pad_0 = const()[name = tensor("op_1547_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1547_dilations_0 = const()[name = tensor("op_1547_dilations_0"), val = tensor([1, 1])]; + tensor var_1547_groups_0 = const()[name = tensor("op_1547_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22282112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22675392))), name = tensor("layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_3_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22675584)))]; + tensor var_1547_cast_fp16 = conv(bias = layers_3_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_1547_dilations_0, groups = var_1547_groups_0, pad = var_1547_pad_0, pad_type = var_1547_pad_type_0, strides = var_1547_strides_0, weight = layers_3_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_103_cast_fp16)[name = tensor("op_1547_cast_fp16")]; + tensor var_1553_pad_type_0 = const()[name = tensor("op_1553_pad_type_0"), val = tensor("valid")]; + tensor var_1553_strides_0 = const()[name = tensor("op_1553_strides_0"), val = tensor([1, 1])]; + tensor var_1553_pad_0 = const()[name = tensor("op_1553_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1553_dilations_0 = const()[name = tensor("op_1553_dilations_0"), val = tensor([1, 1])]; + tensor var_1553_groups_0 = const()[name = tensor("op_1553_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22695360))), name = tensor("layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22677696))), shape = tensor([1024, 512, 1, 1])]; + tensor var_1553_cast_fp16 = conv(dilations = var_1553_dilations_0, groups = var_1553_groups_0, pad = var_1553_pad_0, pad_type = var_1553_pad_type_0, strides = var_1553_strides_0, weight = layers_3_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_103_cast_fp16)[name = tensor("op_1553_cast_fp16")]; + tensor input_105_cast_fp16 = add(x = var_1547_cast_fp16, y = var_1553_cast_fp16)[name = tensor("input_105_cast_fp16")]; + tensor input_107_split_num_splits_0 = const()[name = tensor("input_107_split_num_splits_0"), val = tensor(2)]; + tensor input_107_split_axis_0 = const()[name = tensor("input_107_split_axis_0"), val = tensor(1)]; + tensor input_107_split_cast_fp16_0, tensor input_107_split_cast_fp16_1 = split(axis = input_107_split_axis_0, num_splits = input_107_split_num_splits_0, x = input_105_cast_fp16)[name = tensor("input_107_split_cast_fp16")]; + tensor input_107_split_1_sigmoid_cast_fp16 = sigmoid(x = input_107_split_cast_fp16_1)[name = tensor("input_107_split_1_sigmoid_cast_fp16")]; + tensor input_107_cast_fp16 = mul(x = input_107_split_cast_fp16_0, y = input_107_split_1_sigmoid_cast_fp16)[name = tensor("input_107_cast_fp16")]; + tensor input_109_pad_type_0 = const()[name = tensor("input_109_pad_type_0"), val = tensor("custom")]; + tensor input_109_pad_0 = const()[name = tensor("input_109_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_109_groups_0 = const()[name = tensor("input_109_groups_0"), val = tensor(512)]; + tensor input_109_strides_0 = const()[name = tensor("input_109_strides_0"), val = tensor([1, 1])]; + tensor input_109_dilations_0 = const()[name = tensor("input_109_dilations_0"), val = tensor([1, 1])]; + tensor const_197_to_fp16 = const()[name = tensor("const_197_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22760960)))]; + tensor const_198_to_fp16 = const()[name = tensor("const_198_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22770240)))]; + tensor input_111_cast_fp16 = conv(bias = const_198_to_fp16, dilations = input_109_dilations_0, groups = input_109_groups_0, pad = input_109_pad_0, pad_type = input_109_pad_type_0, strides = input_109_strides_0, weight = const_197_to_fp16, x = input_107_cast_fp16)[name = tensor("input_111_cast_fp16")]; + tensor input_113_cast_fp16 = silu(x = input_111_cast_fp16)[name = tensor("input_113_cast_fp16")]; + tensor var_1577_pad_type_0 = const()[name = tensor("op_1577_pad_type_0"), val = tensor("valid")]; + tensor var_1577_strides_0 = const()[name = tensor("op_1577_strides_0"), val = tensor([1, 1])]; + tensor var_1577_pad_0 = const()[name = tensor("op_1577_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1577_dilations_0 = const()[name = tensor("op_1577_dilations_0"), val = tensor([1, 1])]; + tensor var_1577_groups_0 = const()[name = tensor("op_1577_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22771328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22968000))), name = tensor("layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_3_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22968192)))]; + tensor var_1577_cast_fp16 = conv(bias = layers_3_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_1577_dilations_0, groups = var_1577_groups_0, pad = var_1577_pad_0, pad_type = var_1577_pad_type_0, strides = var_1577_strides_0, weight = layers_3_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_113_cast_fp16)[name = tensor("op_1577_cast_fp16")]; + tensor var_1583_pad_type_0 = const()[name = tensor("op_1583_pad_type_0"), val = tensor("valid")]; + tensor var_1583_strides_0 = const()[name = tensor("op_1583_strides_0"), val = tensor([1, 1])]; + tensor var_1583_pad_0 = const()[name = tensor("op_1583_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1583_dilations_0 = const()[name = tensor("op_1583_dilations_0"), val = tensor([1, 1])]; + tensor var_1583_groups_0 = const()[name = tensor("op_1583_groups_0"), val = tensor(1)]; + tensor layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22978048))), name = tensor("layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(22969280))), shape = tensor([512, 512, 1, 1])]; + tensor var_1583_cast_fp16 = conv(dilations = var_1583_dilations_0, groups = var_1583_groups_0, pad = var_1583_pad_0, pad_type = var_1583_pad_type_0, strides = var_1583_strides_0, weight = layers_3_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_113_cast_fp16)[name = tensor("op_1583_cast_fp16")]; + tensor x_23_cast_fp16 = add(x = var_1577_cast_fp16, y = var_1583_cast_fp16)[name = tensor("x_23_cast_fp16")]; + tensor inputs_37_cast_fp16 = add(x = inputs_35_cast_fp16, y = x_23_cast_fp16)[name = tensor("inputs_37_cast_fp16")]; + tensor out_37_axes_0 = const()[name = tensor("out_37_axes_0"), val = tensor([1])]; + tensor var_1594_to_fp16 = const()[name = tensor("op_1594_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_37_cast_fp16 = layer_norm(axes = out_37_axes_0, epsilon = var_1594_to_fp16, x = inputs_37_cast_fp16)[name = tensor("out_37_cast_fp16")]; + tensor input_115_gamma_0_to_fp16 = const()[name = tensor("input_115_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23010880)))]; + tensor input_115_beta_0_to_fp16 = const()[name = tensor("input_115_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23011968)))]; + tensor input_115_epsilon_0_to_fp16 = const()[name = tensor("input_115_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_115_cast_fp16 = batch_norm(beta = input_115_beta_0_to_fp16, epsilon = input_115_epsilon_0_to_fp16, gamma = input_115_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_37_cast_fp16)[name = tensor("input_115_cast_fp16")]; + tensor var_1614_pad_type_0 = const()[name = tensor("op_1614_pad_type_0"), val = tensor("valid")]; + tensor var_1614_strides_0 = const()[name = tensor("op_1614_strides_0"), val = tensor([1, 1])]; + tensor var_1614_pad_0 = const()[name = tensor("op_1614_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1614_dilations_0 = const()[name = tensor("op_1614_dilations_0"), val = tensor([1, 1])]; + tensor var_1614_groups_0 = const()[name = tensor("op_1614_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23013056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23799552))), name = tensor("layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_3_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23799744)))]; + tensor var_1614_cast_fp16 = conv(bias = layers_3_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_1614_dilations_0, groups = var_1614_groups_0, pad = var_1614_pad_0, pad_type = var_1614_pad_type_0, strides = var_1614_strides_0, weight = layers_3_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_115_cast_fp16)[name = tensor("op_1614_cast_fp16")]; + tensor var_1620_pad_type_0 = const()[name = tensor("op_1620_pad_type_0"), val = tensor("valid")]; + tensor var_1620_strides_0 = const()[name = tensor("op_1620_strides_0"), val = tensor([1, 1])]; + tensor var_1620_pad_0 = const()[name = tensor("op_1620_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1620_dilations_0 = const()[name = tensor("op_1620_dilations_0"), val = tensor([1, 1])]; + tensor var_1620_groups_0 = const()[name = tensor("op_1620_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23835840))), name = tensor("layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23803904))), shape = tensor([2048, 512, 1, 1])]; + tensor var_1620_cast_fp16 = conv(dilations = var_1620_dilations_0, groups = var_1620_groups_0, pad = var_1620_pad_0, pad_type = var_1620_pad_type_0, strides = var_1620_strides_0, weight = layers_3_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_115_cast_fp16)[name = tensor("op_1620_cast_fp16")]; + tensor input_117_cast_fp16 = add(x = var_1614_cast_fp16, y = var_1620_cast_fp16)[name = tensor("input_117_cast_fp16")]; + tensor input_119_cast_fp16 = silu(x = input_117_cast_fp16)[name = tensor("input_119_cast_fp16")]; + tensor var_1631_pad_type_0 = const()[name = tensor("op_1631_pad_type_0"), val = tensor("valid")]; + tensor var_1631_strides_0 = const()[name = tensor("op_1631_strides_0"), val = tensor([1, 1])]; + tensor var_1631_pad_0 = const()[name = tensor("op_1631_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1631_dilations_0 = const()[name = tensor("op_1631_dilations_0"), val = tensor([1, 1])]; + tensor var_1631_groups_0 = const()[name = tensor("op_1631_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(23966976))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24753472))), name = tensor("layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_3_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_3_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24753664)))]; + tensor var_1631_cast_fp16 = conv(bias = layers_3_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_1631_dilations_0, groups = var_1631_groups_0, pad = var_1631_pad_0, pad_type = var_1631_pad_type_0, strides = var_1631_strides_0, weight = layers_3_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_119_cast_fp16)[name = tensor("op_1631_cast_fp16")]; + tensor var_1637_pad_type_0 = const()[name = tensor("op_1637_pad_type_0"), val = tensor("valid")]; + tensor var_1637_strides_0 = const()[name = tensor("op_1637_strides_0"), val = tensor([1, 1])]; + tensor var_1637_pad_0 = const()[name = tensor("op_1637_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1637_dilations_0 = const()[name = tensor("op_1637_dilations_0"), val = tensor([1, 1])]; + tensor var_1637_groups_0 = const()[name = tensor("op_1637_groups_0"), val = tensor(1)]; + tensor layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24790848))), name = tensor("layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24754752))), shape = tensor([512, 2048, 1, 1])]; + tensor var_1637_cast_fp16 = conv(dilations = var_1637_dilations_0, groups = var_1637_groups_0, pad = var_1637_pad_0, pad_type = var_1637_pad_type_0, strides = var_1637_strides_0, weight = layers_3_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_119_cast_fp16)[name = tensor("op_1637_cast_fp16")]; + tensor x_25_cast_fp16 = add(x = var_1631_cast_fp16, y = var_1637_cast_fp16)[name = tensor("x_25_cast_fp16")]; + tensor var_1639_to_fp16 = const()[name = tensor("op_1639_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1640_cast_fp16 = mul(x = x_25_cast_fp16, y = var_1639_to_fp16)[name = tensor("op_1640_cast_fp16")]; + tensor inputs_39_cast_fp16 = add(x = inputs_37_cast_fp16, y = var_1640_cast_fp16)[name = tensor("inputs_39_cast_fp16")]; + tensor out_39_axes_0 = const()[name = tensor("out_39_axes_0"), val = tensor([1])]; + tensor var_1650_to_fp16 = const()[name = tensor("op_1650_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_39_cast_fp16 = layer_norm(axes = out_39_axes_0, epsilon = var_1650_to_fp16, x = inputs_39_cast_fp16)[name = tensor("out_39_cast_fp16")]; + tensor inputs_41_gamma_0_to_fp16 = const()[name = tensor("inputs_41_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24921984)))]; + tensor inputs_41_beta_0_to_fp16 = const()[name = tensor("inputs_41_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24923072)))]; + tensor inputs_41_epsilon_0_to_fp16 = const()[name = tensor("inputs_41_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_41_cast_fp16 = batch_norm(beta = inputs_41_beta_0_to_fp16, epsilon = inputs_41_epsilon_0_to_fp16, gamma = inputs_41_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_39_cast_fp16)[name = tensor("inputs_41_cast_fp16")]; + tensor var_1664 = const()[name = tensor("op_1664"), val = tensor(3)]; + tensor out_41_axes_0 = const()[name = tensor("out_41_axes_0"), val = tensor([1])]; + tensor var_1695_to_fp16 = const()[name = tensor("op_1695_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_41_cast_fp16 = layer_norm(axes = out_41_axes_0, epsilon = var_1695_to_fp16, x = inputs_41_cast_fp16)[name = tensor("out_41_cast_fp16")]; + tensor input_121_gamma_0_to_fp16 = const()[name = tensor("input_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24924160)))]; + tensor input_121_beta_0_to_fp16 = const()[name = tensor("input_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24925248)))]; + tensor input_121_epsilon_0_to_fp16 = const()[name = tensor("input_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_121_cast_fp16 = batch_norm(beta = input_121_beta_0_to_fp16, epsilon = input_121_epsilon_0_to_fp16, gamma = input_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_41_cast_fp16)[name = tensor("input_121_cast_fp16")]; + tensor var_1715_pad_type_0 = const()[name = tensor("op_1715_pad_type_0"), val = tensor("valid")]; + tensor var_1715_strides_0 = const()[name = tensor("op_1715_strides_0"), val = tensor([1, 1])]; + tensor var_1715_pad_0 = const()[name = tensor("op_1715_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1715_dilations_0 = const()[name = tensor("op_1715_dilations_0"), val = tensor([1, 1])]; + tensor var_1715_groups_0 = const()[name = tensor("op_1715_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(24926336))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25712832))), name = tensor("layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_4_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25713024)))]; + tensor var_1715_cast_fp16 = conv(bias = layers_4_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_1715_dilations_0, groups = var_1715_groups_0, pad = var_1715_pad_0, pad_type = var_1715_pad_type_0, strides = var_1715_strides_0, weight = layers_4_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_121_cast_fp16)[name = tensor("op_1715_cast_fp16")]; + tensor var_1721_pad_type_0 = const()[name = tensor("op_1721_pad_type_0"), val = tensor("valid")]; + tensor var_1721_strides_0 = const()[name = tensor("op_1721_strides_0"), val = tensor([1, 1])]; + tensor var_1721_pad_0 = const()[name = tensor("op_1721_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1721_dilations_0 = const()[name = tensor("op_1721_dilations_0"), val = tensor([1, 1])]; + tensor var_1721_groups_0 = const()[name = tensor("op_1721_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25749248))), name = tensor("layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25717184))), shape = tensor([2048, 512, 1, 1])]; + tensor var_1721_cast_fp16 = conv(dilations = var_1721_dilations_0, groups = var_1721_groups_0, pad = var_1721_pad_0, pad_type = var_1721_pad_type_0, strides = var_1721_strides_0, weight = layers_4_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_121_cast_fp16)[name = tensor("op_1721_cast_fp16")]; + tensor input_123_cast_fp16 = add(x = var_1715_cast_fp16, y = var_1721_cast_fp16)[name = tensor("input_123_cast_fp16")]; + tensor input_125_cast_fp16 = silu(x = input_123_cast_fp16)[name = tensor("input_125_cast_fp16")]; + tensor var_1732_pad_type_0 = const()[name = tensor("op_1732_pad_type_0"), val = tensor("valid")]; + tensor var_1732_strides_0 = const()[name = tensor("op_1732_strides_0"), val = tensor([1, 1])]; + tensor var_1732_pad_0 = const()[name = tensor("op_1732_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1732_dilations_0 = const()[name = tensor("op_1732_dilations_0"), val = tensor([1, 1])]; + tensor var_1732_groups_0 = const()[name = tensor("op_1732_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(25880384))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26666880))), name = tensor("layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_4_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26667072)))]; + tensor var_1732_cast_fp16 = conv(bias = layers_4_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_1732_dilations_0, groups = var_1732_groups_0, pad = var_1732_pad_0, pad_type = var_1732_pad_type_0, strides = var_1732_strides_0, weight = layers_4_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_125_cast_fp16)[name = tensor("op_1732_cast_fp16")]; + tensor var_1738_pad_type_0 = const()[name = tensor("op_1738_pad_type_0"), val = tensor("valid")]; + tensor var_1738_strides_0 = const()[name = tensor("op_1738_strides_0"), val = tensor([1, 1])]; + tensor var_1738_pad_0 = const()[name = tensor("op_1738_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1738_dilations_0 = const()[name = tensor("op_1738_dilations_0"), val = tensor([1, 1])]; + tensor var_1738_groups_0 = const()[name = tensor("op_1738_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26704896))), name = tensor("layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26668160))), shape = tensor([512, 2048, 1, 1])]; + tensor var_1738_cast_fp16 = conv(dilations = var_1738_dilations_0, groups = var_1738_groups_0, pad = var_1738_pad_0, pad_type = var_1738_pad_type_0, strides = var_1738_strides_0, weight = layers_4_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_125_cast_fp16)[name = tensor("op_1738_cast_fp16")]; + tensor x_27_cast_fp16 = add(x = var_1732_cast_fp16, y = var_1738_cast_fp16)[name = tensor("x_27_cast_fp16")]; + tensor var_1740_to_fp16 = const()[name = tensor("op_1740_to_fp16"), val = tensor(0x1p-1)]; + tensor var_1741_cast_fp16 = mul(x = x_27_cast_fp16, y = var_1740_to_fp16)[name = tensor("op_1741_cast_fp16")]; + tensor inputs_43_cast_fp16 = add(x = inputs_41_cast_fp16, y = var_1741_cast_fp16)[name = tensor("inputs_43_cast_fp16")]; + tensor out_43_axes_0 = const()[name = tensor("out_43_axes_0"), val = tensor([1])]; + tensor var_1751_to_fp16 = const()[name = tensor("op_1751_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_43_cast_fp16 = layer_norm(axes = out_43_axes_0, epsilon = var_1751_to_fp16, x = inputs_43_cast_fp16)[name = tensor("out_43_cast_fp16")]; + tensor obj_19_gamma_0_to_fp16 = const()[name = tensor("obj_19_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26836032)))]; + tensor obj_19_beta_0_to_fp16 = const()[name = tensor("obj_19_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26837120)))]; + tensor obj_19_epsilon_0_to_fp16 = const()[name = tensor("obj_19_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_19_cast_fp16 = batch_norm(beta = obj_19_beta_0_to_fp16, epsilon = obj_19_epsilon_0_to_fp16, gamma = obj_19_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_43_cast_fp16)[name = tensor("obj_19_cast_fp16")]; + tensor var_1776_pad_type_0 = const()[name = tensor("op_1776_pad_type_0"), val = tensor("valid")]; + tensor var_1776_strides_0 = const()[name = tensor("op_1776_strides_0"), val = tensor([1, 1])]; + tensor var_1776_pad_0 = const()[name = tensor("op_1776_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1776_dilations_0 = const()[name = tensor("op_1776_dilations_0"), val = tensor([1, 1])]; + tensor var_1776_groups_0 = const()[name = tensor("op_1776_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(26838208))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27034880))), name = tensor("layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_4_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27035072)))]; + tensor var_1776_cast_fp16 = conv(bias = layers_4_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_1776_dilations_0, groups = var_1776_groups_0, pad = var_1776_pad_0, pad_type = var_1776_pad_type_0, strides = var_1776_strides_0, weight = layers_4_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1776_cast_fp16")]; + tensor var_1782_pad_type_0 = const()[name = tensor("op_1782_pad_type_0"), val = tensor("valid")]; + tensor var_1782_strides_0 = const()[name = tensor("op_1782_strides_0"), val = tensor([1, 1])]; + tensor var_1782_pad_0 = const()[name = tensor("op_1782_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1782_dilations_0 = const()[name = tensor("op_1782_dilations_0"), val = tensor([1, 1])]; + tensor var_1782_groups_0 = const()[name = tensor("op_1782_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27044608))), name = tensor("layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27036160))), shape = tensor([512, 512, 1, 1])]; + tensor var_1782_cast_fp16 = conv(dilations = var_1782_dilations_0, groups = var_1782_groups_0, pad = var_1782_pad_0, pad_type = var_1782_pad_type_0, strides = var_1782_strides_0, weight = layers_4_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1782_cast_fp16")]; + tensor query_17_cast_fp16 = add(x = var_1776_cast_fp16, y = var_1782_cast_fp16)[name = tensor("query_17_cast_fp16")]; + tensor var_1791_pad_type_0 = const()[name = tensor("op_1791_pad_type_0"), val = tensor("valid")]; + tensor var_1791_strides_0 = const()[name = tensor("op_1791_strides_0"), val = tensor([1, 1])]; + tensor var_1791_pad_0 = const()[name = tensor("op_1791_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1791_dilations_0 = const()[name = tensor("op_1791_dilations_0"), val = tensor([1, 1])]; + tensor var_1791_groups_0 = const()[name = tensor("op_1791_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27077440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27274112))), name = tensor("layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1791_cast_fp16 = conv(dilations = var_1791_dilations_0, groups = var_1791_groups_0, pad = var_1791_pad_0, pad_type = var_1791_pad_type_0, strides = var_1791_strides_0, weight = layers_4_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1791_cast_fp16")]; + tensor var_1797_pad_type_0 = const()[name = tensor("op_1797_pad_type_0"), val = tensor("valid")]; + tensor var_1797_strides_0 = const()[name = tensor("op_1797_strides_0"), val = tensor([1, 1])]; + tensor var_1797_pad_0 = const()[name = tensor("op_1797_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1797_dilations_0 = const()[name = tensor("op_1797_dilations_0"), val = tensor([1, 1])]; + tensor var_1797_groups_0 = const()[name = tensor("op_1797_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27283968))), name = tensor("layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27274304))), shape = tensor([512, 512, 1, 1])]; + tensor var_1797_cast_fp16 = conv(dilations = var_1797_dilations_0, groups = var_1797_groups_0, pad = var_1797_pad_0, pad_type = var_1797_pad_type_0, strides = var_1797_strides_0, weight = layers_4_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1797_cast_fp16")]; + tensor key_9_cast_fp16 = add(x = var_1791_cast_fp16, y = var_1797_cast_fp16)[name = tensor("key_9_cast_fp16")]; + tensor var_1807_pad_type_0 = const()[name = tensor("op_1807_pad_type_0"), val = tensor("valid")]; + tensor var_1807_strides_0 = const()[name = tensor("op_1807_strides_0"), val = tensor([1, 1])]; + tensor var_1807_pad_0 = const()[name = tensor("op_1807_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1807_dilations_0 = const()[name = tensor("op_1807_dilations_0"), val = tensor([1, 1])]; + tensor var_1807_groups_0 = const()[name = tensor("op_1807_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27316800))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27513472))), name = tensor("layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_4_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27513664)))]; + tensor var_1807_cast_fp16 = conv(bias = layers_4_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_1807_dilations_0, groups = var_1807_groups_0, pad = var_1807_pad_0, pad_type = var_1807_pad_type_0, strides = var_1807_strides_0, weight = layers_4_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_19_cast_fp16)[name = tensor("op_1807_cast_fp16")]; + tensor var_1813_pad_type_0 = const()[name = tensor("op_1813_pad_type_0"), val = tensor("valid")]; + tensor var_1813_strides_0 = const()[name = tensor("op_1813_strides_0"), val = tensor([1, 1])]; + tensor var_1813_pad_0 = const()[name = tensor("op_1813_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1813_dilations_0 = const()[name = tensor("op_1813_dilations_0"), val = tensor([1, 1])]; + tensor var_1813_groups_0 = const()[name = tensor("op_1813_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27522816))), name = tensor("layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27514752))), shape = tensor([512, 512, 1, 1])]; + tensor var_1813_cast_fp16 = conv(dilations = var_1813_dilations_0, groups = var_1813_groups_0, pad = var_1813_pad_0, pad_type = var_1813_pad_type_0, strides = var_1813_strides_0, weight = layers_4_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_19_cast_fp16)[name = tensor("op_1813_cast_fp16")]; + tensor value_9_cast_fp16 = add(x = var_1807_cast_fp16, y = var_1813_cast_fp16)[name = tensor("value_9_cast_fp16")]; + tensor var_1816_to_fp16 = const()[name = tensor("op_1816_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27555648)))]; + tensor query_19_cast_fp16 = add(x = query_17_cast_fp16, y = var_1816_to_fp16)[name = tensor("query_19_cast_fp16")]; + tensor var_1819_to_fp16 = const()[name = tensor("op_1819_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27556736)))]; + tensor q_with_bias_v_9_cast_fp16 = add(x = query_17_cast_fp16, y = var_1819_to_fp16)[name = tensor("q_with_bias_v_9_cast_fp16")]; + tensor var_1829_pad_type_0 = const()[name = tensor("op_1829_pad_type_0"), val = tensor("valid")]; + tensor var_1829_strides_0 = const()[name = tensor("op_1829_strides_0"), val = tensor([1, 1])]; + tensor var_1829_pad_0 = const()[name = tensor("op_1829_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1829_dilations_0 = const()[name = tensor("op_1829_dilations_0"), val = tensor([1, 1])]; + tensor var_1829_groups_0 = const()[name = tensor("op_1829_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27557824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27754496))), name = tensor("layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_1829_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_1829_dilations_0, groups = var_1829_groups_0, pad = var_1829_pad_0, pad_type = var_1829_pad_type_0, strides = var_1829_strides_0, weight = layers_4_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_1829_cast_fp16")]; + tensor var_1835_pad_type_0 = const()[name = tensor("op_1835_pad_type_0"), val = tensor("valid")]; + tensor var_1835_strides_0 = const()[name = tensor("op_1835_strides_0"), val = tensor([1, 1])]; + tensor var_1835_pad_0 = const()[name = tensor("op_1835_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1835_dilations_0 = const()[name = tensor("op_1835_dilations_0"), val = tensor([1, 1])]; + tensor var_1835_groups_0 = const()[name = tensor("op_1835_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27776128))), name = tensor("layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27754688))), shape = tensor([512, 512, 1, 1])]; + tensor var_1835_cast_fp16 = conv(dilations = var_1835_dilations_0, groups = var_1835_groups_0, pad = var_1835_pad_0, pad_type = var_1835_pad_type_0, strides = var_1835_strides_0, weight = layers_4_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_1835_cast_fp16")]; + tensor p_9_cast_fp16 = add(x = var_1829_cast_fp16, y = var_1835_cast_fp16)[name = tensor("p_9_cast_fp16")]; + tensor var_1839 = const()[name = tensor("op_1839"), val = tensor([1, 8, 64, 188])]; + tensor var_1840_cast_fp16 = reshape(shape = var_1839, x = q_with_bias_v_9_cast_fp16)[name = tensor("op_1840_cast_fp16")]; + tensor var_1841 = const()[name = tensor("op_1841"), val = tensor([1, 8, 64, -1])]; + tensor var_1842_cast_fp16 = reshape(shape = var_1841, x = p_9_cast_fp16)[name = tensor("op_1842_cast_fp16")]; + tensor matrix_bd_33_transpose_x_0 = const()[name = tensor("matrix_bd_33_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_33_transpose_y_0 = const()[name = tensor("matrix_bd_33_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_33_cast_fp16 = matmul(transpose_x = matrix_bd_33_transpose_x_0, transpose_y = matrix_bd_33_transpose_y_0, x = var_1840_cast_fp16, y = var_1842_cast_fp16)[name = tensor("matrix_bd_33_cast_fp16")]; + tensor matrix_bd_35_pad_0 = const()[name = tensor("matrix_bd_35_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_35_mode_0 = const()[name = tensor("matrix_bd_35_mode_0"), val = tensor("constant")]; + tensor const_54_to_fp16 = const()[name = tensor("const_54_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_35_cast_fp16 = pad(constant_val = const_54_to_fp16, mode = matrix_bd_35_mode_0, pad = matrix_bd_35_pad_0, x = matrix_bd_33_cast_fp16)[name = tensor("matrix_bd_35_cast_fp16")]; + tensor var_1851 = const()[name = tensor("op_1851"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_37_cast_fp16 = reshape(shape = var_1851, x = matrix_bd_35_cast_fp16)[name = tensor("matrix_bd_37_cast_fp16")]; + tensor var_1855_begin_0 = const()[name = tensor("op_1855_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_1855_end_0 = const()[name = tensor("op_1855_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_1855_end_mask_0 = const()[name = tensor("op_1855_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_1855_cast_fp16 = slice_by_index(begin = var_1855_begin_0, end = var_1855_end_0, end_mask = var_1855_end_mask_0, x = matrix_bd_37_cast_fp16)[name = tensor("op_1855_cast_fp16")]; + tensor var_1856 = const()[name = tensor("op_1856"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_39_cast_fp16 = reshape(shape = var_1856, x = var_1855_cast_fp16)[name = tensor("matrix_bd_39_cast_fp16")]; + tensor var_1861_begin_0 = const()[name = tensor("op_1861_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1861_end_0 = const()[name = tensor("op_1861_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_1861_end_mask_0 = const()[name = tensor("op_1861_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_1861_cast_fp16 = slice_by_index(begin = var_1861_begin_0, end = var_1861_end_0, end_mask = var_1861_end_mask_0, x = matrix_bd_39_cast_fp16)[name = tensor("op_1861_cast_fp16")]; + tensor var_1862_to_fp16 = const()[name = tensor("op_1862_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_9_cast_fp16 = mul(x = var_1861_cast_fp16, y = var_1862_to_fp16)[name = tensor("qk_mask_9_cast_fp16")]; + tensor var_1866 = const()[name = tensor("op_1866"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_9_cast_fp16 = reshape(shape = var_1866, x = query_19_cast_fp16)[name = tensor("mh_q_9_cast_fp16")]; + tensor var_1868_to_fp16 = const()[name = tensor("op_1868_to_fp16"), val = tensor(0x1p-3)]; + tensor var_1869_cast_fp16 = mul(x = mh_q_9_cast_fp16, y = var_1868_to_fp16)[name = tensor("op_1869_cast_fp16")]; + tensor var_1872 = const()[name = tensor("op_1872"), val = tensor([1, 8, 64, 188])]; + tensor var_1873_cast_fp16 = reshape(shape = var_1872, x = key_9_cast_fp16)[name = tensor("op_1873_cast_fp16")]; + tensor mh_w_17_transpose_x_0 = const()[name = tensor("mh_w_17_transpose_x_0"), val = tensor(true)]; + tensor mh_w_17_transpose_y_0 = const()[name = tensor("mh_w_17_transpose_y_0"), val = tensor(false)]; + tensor mh_w_17_cast_fp16 = matmul(transpose_x = mh_w_17_transpose_x_0, transpose_y = mh_w_17_transpose_y_0, x = var_1869_cast_fp16, y = var_1873_cast_fp16)[name = tensor("mh_w_17_cast_fp16")]; + tensor mh_w_19_cast_fp16 = add(x = mh_w_17_cast_fp16, y = qk_mask_9_cast_fp16)[name = tensor("mh_w_19_cast_fp16")]; + tensor var_1877_cast_fp16 = softmax(axis = var_1664, x = mh_w_19_cast_fp16)[name = tensor("op_1877_cast_fp16")]; + tensor var_1878 = const()[name = tensor("op_1878"), val = tensor([1, 8, 64, 188])]; + tensor var_1879_cast_fp16 = reshape(shape = var_1878, x = value_9_cast_fp16)[name = tensor("op_1879_cast_fp16")]; + tensor attn_9_transpose_x_0 = const()[name = tensor("attn_9_transpose_x_0"), val = tensor(false)]; + tensor attn_9_transpose_y_0 = const()[name = tensor("attn_9_transpose_y_0"), val = tensor(true)]; + tensor attn_9_cast_fp16 = matmul(transpose_x = attn_9_transpose_x_0, transpose_y = attn_9_transpose_y_0, x = var_1879_cast_fp16, y = var_1877_cast_fp16)[name = tensor("attn_9_cast_fp16")]; + tensor var_1882 = const()[name = tensor("op_1882"), val = tensor([1, 512, 1, 188])]; + tensor input_127_cast_fp16 = reshape(shape = var_1882, x = attn_9_cast_fp16)[name = tensor("input_127_cast_fp16")]; + tensor var_1892_pad_type_0 = const()[name = tensor("op_1892_pad_type_0"), val = tensor("valid")]; + tensor var_1892_strides_0 = const()[name = tensor("op_1892_strides_0"), val = tensor([1, 1])]; + tensor var_1892_pad_0 = const()[name = tensor("op_1892_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1892_dilations_0 = const()[name = tensor("op_1892_dilations_0"), val = tensor([1, 1])]; + tensor var_1892_groups_0 = const()[name = tensor("op_1892_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(27808960))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28005632))), name = tensor("layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_4_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28005824)))]; + tensor var_1892_cast_fp16 = conv(bias = layers_4_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_1892_dilations_0, groups = var_1892_groups_0, pad = var_1892_pad_0, pad_type = var_1892_pad_type_0, strides = var_1892_strides_0, weight = layers_4_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_127_cast_fp16)[name = tensor("op_1892_cast_fp16")]; + tensor var_1898_pad_type_0 = const()[name = tensor("op_1898_pad_type_0"), val = tensor("valid")]; + tensor var_1898_strides_0 = const()[name = tensor("op_1898_strides_0"), val = tensor([1, 1])]; + tensor var_1898_pad_0 = const()[name = tensor("op_1898_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1898_dilations_0 = const()[name = tensor("op_1898_dilations_0"), val = tensor([1, 1])]; + tensor var_1898_groups_0 = const()[name = tensor("op_1898_groups_0"), val = tensor(1)]; + tensor layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28014848))), name = tensor("layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28006912))), shape = tensor([512, 512, 1, 1])]; + tensor var_1898_cast_fp16 = conv(dilations = var_1898_dilations_0, groups = var_1898_groups_0, pad = var_1898_pad_0, pad_type = var_1898_pad_type_0, strides = var_1898_strides_0, weight = layers_4_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_127_cast_fp16)[name = tensor("op_1898_cast_fp16")]; + tensor obj_21_cast_fp16 = add(x = var_1892_cast_fp16, y = var_1898_cast_fp16)[name = tensor("obj_21_cast_fp16")]; + tensor inputs_45_cast_fp16 = add(x = inputs_43_cast_fp16, y = obj_21_cast_fp16)[name = tensor("inputs_45_cast_fp16")]; + tensor out_45_axes_0 = const()[name = tensor("out_45_axes_0"), val = tensor([1])]; + tensor var_1909_to_fp16 = const()[name = tensor("op_1909_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_45_cast_fp16 = layer_norm(axes = out_45_axes_0, epsilon = var_1909_to_fp16, x = inputs_45_cast_fp16)[name = tensor("out_45_cast_fp16")]; + tensor input_129_gamma_0_to_fp16 = const()[name = tensor("input_129_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28047680)))]; + tensor input_129_beta_0_to_fp16 = const()[name = tensor("input_129_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28048768)))]; + tensor input_129_epsilon_0_to_fp16 = const()[name = tensor("input_129_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_129_cast_fp16 = batch_norm(beta = input_129_beta_0_to_fp16, epsilon = input_129_epsilon_0_to_fp16, gamma = input_129_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_45_cast_fp16)[name = tensor("input_129_cast_fp16")]; + tensor var_1931_pad_type_0 = const()[name = tensor("op_1931_pad_type_0"), val = tensor("valid")]; + tensor var_1931_strides_0 = const()[name = tensor("op_1931_strides_0"), val = tensor([1, 1])]; + tensor var_1931_pad_0 = const()[name = tensor("op_1931_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1931_dilations_0 = const()[name = tensor("op_1931_dilations_0"), val = tensor([1, 1])]; + tensor var_1931_groups_0 = const()[name = tensor("op_1931_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28049856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28443136))), name = tensor("layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_4_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28443328)))]; + tensor var_1931_cast_fp16 = conv(bias = layers_4_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_1931_dilations_0, groups = var_1931_groups_0, pad = var_1931_pad_0, pad_type = var_1931_pad_type_0, strides = var_1931_strides_0, weight = layers_4_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_129_cast_fp16)[name = tensor("op_1931_cast_fp16")]; + tensor var_1937_pad_type_0 = const()[name = tensor("op_1937_pad_type_0"), val = tensor("valid")]; + tensor var_1937_strides_0 = const()[name = tensor("op_1937_strides_0"), val = tensor([1, 1])]; + tensor var_1937_pad_0 = const()[name = tensor("op_1937_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1937_dilations_0 = const()[name = tensor("op_1937_dilations_0"), val = tensor([1, 1])]; + tensor var_1937_groups_0 = const()[name = tensor("op_1937_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28463360))), name = tensor("layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28445440))), shape = tensor([1024, 512, 1, 1])]; + tensor var_1937_cast_fp16 = conv(dilations = var_1937_dilations_0, groups = var_1937_groups_0, pad = var_1937_pad_0, pad_type = var_1937_pad_type_0, strides = var_1937_strides_0, weight = layers_4_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_129_cast_fp16)[name = tensor("op_1937_cast_fp16")]; + tensor input_131_cast_fp16 = add(x = var_1931_cast_fp16, y = var_1937_cast_fp16)[name = tensor("input_131_cast_fp16")]; + tensor input_133_split_num_splits_0 = const()[name = tensor("input_133_split_num_splits_0"), val = tensor(2)]; + tensor input_133_split_axis_0 = const()[name = tensor("input_133_split_axis_0"), val = tensor(1)]; + tensor input_133_split_cast_fp16_0, tensor input_133_split_cast_fp16_1 = split(axis = input_133_split_axis_0, num_splits = input_133_split_num_splits_0, x = input_131_cast_fp16)[name = tensor("input_133_split_cast_fp16")]; + tensor input_133_split_1_sigmoid_cast_fp16 = sigmoid(x = input_133_split_cast_fp16_1)[name = tensor("input_133_split_1_sigmoid_cast_fp16")]; + tensor input_133_cast_fp16 = mul(x = input_133_split_cast_fp16_0, y = input_133_split_1_sigmoid_cast_fp16)[name = tensor("input_133_cast_fp16")]; + tensor input_135_pad_type_0 = const()[name = tensor("input_135_pad_type_0"), val = tensor("custom")]; + tensor input_135_pad_0 = const()[name = tensor("input_135_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_135_groups_0 = const()[name = tensor("input_135_groups_0"), val = tensor(512)]; + tensor input_135_strides_0 = const()[name = tensor("input_135_strides_0"), val = tensor([1, 1])]; + tensor input_135_dilations_0 = const()[name = tensor("input_135_dilations_0"), val = tensor([1, 1])]; + tensor const_199_to_fp16 = const()[name = tensor("const_199_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28528960)))]; + tensor const_200_to_fp16 = const()[name = tensor("const_200_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28538240)))]; + tensor input_137_cast_fp16 = conv(bias = const_200_to_fp16, dilations = input_135_dilations_0, groups = input_135_groups_0, pad = input_135_pad_0, pad_type = input_135_pad_type_0, strides = input_135_strides_0, weight = const_199_to_fp16, x = input_133_cast_fp16)[name = tensor("input_137_cast_fp16")]; + tensor input_139_cast_fp16 = silu(x = input_137_cast_fp16)[name = tensor("input_139_cast_fp16")]; + tensor var_1961_pad_type_0 = const()[name = tensor("op_1961_pad_type_0"), val = tensor("valid")]; + tensor var_1961_strides_0 = const()[name = tensor("op_1961_strides_0"), val = tensor([1, 1])]; + tensor var_1961_pad_0 = const()[name = tensor("op_1961_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1961_dilations_0 = const()[name = tensor("op_1961_dilations_0"), val = tensor([1, 1])]; + tensor var_1961_groups_0 = const()[name = tensor("op_1961_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28539328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28736000))), name = tensor("layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_4_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28736192)))]; + tensor var_1961_cast_fp16 = conv(bias = layers_4_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_1961_dilations_0, groups = var_1961_groups_0, pad = var_1961_pad_0, pad_type = var_1961_pad_type_0, strides = var_1961_strides_0, weight = layers_4_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_139_cast_fp16)[name = tensor("op_1961_cast_fp16")]; + tensor var_1967_pad_type_0 = const()[name = tensor("op_1967_pad_type_0"), val = tensor("valid")]; + tensor var_1967_strides_0 = const()[name = tensor("op_1967_strides_0"), val = tensor([1, 1])]; + tensor var_1967_pad_0 = const()[name = tensor("op_1967_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1967_dilations_0 = const()[name = tensor("op_1967_dilations_0"), val = tensor([1, 1])]; + tensor var_1967_groups_0 = const()[name = tensor("op_1967_groups_0"), val = tensor(1)]; + tensor layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28745856))), name = tensor("layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28737280))), shape = tensor([512, 512, 1, 1])]; + tensor var_1967_cast_fp16 = conv(dilations = var_1967_dilations_0, groups = var_1967_groups_0, pad = var_1967_pad_0, pad_type = var_1967_pad_type_0, strides = var_1967_strides_0, weight = layers_4_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_139_cast_fp16)[name = tensor("op_1967_cast_fp16")]; + tensor x_29_cast_fp16 = add(x = var_1961_cast_fp16, y = var_1967_cast_fp16)[name = tensor("x_29_cast_fp16")]; + tensor inputs_47_cast_fp16 = add(x = inputs_45_cast_fp16, y = x_29_cast_fp16)[name = tensor("inputs_47_cast_fp16")]; + tensor out_47_axes_0 = const()[name = tensor("out_47_axes_0"), val = tensor([1])]; + tensor var_1978_to_fp16 = const()[name = tensor("op_1978_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_47_cast_fp16 = layer_norm(axes = out_47_axes_0, epsilon = var_1978_to_fp16, x = inputs_47_cast_fp16)[name = tensor("out_47_cast_fp16")]; + tensor input_141_gamma_0_to_fp16 = const()[name = tensor("input_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28778688)))]; + tensor input_141_beta_0_to_fp16 = const()[name = tensor("input_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28779776)))]; + tensor input_141_epsilon_0_to_fp16 = const()[name = tensor("input_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_141_cast_fp16 = batch_norm(beta = input_141_beta_0_to_fp16, epsilon = input_141_epsilon_0_to_fp16, gamma = input_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_47_cast_fp16)[name = tensor("input_141_cast_fp16")]; + tensor var_1998_pad_type_0 = const()[name = tensor("op_1998_pad_type_0"), val = tensor("valid")]; + tensor var_1998_strides_0 = const()[name = tensor("op_1998_strides_0"), val = tensor([1, 1])]; + tensor var_1998_pad_0 = const()[name = tensor("op_1998_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_1998_dilations_0 = const()[name = tensor("op_1998_dilations_0"), val = tensor([1, 1])]; + tensor var_1998_groups_0 = const()[name = tensor("op_1998_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(28780864))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29567360))), name = tensor("layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_4_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29567552)))]; + tensor var_1998_cast_fp16 = conv(bias = layers_4_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_1998_dilations_0, groups = var_1998_groups_0, pad = var_1998_pad_0, pad_type = var_1998_pad_type_0, strides = var_1998_strides_0, weight = layers_4_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_141_cast_fp16)[name = tensor("op_1998_cast_fp16")]; + tensor var_2004_pad_type_0 = const()[name = tensor("op_2004_pad_type_0"), val = tensor("valid")]; + tensor var_2004_strides_0 = const()[name = tensor("op_2004_strides_0"), val = tensor([1, 1])]; + tensor var_2004_pad_0 = const()[name = tensor("op_2004_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2004_dilations_0 = const()[name = tensor("op_2004_dilations_0"), val = tensor([1, 1])]; + tensor var_2004_groups_0 = const()[name = tensor("op_2004_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29603136))), name = tensor("layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29571712))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2004_cast_fp16 = conv(dilations = var_2004_dilations_0, groups = var_2004_groups_0, pad = var_2004_pad_0, pad_type = var_2004_pad_type_0, strides = var_2004_strides_0, weight = layers_4_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_141_cast_fp16)[name = tensor("op_2004_cast_fp16")]; + tensor input_143_cast_fp16 = add(x = var_1998_cast_fp16, y = var_2004_cast_fp16)[name = tensor("input_143_cast_fp16")]; + tensor input_145_cast_fp16 = silu(x = input_143_cast_fp16)[name = tensor("input_145_cast_fp16")]; + tensor var_2015_pad_type_0 = const()[name = tensor("op_2015_pad_type_0"), val = tensor("valid")]; + tensor var_2015_strides_0 = const()[name = tensor("op_2015_strides_0"), val = tensor([1, 1])]; + tensor var_2015_pad_0 = const()[name = tensor("op_2015_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2015_dilations_0 = const()[name = tensor("op_2015_dilations_0"), val = tensor([1, 1])]; + tensor var_2015_groups_0 = const()[name = tensor("op_2015_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(29734272))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30520768))), name = tensor("layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_4_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_4_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30520960)))]; + tensor var_2015_cast_fp16 = conv(bias = layers_4_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_2015_dilations_0, groups = var_2015_groups_0, pad = var_2015_pad_0, pad_type = var_2015_pad_type_0, strides = var_2015_strides_0, weight = layers_4_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_145_cast_fp16)[name = tensor("op_2015_cast_fp16")]; + tensor var_2021_pad_type_0 = const()[name = tensor("op_2021_pad_type_0"), val = tensor("valid")]; + tensor var_2021_strides_0 = const()[name = tensor("op_2021_strides_0"), val = tensor([1, 1])]; + tensor var_2021_pad_0 = const()[name = tensor("op_2021_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2021_dilations_0 = const()[name = tensor("op_2021_dilations_0"), val = tensor([1, 1])]; + tensor var_2021_groups_0 = const()[name = tensor("op_2021_groups_0"), val = tensor(1)]; + tensor layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30556864))), name = tensor("layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30522048))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2021_cast_fp16 = conv(dilations = var_2021_dilations_0, groups = var_2021_groups_0, pad = var_2021_pad_0, pad_type = var_2021_pad_type_0, strides = var_2021_strides_0, weight = layers_4_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_145_cast_fp16)[name = tensor("op_2021_cast_fp16")]; + tensor x_31_cast_fp16 = add(x = var_2015_cast_fp16, y = var_2021_cast_fp16)[name = tensor("x_31_cast_fp16")]; + tensor var_2023_to_fp16 = const()[name = tensor("op_2023_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2024_cast_fp16 = mul(x = x_31_cast_fp16, y = var_2023_to_fp16)[name = tensor("op_2024_cast_fp16")]; + tensor inputs_49_cast_fp16 = add(x = inputs_47_cast_fp16, y = var_2024_cast_fp16)[name = tensor("inputs_49_cast_fp16")]; + tensor out_49_axes_0 = const()[name = tensor("out_49_axes_0"), val = tensor([1])]; + tensor var_2034_to_fp16 = const()[name = tensor("op_2034_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_49_cast_fp16 = layer_norm(axes = out_49_axes_0, epsilon = var_2034_to_fp16, x = inputs_49_cast_fp16)[name = tensor("out_49_cast_fp16")]; + tensor inputs_51_gamma_0_to_fp16 = const()[name = tensor("inputs_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30688000)))]; + tensor inputs_51_beta_0_to_fp16 = const()[name = tensor("inputs_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30689088)))]; + tensor inputs_51_epsilon_0_to_fp16 = const()[name = tensor("inputs_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_51_cast_fp16 = batch_norm(beta = inputs_51_beta_0_to_fp16, epsilon = inputs_51_epsilon_0_to_fp16, gamma = inputs_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_49_cast_fp16)[name = tensor("inputs_51_cast_fp16")]; + tensor var_2048 = const()[name = tensor("op_2048"), val = tensor(3)]; + tensor out_51_axes_0 = const()[name = tensor("out_51_axes_0"), val = tensor([1])]; + tensor var_2079_to_fp16 = const()[name = tensor("op_2079_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_51_cast_fp16 = layer_norm(axes = out_51_axes_0, epsilon = var_2079_to_fp16, x = inputs_51_cast_fp16)[name = tensor("out_51_cast_fp16")]; + tensor input_147_gamma_0_to_fp16 = const()[name = tensor("input_147_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30690176)))]; + tensor input_147_beta_0_to_fp16 = const()[name = tensor("input_147_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30691264)))]; + tensor input_147_epsilon_0_to_fp16 = const()[name = tensor("input_147_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_147_cast_fp16 = batch_norm(beta = input_147_beta_0_to_fp16, epsilon = input_147_epsilon_0_to_fp16, gamma = input_147_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_51_cast_fp16)[name = tensor("input_147_cast_fp16")]; + tensor var_2099_pad_type_0 = const()[name = tensor("op_2099_pad_type_0"), val = tensor("valid")]; + tensor var_2099_strides_0 = const()[name = tensor("op_2099_strides_0"), val = tensor([1, 1])]; + tensor var_2099_pad_0 = const()[name = tensor("op_2099_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2099_dilations_0 = const()[name = tensor("op_2099_dilations_0"), val = tensor([1, 1])]; + tensor var_2099_groups_0 = const()[name = tensor("op_2099_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(30692352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31478848))), name = tensor("layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_5_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31479040)))]; + tensor var_2099_cast_fp16 = conv(bias = layers_5_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2099_dilations_0, groups = var_2099_groups_0, pad = var_2099_pad_0, pad_type = var_2099_pad_type_0, strides = var_2099_strides_0, weight = layers_5_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_147_cast_fp16)[name = tensor("op_2099_cast_fp16")]; + tensor var_2105_pad_type_0 = const()[name = tensor("op_2105_pad_type_0"), val = tensor("valid")]; + tensor var_2105_strides_0 = const()[name = tensor("op_2105_strides_0"), val = tensor([1, 1])]; + tensor var_2105_pad_0 = const()[name = tensor("op_2105_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2105_dilations_0 = const()[name = tensor("op_2105_dilations_0"), val = tensor([1, 1])]; + tensor var_2105_groups_0 = const()[name = tensor("op_2105_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31515392))), name = tensor("layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31483200))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2105_cast_fp16 = conv(dilations = var_2105_dilations_0, groups = var_2105_groups_0, pad = var_2105_pad_0, pad_type = var_2105_pad_type_0, strides = var_2105_strides_0, weight = layers_5_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_147_cast_fp16)[name = tensor("op_2105_cast_fp16")]; + tensor input_149_cast_fp16 = add(x = var_2099_cast_fp16, y = var_2105_cast_fp16)[name = tensor("input_149_cast_fp16")]; + tensor input_151_cast_fp16 = silu(x = input_149_cast_fp16)[name = tensor("input_151_cast_fp16")]; + tensor var_2116_pad_type_0 = const()[name = tensor("op_2116_pad_type_0"), val = tensor("valid")]; + tensor var_2116_strides_0 = const()[name = tensor("op_2116_strides_0"), val = tensor([1, 1])]; + tensor var_2116_pad_0 = const()[name = tensor("op_2116_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2116_dilations_0 = const()[name = tensor("op_2116_dilations_0"), val = tensor([1, 1])]; + tensor var_2116_groups_0 = const()[name = tensor("op_2116_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(31646528))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32433024))), name = tensor("layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_5_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32433216)))]; + tensor var_2116_cast_fp16 = conv(bias = layers_5_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_2116_dilations_0, groups = var_2116_groups_0, pad = var_2116_pad_0, pad_type = var_2116_pad_type_0, strides = var_2116_strides_0, weight = layers_5_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_151_cast_fp16)[name = tensor("op_2116_cast_fp16")]; + tensor var_2122_pad_type_0 = const()[name = tensor("op_2122_pad_type_0"), val = tensor("valid")]; + tensor var_2122_strides_0 = const()[name = tensor("op_2122_strides_0"), val = tensor([1, 1])]; + tensor var_2122_pad_0 = const()[name = tensor("op_2122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2122_dilations_0 = const()[name = tensor("op_2122_dilations_0"), val = tensor([1, 1])]; + tensor var_2122_groups_0 = const()[name = tensor("op_2122_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32470784))), name = tensor("layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32434304))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2122_cast_fp16 = conv(dilations = var_2122_dilations_0, groups = var_2122_groups_0, pad = var_2122_pad_0, pad_type = var_2122_pad_type_0, strides = var_2122_strides_0, weight = layers_5_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_151_cast_fp16)[name = tensor("op_2122_cast_fp16")]; + tensor x_33_cast_fp16 = add(x = var_2116_cast_fp16, y = var_2122_cast_fp16)[name = tensor("x_33_cast_fp16")]; + tensor var_2124_to_fp16 = const()[name = tensor("op_2124_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2125_cast_fp16 = mul(x = x_33_cast_fp16, y = var_2124_to_fp16)[name = tensor("op_2125_cast_fp16")]; + tensor inputs_53_cast_fp16 = add(x = inputs_51_cast_fp16, y = var_2125_cast_fp16)[name = tensor("inputs_53_cast_fp16")]; + tensor out_53_axes_0 = const()[name = tensor("out_53_axes_0"), val = tensor([1])]; + tensor var_2135_to_fp16 = const()[name = tensor("op_2135_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_53_cast_fp16 = layer_norm(axes = out_53_axes_0, epsilon = var_2135_to_fp16, x = inputs_53_cast_fp16)[name = tensor("out_53_cast_fp16")]; + tensor obj_23_gamma_0_to_fp16 = const()[name = tensor("obj_23_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32601920)))]; + tensor obj_23_beta_0_to_fp16 = const()[name = tensor("obj_23_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32603008)))]; + tensor obj_23_epsilon_0_to_fp16 = const()[name = tensor("obj_23_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_23_cast_fp16 = batch_norm(beta = obj_23_beta_0_to_fp16, epsilon = obj_23_epsilon_0_to_fp16, gamma = obj_23_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_53_cast_fp16)[name = tensor("obj_23_cast_fp16")]; + tensor var_2160_pad_type_0 = const()[name = tensor("op_2160_pad_type_0"), val = tensor("valid")]; + tensor var_2160_strides_0 = const()[name = tensor("op_2160_strides_0"), val = tensor([1, 1])]; + tensor var_2160_pad_0 = const()[name = tensor("op_2160_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2160_dilations_0 = const()[name = tensor("op_2160_dilations_0"), val = tensor([1, 1])]; + tensor var_2160_groups_0 = const()[name = tensor("op_2160_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32604096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32800768))), name = tensor("layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_5_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32800960)))]; + tensor var_2160_cast_fp16 = conv(bias = layers_5_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_2160_dilations_0, groups = var_2160_groups_0, pad = var_2160_pad_0, pad_type = var_2160_pad_type_0, strides = var_2160_strides_0, weight = layers_5_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2160_cast_fp16")]; + tensor var_2166_pad_type_0 = const()[name = tensor("op_2166_pad_type_0"), val = tensor("valid")]; + tensor var_2166_strides_0 = const()[name = tensor("op_2166_strides_0"), val = tensor([1, 1])]; + tensor var_2166_pad_0 = const()[name = tensor("op_2166_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2166_dilations_0 = const()[name = tensor("op_2166_dilations_0"), val = tensor([1, 1])]; + tensor var_2166_groups_0 = const()[name = tensor("op_2166_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32810752))), name = tensor("layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32802048))), shape = tensor([512, 512, 1, 1])]; + tensor var_2166_cast_fp16 = conv(dilations = var_2166_dilations_0, groups = var_2166_groups_0, pad = var_2166_pad_0, pad_type = var_2166_pad_type_0, strides = var_2166_strides_0, weight = layers_5_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2166_cast_fp16")]; + tensor query_21_cast_fp16 = add(x = var_2160_cast_fp16, y = var_2166_cast_fp16)[name = tensor("query_21_cast_fp16")]; + tensor var_2175_pad_type_0 = const()[name = tensor("op_2175_pad_type_0"), val = tensor("valid")]; + tensor var_2175_strides_0 = const()[name = tensor("op_2175_strides_0"), val = tensor([1, 1])]; + tensor var_2175_pad_0 = const()[name = tensor("op_2175_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2175_dilations_0 = const()[name = tensor("op_2175_dilations_0"), val = tensor([1, 1])]; + tensor var_2175_groups_0 = const()[name = tensor("op_2175_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(32843584))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33040256))), name = tensor("layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2175_cast_fp16 = conv(dilations = var_2175_dilations_0, groups = var_2175_groups_0, pad = var_2175_pad_0, pad_type = var_2175_pad_type_0, strides = var_2175_strides_0, weight = layers_5_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2175_cast_fp16")]; + tensor var_2181_pad_type_0 = const()[name = tensor("op_2181_pad_type_0"), val = tensor("valid")]; + tensor var_2181_strides_0 = const()[name = tensor("op_2181_strides_0"), val = tensor([1, 1])]; + tensor var_2181_pad_0 = const()[name = tensor("op_2181_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2181_dilations_0 = const()[name = tensor("op_2181_dilations_0"), val = tensor([1, 1])]; + tensor var_2181_groups_0 = const()[name = tensor("op_2181_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33051712))), name = tensor("layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33040448))), shape = tensor([512, 512, 1, 1])]; + tensor var_2181_cast_fp16 = conv(dilations = var_2181_dilations_0, groups = var_2181_groups_0, pad = var_2181_pad_0, pad_type = var_2181_pad_type_0, strides = var_2181_strides_0, weight = layers_5_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2181_cast_fp16")]; + tensor key_11_cast_fp16 = add(x = var_2175_cast_fp16, y = var_2181_cast_fp16)[name = tensor("key_11_cast_fp16")]; + tensor var_2191_pad_type_0 = const()[name = tensor("op_2191_pad_type_0"), val = tensor("valid")]; + tensor var_2191_strides_0 = const()[name = tensor("op_2191_strides_0"), val = tensor([1, 1])]; + tensor var_2191_pad_0 = const()[name = tensor("op_2191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2191_dilations_0 = const()[name = tensor("op_2191_dilations_0"), val = tensor([1, 1])]; + tensor var_2191_groups_0 = const()[name = tensor("op_2191_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33084544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33281216))), name = tensor("layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_5_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33281408)))]; + tensor var_2191_cast_fp16 = conv(bias = layers_5_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_2191_dilations_0, groups = var_2191_groups_0, pad = var_2191_pad_0, pad_type = var_2191_pad_type_0, strides = var_2191_strides_0, weight = layers_5_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_23_cast_fp16)[name = tensor("op_2191_cast_fp16")]; + tensor var_2197_pad_type_0 = const()[name = tensor("op_2197_pad_type_0"), val = tensor("valid")]; + tensor var_2197_strides_0 = const()[name = tensor("op_2197_strides_0"), val = tensor([1, 1])]; + tensor var_2197_pad_0 = const()[name = tensor("op_2197_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2197_dilations_0 = const()[name = tensor("op_2197_dilations_0"), val = tensor([1, 1])]; + tensor var_2197_groups_0 = const()[name = tensor("op_2197_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33290816))), name = tensor("layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33282496))), shape = tensor([512, 512, 1, 1])]; + tensor var_2197_cast_fp16 = conv(dilations = var_2197_dilations_0, groups = var_2197_groups_0, pad = var_2197_pad_0, pad_type = var_2197_pad_type_0, strides = var_2197_strides_0, weight = layers_5_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_23_cast_fp16)[name = tensor("op_2197_cast_fp16")]; + tensor value_11_cast_fp16 = add(x = var_2191_cast_fp16, y = var_2197_cast_fp16)[name = tensor("value_11_cast_fp16")]; + tensor var_2200_to_fp16 = const()[name = tensor("op_2200_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33323648)))]; + tensor query_23_cast_fp16 = add(x = query_21_cast_fp16, y = var_2200_to_fp16)[name = tensor("query_23_cast_fp16")]; + tensor var_2203_to_fp16 = const()[name = tensor("op_2203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33324736)))]; + tensor q_with_bias_v_11_cast_fp16 = add(x = query_21_cast_fp16, y = var_2203_to_fp16)[name = tensor("q_with_bias_v_11_cast_fp16")]; + tensor var_2213_pad_type_0 = const()[name = tensor("op_2213_pad_type_0"), val = tensor("valid")]; + tensor var_2213_strides_0 = const()[name = tensor("op_2213_strides_0"), val = tensor([1, 1])]; + tensor var_2213_pad_0 = const()[name = tensor("op_2213_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2213_dilations_0 = const()[name = tensor("op_2213_dilations_0"), val = tensor([1, 1])]; + tensor var_2213_groups_0 = const()[name = tensor("op_2213_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33325824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33522496))), name = tensor("layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2213_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2213_dilations_0, groups = var_2213_groups_0, pad = var_2213_pad_0, pad_type = var_2213_pad_type_0, strides = var_2213_strides_0, weight = layers_5_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2213_cast_fp16")]; + tensor var_2219_pad_type_0 = const()[name = tensor("op_2219_pad_type_0"), val = tensor("valid")]; + tensor var_2219_strides_0 = const()[name = tensor("op_2219_strides_0"), val = tensor([1, 1])]; + tensor var_2219_pad_0 = const()[name = tensor("op_2219_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2219_dilations_0 = const()[name = tensor("op_2219_dilations_0"), val = tensor([1, 1])]; + tensor var_2219_groups_0 = const()[name = tensor("op_2219_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33541568))), name = tensor("layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33522688))), shape = tensor([512, 512, 1, 1])]; + tensor var_2219_cast_fp16 = conv(dilations = var_2219_dilations_0, groups = var_2219_groups_0, pad = var_2219_pad_0, pad_type = var_2219_pad_type_0, strides = var_2219_strides_0, weight = layers_5_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2219_cast_fp16")]; + tensor p_11_cast_fp16 = add(x = var_2213_cast_fp16, y = var_2219_cast_fp16)[name = tensor("p_11_cast_fp16")]; + tensor var_2223 = const()[name = tensor("op_2223"), val = tensor([1, 8, 64, 188])]; + tensor var_2224_cast_fp16 = reshape(shape = var_2223, x = q_with_bias_v_11_cast_fp16)[name = tensor("op_2224_cast_fp16")]; + tensor var_2225 = const()[name = tensor("op_2225"), val = tensor([1, 8, 64, -1])]; + tensor var_2226_cast_fp16 = reshape(shape = var_2225, x = p_11_cast_fp16)[name = tensor("op_2226_cast_fp16")]; + tensor matrix_bd_41_transpose_x_0 = const()[name = tensor("matrix_bd_41_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_41_transpose_y_0 = const()[name = tensor("matrix_bd_41_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_41_cast_fp16 = matmul(transpose_x = matrix_bd_41_transpose_x_0, transpose_y = matrix_bd_41_transpose_y_0, x = var_2224_cast_fp16, y = var_2226_cast_fp16)[name = tensor("matrix_bd_41_cast_fp16")]; + tensor matrix_bd_43_pad_0 = const()[name = tensor("matrix_bd_43_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_43_mode_0 = const()[name = tensor("matrix_bd_43_mode_0"), val = tensor("constant")]; + tensor const_65_to_fp16 = const()[name = tensor("const_65_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_43_cast_fp16 = pad(constant_val = const_65_to_fp16, mode = matrix_bd_43_mode_0, pad = matrix_bd_43_pad_0, x = matrix_bd_41_cast_fp16)[name = tensor("matrix_bd_43_cast_fp16")]; + tensor var_2235 = const()[name = tensor("op_2235"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_45_cast_fp16 = reshape(shape = var_2235, x = matrix_bd_43_cast_fp16)[name = tensor("matrix_bd_45_cast_fp16")]; + tensor var_2239_begin_0 = const()[name = tensor("op_2239_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2239_end_0 = const()[name = tensor("op_2239_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_2239_end_mask_0 = const()[name = tensor("op_2239_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2239_cast_fp16 = slice_by_index(begin = var_2239_begin_0, end = var_2239_end_0, end_mask = var_2239_end_mask_0, x = matrix_bd_45_cast_fp16)[name = tensor("op_2239_cast_fp16")]; + tensor var_2240 = const()[name = tensor("op_2240"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_47_cast_fp16 = reshape(shape = var_2240, x = var_2239_cast_fp16)[name = tensor("matrix_bd_47_cast_fp16")]; + tensor var_2245_begin_0 = const()[name = tensor("op_2245_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2245_end_0 = const()[name = tensor("op_2245_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_2245_end_mask_0 = const()[name = tensor("op_2245_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2245_cast_fp16 = slice_by_index(begin = var_2245_begin_0, end = var_2245_end_0, end_mask = var_2245_end_mask_0, x = matrix_bd_47_cast_fp16)[name = tensor("op_2245_cast_fp16")]; + tensor var_2246_to_fp16 = const()[name = tensor("op_2246_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_11_cast_fp16 = mul(x = var_2245_cast_fp16, y = var_2246_to_fp16)[name = tensor("qk_mask_11_cast_fp16")]; + tensor var_2250 = const()[name = tensor("op_2250"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_11_cast_fp16 = reshape(shape = var_2250, x = query_23_cast_fp16)[name = tensor("mh_q_11_cast_fp16")]; + tensor var_2252_to_fp16 = const()[name = tensor("op_2252_to_fp16"), val = tensor(0x1p-3)]; + tensor var_2253_cast_fp16 = mul(x = mh_q_11_cast_fp16, y = var_2252_to_fp16)[name = tensor("op_2253_cast_fp16")]; + tensor var_2256 = const()[name = tensor("op_2256"), val = tensor([1, 8, 64, 188])]; + tensor var_2257_cast_fp16 = reshape(shape = var_2256, x = key_11_cast_fp16)[name = tensor("op_2257_cast_fp16")]; + tensor mh_w_21_transpose_x_0 = const()[name = tensor("mh_w_21_transpose_x_0"), val = tensor(true)]; + tensor mh_w_21_transpose_y_0 = const()[name = tensor("mh_w_21_transpose_y_0"), val = tensor(false)]; + tensor mh_w_21_cast_fp16 = matmul(transpose_x = mh_w_21_transpose_x_0, transpose_y = mh_w_21_transpose_y_0, x = var_2253_cast_fp16, y = var_2257_cast_fp16)[name = tensor("mh_w_21_cast_fp16")]; + tensor mh_w_23_cast_fp16 = add(x = mh_w_21_cast_fp16, y = qk_mask_11_cast_fp16)[name = tensor("mh_w_23_cast_fp16")]; + tensor var_2261_cast_fp16 = softmax(axis = var_2048, x = mh_w_23_cast_fp16)[name = tensor("op_2261_cast_fp16")]; + tensor var_2262 = const()[name = tensor("op_2262"), val = tensor([1, 8, 64, 188])]; + tensor var_2263_cast_fp16 = reshape(shape = var_2262, x = value_11_cast_fp16)[name = tensor("op_2263_cast_fp16")]; + tensor attn_11_transpose_x_0 = const()[name = tensor("attn_11_transpose_x_0"), val = tensor(false)]; + tensor attn_11_transpose_y_0 = const()[name = tensor("attn_11_transpose_y_0"), val = tensor(true)]; + tensor attn_11_cast_fp16 = matmul(transpose_x = attn_11_transpose_x_0, transpose_y = attn_11_transpose_y_0, x = var_2263_cast_fp16, y = var_2261_cast_fp16)[name = tensor("attn_11_cast_fp16")]; + tensor var_2266 = const()[name = tensor("op_2266"), val = tensor([1, 512, 1, 188])]; + tensor input_153_cast_fp16 = reshape(shape = var_2266, x = attn_11_cast_fp16)[name = tensor("input_153_cast_fp16")]; + tensor var_2276_pad_type_0 = const()[name = tensor("op_2276_pad_type_0"), val = tensor("valid")]; + tensor var_2276_strides_0 = const()[name = tensor("op_2276_strides_0"), val = tensor([1, 1])]; + tensor var_2276_pad_0 = const()[name = tensor("op_2276_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2276_dilations_0 = const()[name = tensor("op_2276_dilations_0"), val = tensor([1, 1])]; + tensor var_2276_groups_0 = const()[name = tensor("op_2276_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33574400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33771072))), name = tensor("layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_5_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33771264)))]; + tensor var_2276_cast_fp16 = conv(bias = layers_5_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_2276_dilations_0, groups = var_2276_groups_0, pad = var_2276_pad_0, pad_type = var_2276_pad_type_0, strides = var_2276_strides_0, weight = layers_5_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_153_cast_fp16)[name = tensor("op_2276_cast_fp16")]; + tensor var_2282_pad_type_0 = const()[name = tensor("op_2282_pad_type_0"), val = tensor("valid")]; + tensor var_2282_strides_0 = const()[name = tensor("op_2282_strides_0"), val = tensor([1, 1])]; + tensor var_2282_pad_0 = const()[name = tensor("op_2282_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2282_dilations_0 = const()[name = tensor("op_2282_dilations_0"), val = tensor([1, 1])]; + tensor var_2282_groups_0 = const()[name = tensor("op_2282_groups_0"), val = tensor(1)]; + tensor layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33780736))), name = tensor("layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33772352))), shape = tensor([512, 512, 1, 1])]; + tensor var_2282_cast_fp16 = conv(dilations = var_2282_dilations_0, groups = var_2282_groups_0, pad = var_2282_pad_0, pad_type = var_2282_pad_type_0, strides = var_2282_strides_0, weight = layers_5_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_153_cast_fp16)[name = tensor("op_2282_cast_fp16")]; + tensor obj_25_cast_fp16 = add(x = var_2276_cast_fp16, y = var_2282_cast_fp16)[name = tensor("obj_25_cast_fp16")]; + tensor inputs_55_cast_fp16 = add(x = inputs_53_cast_fp16, y = obj_25_cast_fp16)[name = tensor("inputs_55_cast_fp16")]; + tensor out_55_axes_0 = const()[name = tensor("out_55_axes_0"), val = tensor([1])]; + tensor var_2293_to_fp16 = const()[name = tensor("op_2293_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_55_cast_fp16 = layer_norm(axes = out_55_axes_0, epsilon = var_2293_to_fp16, x = inputs_55_cast_fp16)[name = tensor("out_55_cast_fp16")]; + tensor input_155_gamma_0_to_fp16 = const()[name = tensor("input_155_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33813568)))]; + tensor input_155_beta_0_to_fp16 = const()[name = tensor("input_155_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33814656)))]; + tensor input_155_epsilon_0_to_fp16 = const()[name = tensor("input_155_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_155_cast_fp16 = batch_norm(beta = input_155_beta_0_to_fp16, epsilon = input_155_epsilon_0_to_fp16, gamma = input_155_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_55_cast_fp16)[name = tensor("input_155_cast_fp16")]; + tensor var_2315_pad_type_0 = const()[name = tensor("op_2315_pad_type_0"), val = tensor("valid")]; + tensor var_2315_strides_0 = const()[name = tensor("op_2315_strides_0"), val = tensor([1, 1])]; + tensor var_2315_pad_0 = const()[name = tensor("op_2315_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2315_dilations_0 = const()[name = tensor("op_2315_dilations_0"), val = tensor([1, 1])]; + tensor var_2315_groups_0 = const()[name = tensor("op_2315_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(33815744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34209024))), name = tensor("layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_5_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34209216)))]; + tensor var_2315_cast_fp16 = conv(bias = layers_5_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_2315_dilations_0, groups = var_2315_groups_0, pad = var_2315_pad_0, pad_type = var_2315_pad_type_0, strides = var_2315_strides_0, weight = layers_5_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_155_cast_fp16)[name = tensor("op_2315_cast_fp16")]; + tensor var_2321_pad_type_0 = const()[name = tensor("op_2321_pad_type_0"), val = tensor("valid")]; + tensor var_2321_strides_0 = const()[name = tensor("op_2321_strides_0"), val = tensor([1, 1])]; + tensor var_2321_pad_0 = const()[name = tensor("op_2321_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2321_dilations_0 = const()[name = tensor("op_2321_dilations_0"), val = tensor([1, 1])]; + tensor var_2321_groups_0 = const()[name = tensor("op_2321_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34228480))), name = tensor("layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34211328))), shape = tensor([1024, 512, 1, 1])]; + tensor var_2321_cast_fp16 = conv(dilations = var_2321_dilations_0, groups = var_2321_groups_0, pad = var_2321_pad_0, pad_type = var_2321_pad_type_0, strides = var_2321_strides_0, weight = layers_5_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_155_cast_fp16)[name = tensor("op_2321_cast_fp16")]; + tensor input_157_cast_fp16 = add(x = var_2315_cast_fp16, y = var_2321_cast_fp16)[name = tensor("input_157_cast_fp16")]; + tensor input_159_split_num_splits_0 = const()[name = tensor("input_159_split_num_splits_0"), val = tensor(2)]; + tensor input_159_split_axis_0 = const()[name = tensor("input_159_split_axis_0"), val = tensor(1)]; + tensor input_159_split_cast_fp16_0, tensor input_159_split_cast_fp16_1 = split(axis = input_159_split_axis_0, num_splits = input_159_split_num_splits_0, x = input_157_cast_fp16)[name = tensor("input_159_split_cast_fp16")]; + tensor input_159_split_1_sigmoid_cast_fp16 = sigmoid(x = input_159_split_cast_fp16_1)[name = tensor("input_159_split_1_sigmoid_cast_fp16")]; + tensor input_159_cast_fp16 = mul(x = input_159_split_cast_fp16_0, y = input_159_split_1_sigmoid_cast_fp16)[name = tensor("input_159_cast_fp16")]; + tensor input_161_pad_type_0 = const()[name = tensor("input_161_pad_type_0"), val = tensor("custom")]; + tensor input_161_pad_0 = const()[name = tensor("input_161_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_161_groups_0 = const()[name = tensor("input_161_groups_0"), val = tensor(512)]; + tensor input_161_strides_0 = const()[name = tensor("input_161_strides_0"), val = tensor([1, 1])]; + tensor input_161_dilations_0 = const()[name = tensor("input_161_dilations_0"), val = tensor([1, 1])]; + tensor const_201_to_fp16 = const()[name = tensor("const_201_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34294080)))]; + tensor const_202_to_fp16 = const()[name = tensor("const_202_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34303360)))]; + tensor input_163_cast_fp16 = conv(bias = const_202_to_fp16, dilations = input_161_dilations_0, groups = input_161_groups_0, pad = input_161_pad_0, pad_type = input_161_pad_type_0, strides = input_161_strides_0, weight = const_201_to_fp16, x = input_159_cast_fp16)[name = tensor("input_163_cast_fp16")]; + tensor input_165_cast_fp16 = silu(x = input_163_cast_fp16)[name = tensor("input_165_cast_fp16")]; + tensor var_2345_pad_type_0 = const()[name = tensor("op_2345_pad_type_0"), val = tensor("valid")]; + tensor var_2345_strides_0 = const()[name = tensor("op_2345_strides_0"), val = tensor([1, 1])]; + tensor var_2345_pad_0 = const()[name = tensor("op_2345_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2345_dilations_0 = const()[name = tensor("op_2345_dilations_0"), val = tensor([1, 1])]; + tensor var_2345_groups_0 = const()[name = tensor("op_2345_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34304448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34501120))), name = tensor("layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_5_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34501312)))]; + tensor var_2345_cast_fp16 = conv(bias = layers_5_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_2345_dilations_0, groups = var_2345_groups_0, pad = var_2345_pad_0, pad_type = var_2345_pad_type_0, strides = var_2345_strides_0, weight = layers_5_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_165_cast_fp16)[name = tensor("op_2345_cast_fp16")]; + tensor var_2351_pad_type_0 = const()[name = tensor("op_2351_pad_type_0"), val = tensor("valid")]; + tensor var_2351_strides_0 = const()[name = tensor("op_2351_strides_0"), val = tensor([1, 1])]; + tensor var_2351_pad_0 = const()[name = tensor("op_2351_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2351_dilations_0 = const()[name = tensor("op_2351_dilations_0"), val = tensor([1, 1])]; + tensor var_2351_groups_0 = const()[name = tensor("op_2351_groups_0"), val = tensor(1)]; + tensor layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34510656))), name = tensor("layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34502400))), shape = tensor([512, 512, 1, 1])]; + tensor var_2351_cast_fp16 = conv(dilations = var_2351_dilations_0, groups = var_2351_groups_0, pad = var_2351_pad_0, pad_type = var_2351_pad_type_0, strides = var_2351_strides_0, weight = layers_5_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_165_cast_fp16)[name = tensor("op_2351_cast_fp16")]; + tensor x_35_cast_fp16 = add(x = var_2345_cast_fp16, y = var_2351_cast_fp16)[name = tensor("x_35_cast_fp16")]; + tensor inputs_57_cast_fp16 = add(x = inputs_55_cast_fp16, y = x_35_cast_fp16)[name = tensor("inputs_57_cast_fp16")]; + tensor out_57_axes_0 = const()[name = tensor("out_57_axes_0"), val = tensor([1])]; + tensor var_2362_to_fp16 = const()[name = tensor("op_2362_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_57_cast_fp16 = layer_norm(axes = out_57_axes_0, epsilon = var_2362_to_fp16, x = inputs_57_cast_fp16)[name = tensor("out_57_cast_fp16")]; + tensor input_167_gamma_0_to_fp16 = const()[name = tensor("input_167_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34543488)))]; + tensor input_167_beta_0_to_fp16 = const()[name = tensor("input_167_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34544576)))]; + tensor input_167_epsilon_0_to_fp16 = const()[name = tensor("input_167_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_167_cast_fp16 = batch_norm(beta = input_167_beta_0_to_fp16, epsilon = input_167_epsilon_0_to_fp16, gamma = input_167_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_57_cast_fp16)[name = tensor("input_167_cast_fp16")]; + tensor var_2382_pad_type_0 = const()[name = tensor("op_2382_pad_type_0"), val = tensor("valid")]; + tensor var_2382_strides_0 = const()[name = tensor("op_2382_strides_0"), val = tensor([1, 1])]; + tensor var_2382_pad_0 = const()[name = tensor("op_2382_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2382_dilations_0 = const()[name = tensor("op_2382_dilations_0"), val = tensor([1, 1])]; + tensor var_2382_groups_0 = const()[name = tensor("op_2382_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(34545664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35332160))), name = tensor("layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_5_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35332352)))]; + tensor var_2382_cast_fp16 = conv(bias = layers_5_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_2382_dilations_0, groups = var_2382_groups_0, pad = var_2382_pad_0, pad_type = var_2382_pad_type_0, strides = var_2382_strides_0, weight = layers_5_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_167_cast_fp16)[name = tensor("op_2382_cast_fp16")]; + tensor var_2388_pad_type_0 = const()[name = tensor("op_2388_pad_type_0"), val = tensor("valid")]; + tensor var_2388_strides_0 = const()[name = tensor("op_2388_strides_0"), val = tensor([1, 1])]; + tensor var_2388_pad_0 = const()[name = tensor("op_2388_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2388_dilations_0 = const()[name = tensor("op_2388_dilations_0"), val = tensor([1, 1])]; + tensor var_2388_groups_0 = const()[name = tensor("op_2388_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35369792))), name = tensor("layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35336512))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2388_cast_fp16 = conv(dilations = var_2388_dilations_0, groups = var_2388_groups_0, pad = var_2388_pad_0, pad_type = var_2388_pad_type_0, strides = var_2388_strides_0, weight = layers_5_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_167_cast_fp16)[name = tensor("op_2388_cast_fp16")]; + tensor input_169_cast_fp16 = add(x = var_2382_cast_fp16, y = var_2388_cast_fp16)[name = tensor("input_169_cast_fp16")]; + tensor input_171_cast_fp16 = silu(x = input_169_cast_fp16)[name = tensor("input_171_cast_fp16")]; + tensor var_2399_pad_type_0 = const()[name = tensor("op_2399_pad_type_0"), val = tensor("valid")]; + tensor var_2399_strides_0 = const()[name = tensor("op_2399_strides_0"), val = tensor([1, 1])]; + tensor var_2399_pad_0 = const()[name = tensor("op_2399_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2399_dilations_0 = const()[name = tensor("op_2399_dilations_0"), val = tensor([1, 1])]; + tensor var_2399_groups_0 = const()[name = tensor("op_2399_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(35500928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36287424))), name = tensor("layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_5_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_5_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36287616)))]; + tensor var_2399_cast_fp16 = conv(bias = layers_5_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_2399_dilations_0, groups = var_2399_groups_0, pad = var_2399_pad_0, pad_type = var_2399_pad_type_0, strides = var_2399_strides_0, weight = layers_5_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_171_cast_fp16)[name = tensor("op_2399_cast_fp16")]; + tensor var_2405_pad_type_0 = const()[name = tensor("op_2405_pad_type_0"), val = tensor("valid")]; + tensor var_2405_strides_0 = const()[name = tensor("op_2405_strides_0"), val = tensor([1, 1])]; + tensor var_2405_pad_0 = const()[name = tensor("op_2405_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2405_dilations_0 = const()[name = tensor("op_2405_dilations_0"), val = tensor([1, 1])]; + tensor var_2405_groups_0 = const()[name = tensor("op_2405_groups_0"), val = tensor(1)]; + tensor layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36328256))), name = tensor("layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36288704))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2405_cast_fp16 = conv(dilations = var_2405_dilations_0, groups = var_2405_groups_0, pad = var_2405_pad_0, pad_type = var_2405_pad_type_0, strides = var_2405_strides_0, weight = layers_5_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_171_cast_fp16)[name = tensor("op_2405_cast_fp16")]; + tensor x_37_cast_fp16 = add(x = var_2399_cast_fp16, y = var_2405_cast_fp16)[name = tensor("x_37_cast_fp16")]; + tensor var_2407_to_fp16 = const()[name = tensor("op_2407_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2408_cast_fp16 = mul(x = x_37_cast_fp16, y = var_2407_to_fp16)[name = tensor("op_2408_cast_fp16")]; + tensor inputs_59_cast_fp16 = add(x = inputs_57_cast_fp16, y = var_2408_cast_fp16)[name = tensor("inputs_59_cast_fp16")]; + tensor out_59_axes_0 = const()[name = tensor("out_59_axes_0"), val = tensor([1])]; + tensor var_2418_to_fp16 = const()[name = tensor("op_2418_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_59_cast_fp16 = layer_norm(axes = out_59_axes_0, epsilon = var_2418_to_fp16, x = inputs_59_cast_fp16)[name = tensor("out_59_cast_fp16")]; + tensor inputs_61_gamma_0_to_fp16 = const()[name = tensor("inputs_61_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36459392)))]; + tensor inputs_61_beta_0_to_fp16 = const()[name = tensor("inputs_61_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36460480)))]; + tensor inputs_61_epsilon_0_to_fp16 = const()[name = tensor("inputs_61_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_61_cast_fp16 = batch_norm(beta = inputs_61_beta_0_to_fp16, epsilon = inputs_61_epsilon_0_to_fp16, gamma = inputs_61_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_59_cast_fp16)[name = tensor("inputs_61_cast_fp16")]; + tensor var_2432 = const()[name = tensor("op_2432"), val = tensor(3)]; + tensor out_61_axes_0 = const()[name = tensor("out_61_axes_0"), val = tensor([1])]; + tensor var_2463_to_fp16 = const()[name = tensor("op_2463_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_61_cast_fp16 = layer_norm(axes = out_61_axes_0, epsilon = var_2463_to_fp16, x = inputs_61_cast_fp16)[name = tensor("out_61_cast_fp16")]; + tensor input_173_gamma_0_to_fp16 = const()[name = tensor("input_173_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36461568)))]; + tensor input_173_beta_0_to_fp16 = const()[name = tensor("input_173_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36462656)))]; + tensor input_173_epsilon_0_to_fp16 = const()[name = tensor("input_173_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_173_cast_fp16 = batch_norm(beta = input_173_beta_0_to_fp16, epsilon = input_173_epsilon_0_to_fp16, gamma = input_173_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_61_cast_fp16)[name = tensor("input_173_cast_fp16")]; + tensor var_2483_pad_type_0 = const()[name = tensor("op_2483_pad_type_0"), val = tensor("valid")]; + tensor var_2483_strides_0 = const()[name = tensor("op_2483_strides_0"), val = tensor([1, 1])]; + tensor var_2483_pad_0 = const()[name = tensor("op_2483_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2483_dilations_0 = const()[name = tensor("op_2483_dilations_0"), val = tensor([1, 1])]; + tensor var_2483_groups_0 = const()[name = tensor("op_2483_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(36463744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37250240))), name = tensor("layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_6_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37250432)))]; + tensor var_2483_cast_fp16 = conv(bias = layers_6_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2483_dilations_0, groups = var_2483_groups_0, pad = var_2483_pad_0, pad_type = var_2483_pad_type_0, strides = var_2483_strides_0, weight = layers_6_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_173_cast_fp16)[name = tensor("op_2483_cast_fp16")]; + tensor var_2489_pad_type_0 = const()[name = tensor("op_2489_pad_type_0"), val = tensor("valid")]; + tensor var_2489_strides_0 = const()[name = tensor("op_2489_strides_0"), val = tensor([1, 1])]; + tensor var_2489_pad_0 = const()[name = tensor("op_2489_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2489_dilations_0 = const()[name = tensor("op_2489_dilations_0"), val = tensor([1, 1])]; + tensor var_2489_groups_0 = const()[name = tensor("op_2489_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37288576))), name = tensor("layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37254592))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2489_cast_fp16 = conv(dilations = var_2489_dilations_0, groups = var_2489_groups_0, pad = var_2489_pad_0, pad_type = var_2489_pad_type_0, strides = var_2489_strides_0, weight = layers_6_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_173_cast_fp16)[name = tensor("op_2489_cast_fp16")]; + tensor input_175_cast_fp16 = add(x = var_2483_cast_fp16, y = var_2489_cast_fp16)[name = tensor("input_175_cast_fp16")]; + tensor input_177_cast_fp16 = silu(x = input_175_cast_fp16)[name = tensor("input_177_cast_fp16")]; + tensor var_2500_pad_type_0 = const()[name = tensor("op_2500_pad_type_0"), val = tensor("valid")]; + tensor var_2500_strides_0 = const()[name = tensor("op_2500_strides_0"), val = tensor([1, 1])]; + tensor var_2500_pad_0 = const()[name = tensor("op_2500_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2500_dilations_0 = const()[name = tensor("op_2500_dilations_0"), val = tensor([1, 1])]; + tensor var_2500_groups_0 = const()[name = tensor("op_2500_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(37419712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38206208))), name = tensor("layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_6_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38206400)))]; + tensor var_2500_cast_fp16 = conv(bias = layers_6_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_2500_dilations_0, groups = var_2500_groups_0, pad = var_2500_pad_0, pad_type = var_2500_pad_type_0, strides = var_2500_strides_0, weight = layers_6_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_177_cast_fp16)[name = tensor("op_2500_cast_fp16")]; + tensor var_2506_pad_type_0 = const()[name = tensor("op_2506_pad_type_0"), val = tensor("valid")]; + tensor var_2506_strides_0 = const()[name = tensor("op_2506_strides_0"), val = tensor([1, 1])]; + tensor var_2506_pad_0 = const()[name = tensor("op_2506_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2506_dilations_0 = const()[name = tensor("op_2506_dilations_0"), val = tensor([1, 1])]; + tensor var_2506_groups_0 = const()[name = tensor("op_2506_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38248448))), name = tensor("layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38207488))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2506_cast_fp16 = conv(dilations = var_2506_dilations_0, groups = var_2506_groups_0, pad = var_2506_pad_0, pad_type = var_2506_pad_type_0, strides = var_2506_strides_0, weight = layers_6_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_177_cast_fp16)[name = tensor("op_2506_cast_fp16")]; + tensor x_39_cast_fp16 = add(x = var_2500_cast_fp16, y = var_2506_cast_fp16)[name = tensor("x_39_cast_fp16")]; + tensor var_2508_to_fp16 = const()[name = tensor("op_2508_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2509_cast_fp16 = mul(x = x_39_cast_fp16, y = var_2508_to_fp16)[name = tensor("op_2509_cast_fp16")]; + tensor inputs_63_cast_fp16 = add(x = inputs_61_cast_fp16, y = var_2509_cast_fp16)[name = tensor("inputs_63_cast_fp16")]; + tensor out_63_axes_0 = const()[name = tensor("out_63_axes_0"), val = tensor([1])]; + tensor var_2519_to_fp16 = const()[name = tensor("op_2519_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_63_cast_fp16 = layer_norm(axes = out_63_axes_0, epsilon = var_2519_to_fp16, x = inputs_63_cast_fp16)[name = tensor("out_63_cast_fp16")]; + tensor obj_27_gamma_0_to_fp16 = const()[name = tensor("obj_27_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38379584)))]; + tensor obj_27_beta_0_to_fp16 = const()[name = tensor("obj_27_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38380672)))]; + tensor obj_27_epsilon_0_to_fp16 = const()[name = tensor("obj_27_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_27_cast_fp16 = batch_norm(beta = obj_27_beta_0_to_fp16, epsilon = obj_27_epsilon_0_to_fp16, gamma = obj_27_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_63_cast_fp16)[name = tensor("obj_27_cast_fp16")]; + tensor var_2544_pad_type_0 = const()[name = tensor("op_2544_pad_type_0"), val = tensor("valid")]; + tensor var_2544_strides_0 = const()[name = tensor("op_2544_strides_0"), val = tensor([1, 1])]; + tensor var_2544_pad_0 = const()[name = tensor("op_2544_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2544_dilations_0 = const()[name = tensor("op_2544_dilations_0"), val = tensor([1, 1])]; + tensor var_2544_groups_0 = const()[name = tensor("op_2544_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38381760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38578432))), name = tensor("layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_6_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38578624)))]; + tensor var_2544_cast_fp16 = conv(bias = layers_6_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_2544_dilations_0, groups = var_2544_groups_0, pad = var_2544_pad_0, pad_type = var_2544_pad_type_0, strides = var_2544_strides_0, weight = layers_6_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2544_cast_fp16")]; + tensor var_2550_pad_type_0 = const()[name = tensor("op_2550_pad_type_0"), val = tensor("valid")]; + tensor var_2550_strides_0 = const()[name = tensor("op_2550_strides_0"), val = tensor([1, 1])]; + tensor var_2550_pad_0 = const()[name = tensor("op_2550_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2550_dilations_0 = const()[name = tensor("op_2550_dilations_0"), val = tensor([1, 1])]; + tensor var_2550_groups_0 = const()[name = tensor("op_2550_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38588992))), name = tensor("layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38579712))), shape = tensor([512, 512, 1, 1])]; + tensor var_2550_cast_fp16 = conv(dilations = var_2550_dilations_0, groups = var_2550_groups_0, pad = var_2550_pad_0, pad_type = var_2550_pad_type_0, strides = var_2550_strides_0, weight = layers_6_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2550_cast_fp16")]; + tensor query_25_cast_fp16 = add(x = var_2544_cast_fp16, y = var_2550_cast_fp16)[name = tensor("query_25_cast_fp16")]; + tensor var_2559_pad_type_0 = const()[name = tensor("op_2559_pad_type_0"), val = tensor("valid")]; + tensor var_2559_strides_0 = const()[name = tensor("op_2559_strides_0"), val = tensor([1, 1])]; + tensor var_2559_pad_0 = const()[name = tensor("op_2559_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2559_dilations_0 = const()[name = tensor("op_2559_dilations_0"), val = tensor([1, 1])]; + tensor var_2559_groups_0 = const()[name = tensor("op_2559_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38621824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38818496))), name = tensor("layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2559_cast_fp16 = conv(dilations = var_2559_dilations_0, groups = var_2559_groups_0, pad = var_2559_pad_0, pad_type = var_2559_pad_type_0, strides = var_2559_strides_0, weight = layers_6_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2559_cast_fp16")]; + tensor var_2565_pad_type_0 = const()[name = tensor("op_2565_pad_type_0"), val = tensor("valid")]; + tensor var_2565_strides_0 = const()[name = tensor("op_2565_strides_0"), val = tensor([1, 1])]; + tensor var_2565_pad_0 = const()[name = tensor("op_2565_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2565_dilations_0 = const()[name = tensor("op_2565_dilations_0"), val = tensor([1, 1])]; + tensor var_2565_groups_0 = const()[name = tensor("op_2565_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38829632))), name = tensor("layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38818688))), shape = tensor([512, 512, 1, 1])]; + tensor var_2565_cast_fp16 = conv(dilations = var_2565_dilations_0, groups = var_2565_groups_0, pad = var_2565_pad_0, pad_type = var_2565_pad_type_0, strides = var_2565_strides_0, weight = layers_6_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2565_cast_fp16")]; + tensor key_13_cast_fp16 = add(x = var_2559_cast_fp16, y = var_2565_cast_fp16)[name = tensor("key_13_cast_fp16")]; + tensor var_2575_pad_type_0 = const()[name = tensor("op_2575_pad_type_0"), val = tensor("valid")]; + tensor var_2575_strides_0 = const()[name = tensor("op_2575_strides_0"), val = tensor([1, 1])]; + tensor var_2575_pad_0 = const()[name = tensor("op_2575_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2575_dilations_0 = const()[name = tensor("op_2575_dilations_0"), val = tensor([1, 1])]; + tensor var_2575_groups_0 = const()[name = tensor("op_2575_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(38862464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39059136))), name = tensor("layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_6_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39059328)))]; + tensor var_2575_cast_fp16 = conv(bias = layers_6_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_2575_dilations_0, groups = var_2575_groups_0, pad = var_2575_pad_0, pad_type = var_2575_pad_type_0, strides = var_2575_strides_0, weight = layers_6_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_27_cast_fp16)[name = tensor("op_2575_cast_fp16")]; + tensor var_2581_pad_type_0 = const()[name = tensor("op_2581_pad_type_0"), val = tensor("valid")]; + tensor var_2581_strides_0 = const()[name = tensor("op_2581_strides_0"), val = tensor([1, 1])]; + tensor var_2581_pad_0 = const()[name = tensor("op_2581_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2581_dilations_0 = const()[name = tensor("op_2581_dilations_0"), val = tensor([1, 1])]; + tensor var_2581_groups_0 = const()[name = tensor("op_2581_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39068864))), name = tensor("layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39060416))), shape = tensor([512, 512, 1, 1])]; + tensor var_2581_cast_fp16 = conv(dilations = var_2581_dilations_0, groups = var_2581_groups_0, pad = var_2581_pad_0, pad_type = var_2581_pad_type_0, strides = var_2581_strides_0, weight = layers_6_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_27_cast_fp16)[name = tensor("op_2581_cast_fp16")]; + tensor value_13_cast_fp16 = add(x = var_2575_cast_fp16, y = var_2581_cast_fp16)[name = tensor("value_13_cast_fp16")]; + tensor var_2584_to_fp16 = const()[name = tensor("op_2584_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39101696)))]; + tensor query_27_cast_fp16 = add(x = query_25_cast_fp16, y = var_2584_to_fp16)[name = tensor("query_27_cast_fp16")]; + tensor var_2587_to_fp16 = const()[name = tensor("op_2587_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39102784)))]; + tensor q_with_bias_v_13_cast_fp16 = add(x = query_25_cast_fp16, y = var_2587_to_fp16)[name = tensor("q_with_bias_v_13_cast_fp16")]; + tensor var_2597_pad_type_0 = const()[name = tensor("op_2597_pad_type_0"), val = tensor("valid")]; + tensor var_2597_strides_0 = const()[name = tensor("op_2597_strides_0"), val = tensor([1, 1])]; + tensor var_2597_pad_0 = const()[name = tensor("op_2597_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2597_dilations_0 = const()[name = tensor("op_2597_dilations_0"), val = tensor([1, 1])]; + tensor var_2597_groups_0 = const()[name = tensor("op_2597_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39103872))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39300544))), name = tensor("layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2597_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2597_dilations_0, groups = var_2597_groups_0, pad = var_2597_pad_0, pad_type = var_2597_pad_type_0, strides = var_2597_strides_0, weight = layers_6_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2597_cast_fp16")]; + tensor var_2603_pad_type_0 = const()[name = tensor("op_2603_pad_type_0"), val = tensor("valid")]; + tensor var_2603_strides_0 = const()[name = tensor("op_2603_strides_0"), val = tensor([1, 1])]; + tensor var_2603_pad_0 = const()[name = tensor("op_2603_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2603_dilations_0 = const()[name = tensor("op_2603_dilations_0"), val = tensor([1, 1])]; + tensor var_2603_groups_0 = const()[name = tensor("op_2603_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39321024))), name = tensor("layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39300736))), shape = tensor([512, 512, 1, 1])]; + tensor var_2603_cast_fp16 = conv(dilations = var_2603_dilations_0, groups = var_2603_groups_0, pad = var_2603_pad_0, pad_type = var_2603_pad_type_0, strides = var_2603_strides_0, weight = layers_6_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2603_cast_fp16")]; + tensor p_13_cast_fp16 = add(x = var_2597_cast_fp16, y = var_2603_cast_fp16)[name = tensor("p_13_cast_fp16")]; + tensor var_2607 = const()[name = tensor("op_2607"), val = tensor([1, 8, 64, 188])]; + tensor var_2608_cast_fp16 = reshape(shape = var_2607, x = q_with_bias_v_13_cast_fp16)[name = tensor("op_2608_cast_fp16")]; + tensor var_2609 = const()[name = tensor("op_2609"), val = tensor([1, 8, 64, -1])]; + tensor var_2610_cast_fp16 = reshape(shape = var_2609, x = p_13_cast_fp16)[name = tensor("op_2610_cast_fp16")]; + tensor matrix_bd_49_transpose_x_0 = const()[name = tensor("matrix_bd_49_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_49_transpose_y_0 = const()[name = tensor("matrix_bd_49_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_49_cast_fp16 = matmul(transpose_x = matrix_bd_49_transpose_x_0, transpose_y = matrix_bd_49_transpose_y_0, x = var_2608_cast_fp16, y = var_2610_cast_fp16)[name = tensor("matrix_bd_49_cast_fp16")]; + tensor matrix_bd_51_pad_0 = const()[name = tensor("matrix_bd_51_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_51_mode_0 = const()[name = tensor("matrix_bd_51_mode_0"), val = tensor("constant")]; + tensor const_76_to_fp16 = const()[name = tensor("const_76_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_51_cast_fp16 = pad(constant_val = const_76_to_fp16, mode = matrix_bd_51_mode_0, pad = matrix_bd_51_pad_0, x = matrix_bd_49_cast_fp16)[name = tensor("matrix_bd_51_cast_fp16")]; + tensor var_2619 = const()[name = tensor("op_2619"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_53_cast_fp16 = reshape(shape = var_2619, x = matrix_bd_51_cast_fp16)[name = tensor("matrix_bd_53_cast_fp16")]; + tensor var_2623_begin_0 = const()[name = tensor("op_2623_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_2623_end_0 = const()[name = tensor("op_2623_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_2623_end_mask_0 = const()[name = tensor("op_2623_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_2623_cast_fp16 = slice_by_index(begin = var_2623_begin_0, end = var_2623_end_0, end_mask = var_2623_end_mask_0, x = matrix_bd_53_cast_fp16)[name = tensor("op_2623_cast_fp16")]; + tensor var_2624 = const()[name = tensor("op_2624"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_55_cast_fp16 = reshape(shape = var_2624, x = var_2623_cast_fp16)[name = tensor("matrix_bd_55_cast_fp16")]; + tensor var_2629_begin_0 = const()[name = tensor("op_2629_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2629_end_0 = const()[name = tensor("op_2629_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_2629_end_mask_0 = const()[name = tensor("op_2629_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_2629_cast_fp16 = slice_by_index(begin = var_2629_begin_0, end = var_2629_end_0, end_mask = var_2629_end_mask_0, x = matrix_bd_55_cast_fp16)[name = tensor("op_2629_cast_fp16")]; + tensor var_2630_to_fp16 = const()[name = tensor("op_2630_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_13_cast_fp16 = mul(x = var_2629_cast_fp16, y = var_2630_to_fp16)[name = tensor("qk_mask_13_cast_fp16")]; + tensor var_2634 = const()[name = tensor("op_2634"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_13_cast_fp16 = reshape(shape = var_2634, x = query_27_cast_fp16)[name = tensor("mh_q_13_cast_fp16")]; + tensor var_2636_to_fp16 = const()[name = tensor("op_2636_to_fp16"), val = tensor(0x1p-3)]; + tensor var_2637_cast_fp16 = mul(x = mh_q_13_cast_fp16, y = var_2636_to_fp16)[name = tensor("op_2637_cast_fp16")]; + tensor var_2640 = const()[name = tensor("op_2640"), val = tensor([1, 8, 64, 188])]; + tensor var_2641_cast_fp16 = reshape(shape = var_2640, x = key_13_cast_fp16)[name = tensor("op_2641_cast_fp16")]; + tensor mh_w_25_transpose_x_0 = const()[name = tensor("mh_w_25_transpose_x_0"), val = tensor(true)]; + tensor mh_w_25_transpose_y_0 = const()[name = tensor("mh_w_25_transpose_y_0"), val = tensor(false)]; + tensor mh_w_25_cast_fp16 = matmul(transpose_x = mh_w_25_transpose_x_0, transpose_y = mh_w_25_transpose_y_0, x = var_2637_cast_fp16, y = var_2641_cast_fp16)[name = tensor("mh_w_25_cast_fp16")]; + tensor mh_w_27_cast_fp16 = add(x = mh_w_25_cast_fp16, y = qk_mask_13_cast_fp16)[name = tensor("mh_w_27_cast_fp16")]; + tensor var_2645_cast_fp16 = softmax(axis = var_2432, x = mh_w_27_cast_fp16)[name = tensor("op_2645_cast_fp16")]; + tensor var_2646 = const()[name = tensor("op_2646"), val = tensor([1, 8, 64, 188])]; + tensor var_2647_cast_fp16 = reshape(shape = var_2646, x = value_13_cast_fp16)[name = tensor("op_2647_cast_fp16")]; + tensor attn_13_transpose_x_0 = const()[name = tensor("attn_13_transpose_x_0"), val = tensor(false)]; + tensor attn_13_transpose_y_0 = const()[name = tensor("attn_13_transpose_y_0"), val = tensor(true)]; + tensor attn_13_cast_fp16 = matmul(transpose_x = attn_13_transpose_x_0, transpose_y = attn_13_transpose_y_0, x = var_2647_cast_fp16, y = var_2645_cast_fp16)[name = tensor("attn_13_cast_fp16")]; + tensor var_2650 = const()[name = tensor("op_2650"), val = tensor([1, 512, 1, 188])]; + tensor input_179_cast_fp16 = reshape(shape = var_2650, x = attn_13_cast_fp16)[name = tensor("input_179_cast_fp16")]; + tensor var_2660_pad_type_0 = const()[name = tensor("op_2660_pad_type_0"), val = tensor("valid")]; + tensor var_2660_strides_0 = const()[name = tensor("op_2660_strides_0"), val = tensor([1, 1])]; + tensor var_2660_pad_0 = const()[name = tensor("op_2660_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2660_dilations_0 = const()[name = tensor("op_2660_dilations_0"), val = tensor([1, 1])]; + tensor var_2660_groups_0 = const()[name = tensor("op_2660_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39353856))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39550528))), name = tensor("layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_6_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39550720)))]; + tensor var_2660_cast_fp16 = conv(bias = layers_6_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_2660_dilations_0, groups = var_2660_groups_0, pad = var_2660_pad_0, pad_type = var_2660_pad_type_0, strides = var_2660_strides_0, weight = layers_6_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_179_cast_fp16)[name = tensor("op_2660_cast_fp16")]; + tensor var_2666_pad_type_0 = const()[name = tensor("op_2666_pad_type_0"), val = tensor("valid")]; + tensor var_2666_strides_0 = const()[name = tensor("op_2666_strides_0"), val = tensor([1, 1])]; + tensor var_2666_pad_0 = const()[name = tensor("op_2666_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2666_dilations_0 = const()[name = tensor("op_2666_dilations_0"), val = tensor([1, 1])]; + tensor var_2666_groups_0 = const()[name = tensor("op_2666_groups_0"), val = tensor(1)]; + tensor layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39559936))), name = tensor("layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39551808))), shape = tensor([512, 512, 1, 1])]; + tensor var_2666_cast_fp16 = conv(dilations = var_2666_dilations_0, groups = var_2666_groups_0, pad = var_2666_pad_0, pad_type = var_2666_pad_type_0, strides = var_2666_strides_0, weight = layers_6_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_179_cast_fp16)[name = tensor("op_2666_cast_fp16")]; + tensor obj_29_cast_fp16 = add(x = var_2660_cast_fp16, y = var_2666_cast_fp16)[name = tensor("obj_29_cast_fp16")]; + tensor inputs_65_cast_fp16 = add(x = inputs_63_cast_fp16, y = obj_29_cast_fp16)[name = tensor("inputs_65_cast_fp16")]; + tensor out_65_axes_0 = const()[name = tensor("out_65_axes_0"), val = tensor([1])]; + tensor var_2677_to_fp16 = const()[name = tensor("op_2677_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_65_cast_fp16 = layer_norm(axes = out_65_axes_0, epsilon = var_2677_to_fp16, x = inputs_65_cast_fp16)[name = tensor("out_65_cast_fp16")]; + tensor input_181_gamma_0_to_fp16 = const()[name = tensor("input_181_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39592768)))]; + tensor input_181_beta_0_to_fp16 = const()[name = tensor("input_181_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39593856)))]; + tensor input_181_epsilon_0_to_fp16 = const()[name = tensor("input_181_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_181_cast_fp16 = batch_norm(beta = input_181_beta_0_to_fp16, epsilon = input_181_epsilon_0_to_fp16, gamma = input_181_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_65_cast_fp16)[name = tensor("input_181_cast_fp16")]; + tensor var_2699_pad_type_0 = const()[name = tensor("op_2699_pad_type_0"), val = tensor("valid")]; + tensor var_2699_strides_0 = const()[name = tensor("op_2699_strides_0"), val = tensor([1, 1])]; + tensor var_2699_pad_0 = const()[name = tensor("op_2699_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2699_dilations_0 = const()[name = tensor("op_2699_dilations_0"), val = tensor([1, 1])]; + tensor var_2699_groups_0 = const()[name = tensor("op_2699_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39594944))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39988224))), name = tensor("layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_6_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39988416)))]; + tensor var_2699_cast_fp16 = conv(bias = layers_6_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_2699_dilations_0, groups = var_2699_groups_0, pad = var_2699_pad_0, pad_type = var_2699_pad_type_0, strides = var_2699_strides_0, weight = layers_6_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_181_cast_fp16)[name = tensor("op_2699_cast_fp16")]; + tensor var_2705_pad_type_0 = const()[name = tensor("op_2705_pad_type_0"), val = tensor("valid")]; + tensor var_2705_strides_0 = const()[name = tensor("op_2705_strides_0"), val = tensor([1, 1])]; + tensor var_2705_pad_0 = const()[name = tensor("op_2705_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2705_dilations_0 = const()[name = tensor("op_2705_dilations_0"), val = tensor([1, 1])]; + tensor var_2705_groups_0 = const()[name = tensor("op_2705_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40007936))), name = tensor("layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(39990528))), shape = tensor([1024, 512, 1, 1])]; + tensor var_2705_cast_fp16 = conv(dilations = var_2705_dilations_0, groups = var_2705_groups_0, pad = var_2705_pad_0, pad_type = var_2705_pad_type_0, strides = var_2705_strides_0, weight = layers_6_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_181_cast_fp16)[name = tensor("op_2705_cast_fp16")]; + tensor input_183_cast_fp16 = add(x = var_2699_cast_fp16, y = var_2705_cast_fp16)[name = tensor("input_183_cast_fp16")]; + tensor input_185_split_num_splits_0 = const()[name = tensor("input_185_split_num_splits_0"), val = tensor(2)]; + tensor input_185_split_axis_0 = const()[name = tensor("input_185_split_axis_0"), val = tensor(1)]; + tensor input_185_split_cast_fp16_0, tensor input_185_split_cast_fp16_1 = split(axis = input_185_split_axis_0, num_splits = input_185_split_num_splits_0, x = input_183_cast_fp16)[name = tensor("input_185_split_cast_fp16")]; + tensor input_185_split_1_sigmoid_cast_fp16 = sigmoid(x = input_185_split_cast_fp16_1)[name = tensor("input_185_split_1_sigmoid_cast_fp16")]; + tensor input_185_cast_fp16 = mul(x = input_185_split_cast_fp16_0, y = input_185_split_1_sigmoid_cast_fp16)[name = tensor("input_185_cast_fp16")]; + tensor input_187_pad_type_0 = const()[name = tensor("input_187_pad_type_0"), val = tensor("custom")]; + tensor input_187_pad_0 = const()[name = tensor("input_187_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_187_groups_0 = const()[name = tensor("input_187_groups_0"), val = tensor(512)]; + tensor input_187_strides_0 = const()[name = tensor("input_187_strides_0"), val = tensor([1, 1])]; + tensor input_187_dilations_0 = const()[name = tensor("input_187_dilations_0"), val = tensor([1, 1])]; + tensor const_203_to_fp16 = const()[name = tensor("const_203_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40073536)))]; + tensor const_204_to_fp16 = const()[name = tensor("const_204_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40082816)))]; + tensor input_189_cast_fp16 = conv(bias = const_204_to_fp16, dilations = input_187_dilations_0, groups = input_187_groups_0, pad = input_187_pad_0, pad_type = input_187_pad_type_0, strides = input_187_strides_0, weight = const_203_to_fp16, x = input_185_cast_fp16)[name = tensor("input_189_cast_fp16")]; + tensor input_191_cast_fp16 = silu(x = input_189_cast_fp16)[name = tensor("input_191_cast_fp16")]; + tensor var_2729_pad_type_0 = const()[name = tensor("op_2729_pad_type_0"), val = tensor("valid")]; + tensor var_2729_strides_0 = const()[name = tensor("op_2729_strides_0"), val = tensor([1, 1])]; + tensor var_2729_pad_0 = const()[name = tensor("op_2729_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2729_dilations_0 = const()[name = tensor("op_2729_dilations_0"), val = tensor([1, 1])]; + tensor var_2729_groups_0 = const()[name = tensor("op_2729_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40083904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40280576))), name = tensor("layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_6_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40280768)))]; + tensor var_2729_cast_fp16 = conv(bias = layers_6_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_2729_dilations_0, groups = var_2729_groups_0, pad = var_2729_pad_0, pad_type = var_2729_pad_type_0, strides = var_2729_strides_0, weight = layers_6_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_191_cast_fp16)[name = tensor("op_2729_cast_fp16")]; + tensor var_2735_pad_type_0 = const()[name = tensor("op_2735_pad_type_0"), val = tensor("valid")]; + tensor var_2735_strides_0 = const()[name = tensor("op_2735_strides_0"), val = tensor([1, 1])]; + tensor var_2735_pad_0 = const()[name = tensor("op_2735_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2735_dilations_0 = const()[name = tensor("op_2735_dilations_0"), val = tensor([1, 1])]; + tensor var_2735_groups_0 = const()[name = tensor("op_2735_groups_0"), val = tensor(1)]; + tensor layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40290496))), name = tensor("layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40281856))), shape = tensor([512, 512, 1, 1])]; + tensor var_2735_cast_fp16 = conv(dilations = var_2735_dilations_0, groups = var_2735_groups_0, pad = var_2735_pad_0, pad_type = var_2735_pad_type_0, strides = var_2735_strides_0, weight = layers_6_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_191_cast_fp16)[name = tensor("op_2735_cast_fp16")]; + tensor x_41_cast_fp16 = add(x = var_2729_cast_fp16, y = var_2735_cast_fp16)[name = tensor("x_41_cast_fp16")]; + tensor inputs_67_cast_fp16 = add(x = inputs_65_cast_fp16, y = x_41_cast_fp16)[name = tensor("inputs_67_cast_fp16")]; + tensor out_67_axes_0 = const()[name = tensor("out_67_axes_0"), val = tensor([1])]; + tensor var_2746_to_fp16 = const()[name = tensor("op_2746_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_67_cast_fp16 = layer_norm(axes = out_67_axes_0, epsilon = var_2746_to_fp16, x = inputs_67_cast_fp16)[name = tensor("out_67_cast_fp16")]; + tensor input_193_gamma_0_to_fp16 = const()[name = tensor("input_193_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40323328)))]; + tensor input_193_beta_0_to_fp16 = const()[name = tensor("input_193_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40324416)))]; + tensor input_193_epsilon_0_to_fp16 = const()[name = tensor("input_193_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_193_cast_fp16 = batch_norm(beta = input_193_beta_0_to_fp16, epsilon = input_193_epsilon_0_to_fp16, gamma = input_193_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_67_cast_fp16)[name = tensor("input_193_cast_fp16")]; + tensor var_2766_pad_type_0 = const()[name = tensor("op_2766_pad_type_0"), val = tensor("valid")]; + tensor var_2766_strides_0 = const()[name = tensor("op_2766_strides_0"), val = tensor([1, 1])]; + tensor var_2766_pad_0 = const()[name = tensor("op_2766_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2766_dilations_0 = const()[name = tensor("op_2766_dilations_0"), val = tensor([1, 1])]; + tensor var_2766_groups_0 = const()[name = tensor("op_2766_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(40325504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41112000))), name = tensor("layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_6_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41112192)))]; + tensor var_2766_cast_fp16 = conv(bias = layers_6_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_2766_dilations_0, groups = var_2766_groups_0, pad = var_2766_pad_0, pad_type = var_2766_pad_type_0, strides = var_2766_strides_0, weight = layers_6_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_193_cast_fp16)[name = tensor("op_2766_cast_fp16")]; + tensor var_2772_pad_type_0 = const()[name = tensor("op_2772_pad_type_0"), val = tensor("valid")]; + tensor var_2772_strides_0 = const()[name = tensor("op_2772_strides_0"), val = tensor([1, 1])]; + tensor var_2772_pad_0 = const()[name = tensor("op_2772_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2772_dilations_0 = const()[name = tensor("op_2772_dilations_0"), val = tensor([1, 1])]; + tensor var_2772_groups_0 = const()[name = tensor("op_2772_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41149888))), name = tensor("layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41116352))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2772_cast_fp16 = conv(dilations = var_2772_dilations_0, groups = var_2772_groups_0, pad = var_2772_pad_0, pad_type = var_2772_pad_type_0, strides = var_2772_strides_0, weight = layers_6_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_193_cast_fp16)[name = tensor("op_2772_cast_fp16")]; + tensor input_195_cast_fp16 = add(x = var_2766_cast_fp16, y = var_2772_cast_fp16)[name = tensor("input_195_cast_fp16")]; + tensor input_197_cast_fp16 = silu(x = input_195_cast_fp16)[name = tensor("input_197_cast_fp16")]; + tensor var_2783_pad_type_0 = const()[name = tensor("op_2783_pad_type_0"), val = tensor("valid")]; + tensor var_2783_strides_0 = const()[name = tensor("op_2783_strides_0"), val = tensor([1, 1])]; + tensor var_2783_pad_0 = const()[name = tensor("op_2783_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2783_dilations_0 = const()[name = tensor("op_2783_dilations_0"), val = tensor([1, 1])]; + tensor var_2783_groups_0 = const()[name = tensor("op_2783_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(41281024))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42067520))), name = tensor("layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_6_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_6_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42067712)))]; + tensor var_2783_cast_fp16 = conv(bias = layers_6_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_2783_dilations_0, groups = var_2783_groups_0, pad = var_2783_pad_0, pad_type = var_2783_pad_type_0, strides = var_2783_strides_0, weight = layers_6_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_197_cast_fp16)[name = tensor("op_2783_cast_fp16")]; + tensor var_2789_pad_type_0 = const()[name = tensor("op_2789_pad_type_0"), val = tensor("valid")]; + tensor var_2789_strides_0 = const()[name = tensor("op_2789_strides_0"), val = tensor([1, 1])]; + tensor var_2789_pad_0 = const()[name = tensor("op_2789_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2789_dilations_0 = const()[name = tensor("op_2789_dilations_0"), val = tensor([1, 1])]; + tensor var_2789_groups_0 = const()[name = tensor("op_2789_groups_0"), val = tensor(1)]; + tensor layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42107840))), name = tensor("layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42068800))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2789_cast_fp16 = conv(dilations = var_2789_dilations_0, groups = var_2789_groups_0, pad = var_2789_pad_0, pad_type = var_2789_pad_type_0, strides = var_2789_strides_0, weight = layers_6_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_197_cast_fp16)[name = tensor("op_2789_cast_fp16")]; + tensor x_43_cast_fp16 = add(x = var_2783_cast_fp16, y = var_2789_cast_fp16)[name = tensor("x_43_cast_fp16")]; + tensor var_2791_to_fp16 = const()[name = tensor("op_2791_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2792_cast_fp16 = mul(x = x_43_cast_fp16, y = var_2791_to_fp16)[name = tensor("op_2792_cast_fp16")]; + tensor inputs_69_cast_fp16 = add(x = inputs_67_cast_fp16, y = var_2792_cast_fp16)[name = tensor("inputs_69_cast_fp16")]; + tensor out_69_axes_0 = const()[name = tensor("out_69_axes_0"), val = tensor([1])]; + tensor var_2802_to_fp16 = const()[name = tensor("op_2802_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_69_cast_fp16 = layer_norm(axes = out_69_axes_0, epsilon = var_2802_to_fp16, x = inputs_69_cast_fp16)[name = tensor("out_69_cast_fp16")]; + tensor inputs_71_gamma_0_to_fp16 = const()[name = tensor("inputs_71_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42238976)))]; + tensor inputs_71_beta_0_to_fp16 = const()[name = tensor("inputs_71_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42240064)))]; + tensor inputs_71_epsilon_0_to_fp16 = const()[name = tensor("inputs_71_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_71_cast_fp16 = batch_norm(beta = inputs_71_beta_0_to_fp16, epsilon = inputs_71_epsilon_0_to_fp16, gamma = inputs_71_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_69_cast_fp16)[name = tensor("inputs_71_cast_fp16")]; + tensor var_2816 = const()[name = tensor("op_2816"), val = tensor(3)]; + tensor out_71_axes_0 = const()[name = tensor("out_71_axes_0"), val = tensor([1])]; + tensor var_2847_to_fp16 = const()[name = tensor("op_2847_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_71_cast_fp16 = layer_norm(axes = out_71_axes_0, epsilon = var_2847_to_fp16, x = inputs_71_cast_fp16)[name = tensor("out_71_cast_fp16")]; + tensor input_199_gamma_0_to_fp16 = const()[name = tensor("input_199_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42241152)))]; + tensor input_199_beta_0_to_fp16 = const()[name = tensor("input_199_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42242240)))]; + tensor input_199_epsilon_0_to_fp16 = const()[name = tensor("input_199_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_199_cast_fp16 = batch_norm(beta = input_199_beta_0_to_fp16, epsilon = input_199_epsilon_0_to_fp16, gamma = input_199_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_71_cast_fp16)[name = tensor("input_199_cast_fp16")]; + tensor var_2867_pad_type_0 = const()[name = tensor("op_2867_pad_type_0"), val = tensor("valid")]; + tensor var_2867_strides_0 = const()[name = tensor("op_2867_strides_0"), val = tensor([1, 1])]; + tensor var_2867_pad_0 = const()[name = tensor("op_2867_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2867_dilations_0 = const()[name = tensor("op_2867_dilations_0"), val = tensor([1, 1])]; + tensor var_2867_groups_0 = const()[name = tensor("op_2867_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(42243328))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43029824))), name = tensor("layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_7_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43030016)))]; + tensor var_2867_cast_fp16 = conv(bias = layers_7_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_2867_dilations_0, groups = var_2867_groups_0, pad = var_2867_pad_0, pad_type = var_2867_pad_type_0, strides = var_2867_strides_0, weight = layers_7_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_199_cast_fp16)[name = tensor("op_2867_cast_fp16")]; + tensor var_2873_pad_type_0 = const()[name = tensor("op_2873_pad_type_0"), val = tensor("valid")]; + tensor var_2873_strides_0 = const()[name = tensor("op_2873_strides_0"), val = tensor([1, 1])]; + tensor var_2873_pad_0 = const()[name = tensor("op_2873_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2873_dilations_0 = const()[name = tensor("op_2873_dilations_0"), val = tensor([1, 1])]; + tensor var_2873_groups_0 = const()[name = tensor("op_2873_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43068288))), name = tensor("layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43034176))), shape = tensor([2048, 512, 1, 1])]; + tensor var_2873_cast_fp16 = conv(dilations = var_2873_dilations_0, groups = var_2873_groups_0, pad = var_2873_pad_0, pad_type = var_2873_pad_type_0, strides = var_2873_strides_0, weight = layers_7_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_199_cast_fp16)[name = tensor("op_2873_cast_fp16")]; + tensor input_201_cast_fp16 = add(x = var_2867_cast_fp16, y = var_2873_cast_fp16)[name = tensor("input_201_cast_fp16")]; + tensor input_203_cast_fp16 = silu(x = input_201_cast_fp16)[name = tensor("input_203_cast_fp16")]; + tensor var_2884_pad_type_0 = const()[name = tensor("op_2884_pad_type_0"), val = tensor("valid")]; + tensor var_2884_strides_0 = const()[name = tensor("op_2884_strides_0"), val = tensor([1, 1])]; + tensor var_2884_pad_0 = const()[name = tensor("op_2884_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2884_dilations_0 = const()[name = tensor("op_2884_dilations_0"), val = tensor([1, 1])]; + tensor var_2884_groups_0 = const()[name = tensor("op_2884_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43199424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43985920))), name = tensor("layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_7_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43986112)))]; + tensor var_2884_cast_fp16 = conv(bias = layers_7_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_2884_dilations_0, groups = var_2884_groups_0, pad = var_2884_pad_0, pad_type = var_2884_pad_type_0, strides = var_2884_strides_0, weight = layers_7_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_203_cast_fp16)[name = tensor("op_2884_cast_fp16")]; + tensor var_2890_pad_type_0 = const()[name = tensor("op_2890_pad_type_0"), val = tensor("valid")]; + tensor var_2890_strides_0 = const()[name = tensor("op_2890_strides_0"), val = tensor([1, 1])]; + tensor var_2890_pad_0 = const()[name = tensor("op_2890_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2890_dilations_0 = const()[name = tensor("op_2890_dilations_0"), val = tensor([1, 1])]; + tensor var_2890_groups_0 = const()[name = tensor("op_2890_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44027328))), name = tensor("layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(43987200))), shape = tensor([512, 2048, 1, 1])]; + tensor var_2890_cast_fp16 = conv(dilations = var_2890_dilations_0, groups = var_2890_groups_0, pad = var_2890_pad_0, pad_type = var_2890_pad_type_0, strides = var_2890_strides_0, weight = layers_7_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_203_cast_fp16)[name = tensor("op_2890_cast_fp16")]; + tensor x_45_cast_fp16 = add(x = var_2884_cast_fp16, y = var_2890_cast_fp16)[name = tensor("x_45_cast_fp16")]; + tensor var_2892_to_fp16 = const()[name = tensor("op_2892_to_fp16"), val = tensor(0x1p-1)]; + tensor var_2893_cast_fp16 = mul(x = x_45_cast_fp16, y = var_2892_to_fp16)[name = tensor("op_2893_cast_fp16")]; + tensor inputs_73_cast_fp16 = add(x = inputs_71_cast_fp16, y = var_2893_cast_fp16)[name = tensor("inputs_73_cast_fp16")]; + tensor out_73_axes_0 = const()[name = tensor("out_73_axes_0"), val = tensor([1])]; + tensor var_2903_to_fp16 = const()[name = tensor("op_2903_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_73_cast_fp16 = layer_norm(axes = out_73_axes_0, epsilon = var_2903_to_fp16, x = inputs_73_cast_fp16)[name = tensor("out_73_cast_fp16")]; + tensor obj_31_gamma_0_to_fp16 = const()[name = tensor("obj_31_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44158464)))]; + tensor obj_31_beta_0_to_fp16 = const()[name = tensor("obj_31_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44159552)))]; + tensor obj_31_epsilon_0_to_fp16 = const()[name = tensor("obj_31_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_31_cast_fp16 = batch_norm(beta = obj_31_beta_0_to_fp16, epsilon = obj_31_epsilon_0_to_fp16, gamma = obj_31_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_73_cast_fp16)[name = tensor("obj_31_cast_fp16")]; + tensor var_2928_pad_type_0 = const()[name = tensor("op_2928_pad_type_0"), val = tensor("valid")]; + tensor var_2928_strides_0 = const()[name = tensor("op_2928_strides_0"), val = tensor([1, 1])]; + tensor var_2928_pad_0 = const()[name = tensor("op_2928_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2928_dilations_0 = const()[name = tensor("op_2928_dilations_0"), val = tensor([1, 1])]; + tensor var_2928_groups_0 = const()[name = tensor("op_2928_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44160640))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44357312))), name = tensor("layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_7_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44357504)))]; + tensor var_2928_cast_fp16 = conv(bias = layers_7_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_2928_dilations_0, groups = var_2928_groups_0, pad = var_2928_pad_0, pad_type = var_2928_pad_type_0, strides = var_2928_strides_0, weight = layers_7_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2928_cast_fp16")]; + tensor var_2934_pad_type_0 = const()[name = tensor("op_2934_pad_type_0"), val = tensor("valid")]; + tensor var_2934_strides_0 = const()[name = tensor("op_2934_strides_0"), val = tensor([1, 1])]; + tensor var_2934_pad_0 = const()[name = tensor("op_2934_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2934_dilations_0 = const()[name = tensor("op_2934_dilations_0"), val = tensor([1, 1])]; + tensor var_2934_groups_0 = const()[name = tensor("op_2934_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44368832))), name = tensor("layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44358592))), shape = tensor([512, 512, 1, 1])]; + tensor var_2934_cast_fp16 = conv(dilations = var_2934_dilations_0, groups = var_2934_groups_0, pad = var_2934_pad_0, pad_type = var_2934_pad_type_0, strides = var_2934_strides_0, weight = layers_7_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2934_cast_fp16")]; + tensor query_29_cast_fp16 = add(x = var_2928_cast_fp16, y = var_2934_cast_fp16)[name = tensor("query_29_cast_fp16")]; + tensor var_2943_pad_type_0 = const()[name = tensor("op_2943_pad_type_0"), val = tensor("valid")]; + tensor var_2943_strides_0 = const()[name = tensor("op_2943_strides_0"), val = tensor([1, 1])]; + tensor var_2943_pad_0 = const()[name = tensor("op_2943_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2943_dilations_0 = const()[name = tensor("op_2943_dilations_0"), val = tensor([1, 1])]; + tensor var_2943_groups_0 = const()[name = tensor("op_2943_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44401664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44598336))), name = tensor("layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2943_cast_fp16 = conv(dilations = var_2943_dilations_0, groups = var_2943_groups_0, pad = var_2943_pad_0, pad_type = var_2943_pad_type_0, strides = var_2943_strides_0, weight = layers_7_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2943_cast_fp16")]; + tensor var_2949_pad_type_0 = const()[name = tensor("op_2949_pad_type_0"), val = tensor("valid")]; + tensor var_2949_strides_0 = const()[name = tensor("op_2949_strides_0"), val = tensor([1, 1])]; + tensor var_2949_pad_0 = const()[name = tensor("op_2949_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2949_dilations_0 = const()[name = tensor("op_2949_dilations_0"), val = tensor([1, 1])]; + tensor var_2949_groups_0 = const()[name = tensor("op_2949_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44611072))), name = tensor("layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44598528))), shape = tensor([512, 512, 1, 1])]; + tensor var_2949_cast_fp16 = conv(dilations = var_2949_dilations_0, groups = var_2949_groups_0, pad = var_2949_pad_0, pad_type = var_2949_pad_type_0, strides = var_2949_strides_0, weight = layers_7_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2949_cast_fp16")]; + tensor key_15_cast_fp16 = add(x = var_2943_cast_fp16, y = var_2949_cast_fp16)[name = tensor("key_15_cast_fp16")]; + tensor var_2959_pad_type_0 = const()[name = tensor("op_2959_pad_type_0"), val = tensor("valid")]; + tensor var_2959_strides_0 = const()[name = tensor("op_2959_strides_0"), val = tensor([1, 1])]; + tensor var_2959_pad_0 = const()[name = tensor("op_2959_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2959_dilations_0 = const()[name = tensor("op_2959_dilations_0"), val = tensor([1, 1])]; + tensor var_2959_groups_0 = const()[name = tensor("op_2959_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44643904))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44840576))), name = tensor("layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_7_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44840768)))]; + tensor var_2959_cast_fp16 = conv(bias = layers_7_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_2959_dilations_0, groups = var_2959_groups_0, pad = var_2959_pad_0, pad_type = var_2959_pad_type_0, strides = var_2959_strides_0, weight = layers_7_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_31_cast_fp16)[name = tensor("op_2959_cast_fp16")]; + tensor var_2965_pad_type_0 = const()[name = tensor("op_2965_pad_type_0"), val = tensor("valid")]; + tensor var_2965_strides_0 = const()[name = tensor("op_2965_strides_0"), val = tensor([1, 1])]; + tensor var_2965_pad_0 = const()[name = tensor("op_2965_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2965_dilations_0 = const()[name = tensor("op_2965_dilations_0"), val = tensor([1, 1])]; + tensor var_2965_groups_0 = const()[name = tensor("op_2965_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44851456))), name = tensor("layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44841856))), shape = tensor([512, 512, 1, 1])]; + tensor var_2965_cast_fp16 = conv(dilations = var_2965_dilations_0, groups = var_2965_groups_0, pad = var_2965_pad_0, pad_type = var_2965_pad_type_0, strides = var_2965_strides_0, weight = layers_7_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_31_cast_fp16)[name = tensor("op_2965_cast_fp16")]; + tensor value_15_cast_fp16 = add(x = var_2959_cast_fp16, y = var_2965_cast_fp16)[name = tensor("value_15_cast_fp16")]; + tensor var_2968_to_fp16 = const()[name = tensor("op_2968_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44884288)))]; + tensor query_31_cast_fp16 = add(x = query_29_cast_fp16, y = var_2968_to_fp16)[name = tensor("query_31_cast_fp16")]; + tensor var_2971_to_fp16 = const()[name = tensor("op_2971_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44885376)))]; + tensor q_with_bias_v_15_cast_fp16 = add(x = query_29_cast_fp16, y = var_2971_to_fp16)[name = tensor("q_with_bias_v_15_cast_fp16")]; + tensor var_2981_pad_type_0 = const()[name = tensor("op_2981_pad_type_0"), val = tensor("valid")]; + tensor var_2981_strides_0 = const()[name = tensor("op_2981_strides_0"), val = tensor([1, 1])]; + tensor var_2981_pad_0 = const()[name = tensor("op_2981_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2981_dilations_0 = const()[name = tensor("op_2981_dilations_0"), val = tensor([1, 1])]; + tensor var_2981_groups_0 = const()[name = tensor("op_2981_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(44886464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45083136))), name = tensor("layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_2981_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_2981_dilations_0, groups = var_2981_groups_0, pad = var_2981_pad_0, pad_type = var_2981_pad_type_0, strides = var_2981_strides_0, weight = layers_7_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_2981_cast_fp16")]; + tensor var_2987_pad_type_0 = const()[name = tensor("op_2987_pad_type_0"), val = tensor("valid")]; + tensor var_2987_strides_0 = const()[name = tensor("op_2987_strides_0"), val = tensor([1, 1])]; + tensor var_2987_pad_0 = const()[name = tensor("op_2987_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_2987_dilations_0 = const()[name = tensor("op_2987_dilations_0"), val = tensor([1, 1])]; + tensor var_2987_groups_0 = const()[name = tensor("op_2987_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45102592))), name = tensor("layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45083328))), shape = tensor([512, 512, 1, 1])]; + tensor var_2987_cast_fp16 = conv(dilations = var_2987_dilations_0, groups = var_2987_groups_0, pad = var_2987_pad_0, pad_type = var_2987_pad_type_0, strides = var_2987_strides_0, weight = layers_7_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_2987_cast_fp16")]; + tensor p_15_cast_fp16 = add(x = var_2981_cast_fp16, y = var_2987_cast_fp16)[name = tensor("p_15_cast_fp16")]; + tensor var_2991 = const()[name = tensor("op_2991"), val = tensor([1, 8, 64, 188])]; + tensor var_2992_cast_fp16 = reshape(shape = var_2991, x = q_with_bias_v_15_cast_fp16)[name = tensor("op_2992_cast_fp16")]; + tensor var_2993 = const()[name = tensor("op_2993"), val = tensor([1, 8, 64, -1])]; + tensor var_2994_cast_fp16 = reshape(shape = var_2993, x = p_15_cast_fp16)[name = tensor("op_2994_cast_fp16")]; + tensor matrix_bd_57_transpose_x_0 = const()[name = tensor("matrix_bd_57_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_57_transpose_y_0 = const()[name = tensor("matrix_bd_57_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_57_cast_fp16 = matmul(transpose_x = matrix_bd_57_transpose_x_0, transpose_y = matrix_bd_57_transpose_y_0, x = var_2992_cast_fp16, y = var_2994_cast_fp16)[name = tensor("matrix_bd_57_cast_fp16")]; + tensor matrix_bd_59_pad_0 = const()[name = tensor("matrix_bd_59_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_59_mode_0 = const()[name = tensor("matrix_bd_59_mode_0"), val = tensor("constant")]; + tensor const_87_to_fp16 = const()[name = tensor("const_87_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_59_cast_fp16 = pad(constant_val = const_87_to_fp16, mode = matrix_bd_59_mode_0, pad = matrix_bd_59_pad_0, x = matrix_bd_57_cast_fp16)[name = tensor("matrix_bd_59_cast_fp16")]; + tensor var_3003 = const()[name = tensor("op_3003"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_61_cast_fp16 = reshape(shape = var_3003, x = matrix_bd_59_cast_fp16)[name = tensor("matrix_bd_61_cast_fp16")]; + tensor var_3007_begin_0 = const()[name = tensor("op_3007_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3007_end_0 = const()[name = tensor("op_3007_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_3007_end_mask_0 = const()[name = tensor("op_3007_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3007_cast_fp16 = slice_by_index(begin = var_3007_begin_0, end = var_3007_end_0, end_mask = var_3007_end_mask_0, x = matrix_bd_61_cast_fp16)[name = tensor("op_3007_cast_fp16")]; + tensor var_3008 = const()[name = tensor("op_3008"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_63_cast_fp16 = reshape(shape = var_3008, x = var_3007_cast_fp16)[name = tensor("matrix_bd_63_cast_fp16")]; + tensor var_3013_begin_0 = const()[name = tensor("op_3013_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3013_end_0 = const()[name = tensor("op_3013_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3013_end_mask_0 = const()[name = tensor("op_3013_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3013_cast_fp16 = slice_by_index(begin = var_3013_begin_0, end = var_3013_end_0, end_mask = var_3013_end_mask_0, x = matrix_bd_63_cast_fp16)[name = tensor("op_3013_cast_fp16")]; + tensor var_3014_to_fp16 = const()[name = tensor("op_3014_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_15_cast_fp16 = mul(x = var_3013_cast_fp16, y = var_3014_to_fp16)[name = tensor("qk_mask_15_cast_fp16")]; + tensor var_3018 = const()[name = tensor("op_3018"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_15_cast_fp16 = reshape(shape = var_3018, x = query_31_cast_fp16)[name = tensor("mh_q_15_cast_fp16")]; + tensor var_3020_to_fp16 = const()[name = tensor("op_3020_to_fp16"), val = tensor(0x1p-3)]; + tensor var_3021_cast_fp16 = mul(x = mh_q_15_cast_fp16, y = var_3020_to_fp16)[name = tensor("op_3021_cast_fp16")]; + tensor var_3024 = const()[name = tensor("op_3024"), val = tensor([1, 8, 64, 188])]; + tensor var_3025_cast_fp16 = reshape(shape = var_3024, x = key_15_cast_fp16)[name = tensor("op_3025_cast_fp16")]; + tensor mh_w_29_transpose_x_0 = const()[name = tensor("mh_w_29_transpose_x_0"), val = tensor(true)]; + tensor mh_w_29_transpose_y_0 = const()[name = tensor("mh_w_29_transpose_y_0"), val = tensor(false)]; + tensor mh_w_29_cast_fp16 = matmul(transpose_x = mh_w_29_transpose_x_0, transpose_y = mh_w_29_transpose_y_0, x = var_3021_cast_fp16, y = var_3025_cast_fp16)[name = tensor("mh_w_29_cast_fp16")]; + tensor mh_w_31_cast_fp16 = add(x = mh_w_29_cast_fp16, y = qk_mask_15_cast_fp16)[name = tensor("mh_w_31_cast_fp16")]; + tensor var_3029_cast_fp16 = softmax(axis = var_2816, x = mh_w_31_cast_fp16)[name = tensor("op_3029_cast_fp16")]; + tensor var_3030 = const()[name = tensor("op_3030"), val = tensor([1, 8, 64, 188])]; + tensor var_3031_cast_fp16 = reshape(shape = var_3030, x = value_15_cast_fp16)[name = tensor("op_3031_cast_fp16")]; + tensor attn_15_transpose_x_0 = const()[name = tensor("attn_15_transpose_x_0"), val = tensor(false)]; + tensor attn_15_transpose_y_0 = const()[name = tensor("attn_15_transpose_y_0"), val = tensor(true)]; + tensor attn_15_cast_fp16 = matmul(transpose_x = attn_15_transpose_x_0, transpose_y = attn_15_transpose_y_0, x = var_3031_cast_fp16, y = var_3029_cast_fp16)[name = tensor("attn_15_cast_fp16")]; + tensor var_3034 = const()[name = tensor("op_3034"), val = tensor([1, 512, 1, 188])]; + tensor input_205_cast_fp16 = reshape(shape = var_3034, x = attn_15_cast_fp16)[name = tensor("input_205_cast_fp16")]; + tensor var_3044_pad_type_0 = const()[name = tensor("op_3044_pad_type_0"), val = tensor("valid")]; + tensor var_3044_strides_0 = const()[name = tensor("op_3044_strides_0"), val = tensor([1, 1])]; + tensor var_3044_pad_0 = const()[name = tensor("op_3044_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3044_dilations_0 = const()[name = tensor("op_3044_dilations_0"), val = tensor([1, 1])]; + tensor var_3044_groups_0 = const()[name = tensor("op_3044_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45135424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45332096))), name = tensor("layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_7_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45332288)))]; + tensor var_3044_cast_fp16 = conv(bias = layers_7_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_3044_dilations_0, groups = var_3044_groups_0, pad = var_3044_pad_0, pad_type = var_3044_pad_type_0, strides = var_3044_strides_0, weight = layers_7_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_205_cast_fp16)[name = tensor("op_3044_cast_fp16")]; + tensor var_3050_pad_type_0 = const()[name = tensor("op_3050_pad_type_0"), val = tensor("valid")]; + tensor var_3050_strides_0 = const()[name = tensor("op_3050_strides_0"), val = tensor([1, 1])]; + tensor var_3050_pad_0 = const()[name = tensor("op_3050_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3050_dilations_0 = const()[name = tensor("op_3050_dilations_0"), val = tensor([1, 1])]; + tensor var_3050_groups_0 = const()[name = tensor("op_3050_groups_0"), val = tensor(1)]; + tensor layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45342656))), name = tensor("layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45333376))), shape = tensor([512, 512, 1, 1])]; + tensor var_3050_cast_fp16 = conv(dilations = var_3050_dilations_0, groups = var_3050_groups_0, pad = var_3050_pad_0, pad_type = var_3050_pad_type_0, strides = var_3050_strides_0, weight = layers_7_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_205_cast_fp16)[name = tensor("op_3050_cast_fp16")]; + tensor obj_33_cast_fp16 = add(x = var_3044_cast_fp16, y = var_3050_cast_fp16)[name = tensor("obj_33_cast_fp16")]; + tensor inputs_75_cast_fp16 = add(x = inputs_73_cast_fp16, y = obj_33_cast_fp16)[name = tensor("inputs_75_cast_fp16")]; + tensor out_75_axes_0 = const()[name = tensor("out_75_axes_0"), val = tensor([1])]; + tensor var_3061_to_fp16 = const()[name = tensor("op_3061_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_75_cast_fp16 = layer_norm(axes = out_75_axes_0, epsilon = var_3061_to_fp16, x = inputs_75_cast_fp16)[name = tensor("out_75_cast_fp16")]; + tensor input_207_gamma_0_to_fp16 = const()[name = tensor("input_207_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45375488)))]; + tensor input_207_beta_0_to_fp16 = const()[name = tensor("input_207_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45376576)))]; + tensor input_207_epsilon_0_to_fp16 = const()[name = tensor("input_207_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_207_cast_fp16 = batch_norm(beta = input_207_beta_0_to_fp16, epsilon = input_207_epsilon_0_to_fp16, gamma = input_207_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_75_cast_fp16)[name = tensor("input_207_cast_fp16")]; + tensor var_3083_pad_type_0 = const()[name = tensor("op_3083_pad_type_0"), val = tensor("valid")]; + tensor var_3083_strides_0 = const()[name = tensor("op_3083_strides_0"), val = tensor([1, 1])]; + tensor var_3083_pad_0 = const()[name = tensor("op_3083_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3083_dilations_0 = const()[name = tensor("op_3083_dilations_0"), val = tensor([1, 1])]; + tensor var_3083_groups_0 = const()[name = tensor("op_3083_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45377664))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45770944))), name = tensor("layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_7_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45771136)))]; + tensor var_3083_cast_fp16 = conv(bias = layers_7_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_3083_dilations_0, groups = var_3083_groups_0, pad = var_3083_pad_0, pad_type = var_3083_pad_type_0, strides = var_3083_strides_0, weight = layers_7_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_207_cast_fp16)[name = tensor("op_3083_cast_fp16")]; + tensor var_3089_pad_type_0 = const()[name = tensor("op_3089_pad_type_0"), val = tensor("valid")]; + tensor var_3089_strides_0 = const()[name = tensor("op_3089_strides_0"), val = tensor([1, 1])]; + tensor var_3089_pad_0 = const()[name = tensor("op_3089_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3089_dilations_0 = const()[name = tensor("op_3089_dilations_0"), val = tensor([1, 1])]; + tensor var_3089_groups_0 = const()[name = tensor("op_3089_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45791168))), name = tensor("layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45773248))), shape = tensor([1024, 512, 1, 1])]; + tensor var_3089_cast_fp16 = conv(dilations = var_3089_dilations_0, groups = var_3089_groups_0, pad = var_3089_pad_0, pad_type = var_3089_pad_type_0, strides = var_3089_strides_0, weight = layers_7_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_207_cast_fp16)[name = tensor("op_3089_cast_fp16")]; + tensor input_209_cast_fp16 = add(x = var_3083_cast_fp16, y = var_3089_cast_fp16)[name = tensor("input_209_cast_fp16")]; + tensor input_211_split_num_splits_0 = const()[name = tensor("input_211_split_num_splits_0"), val = tensor(2)]; + tensor input_211_split_axis_0 = const()[name = tensor("input_211_split_axis_0"), val = tensor(1)]; + tensor input_211_split_cast_fp16_0, tensor input_211_split_cast_fp16_1 = split(axis = input_211_split_axis_0, num_splits = input_211_split_num_splits_0, x = input_209_cast_fp16)[name = tensor("input_211_split_cast_fp16")]; + tensor input_211_split_1_sigmoid_cast_fp16 = sigmoid(x = input_211_split_cast_fp16_1)[name = tensor("input_211_split_1_sigmoid_cast_fp16")]; + tensor input_211_cast_fp16 = mul(x = input_211_split_cast_fp16_0, y = input_211_split_1_sigmoid_cast_fp16)[name = tensor("input_211_cast_fp16")]; + tensor input_213_pad_type_0 = const()[name = tensor("input_213_pad_type_0"), val = tensor("custom")]; + tensor input_213_pad_0 = const()[name = tensor("input_213_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_213_groups_0 = const()[name = tensor("input_213_groups_0"), val = tensor(512)]; + tensor input_213_strides_0 = const()[name = tensor("input_213_strides_0"), val = tensor([1, 1])]; + tensor input_213_dilations_0 = const()[name = tensor("input_213_dilations_0"), val = tensor([1, 1])]; + tensor const_205_to_fp16 = const()[name = tensor("const_205_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45856768)))]; + tensor const_206_to_fp16 = const()[name = tensor("const_206_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45866048)))]; + tensor input_215_cast_fp16 = conv(bias = const_206_to_fp16, dilations = input_213_dilations_0, groups = input_213_groups_0, pad = input_213_pad_0, pad_type = input_213_pad_type_0, strides = input_213_strides_0, weight = const_205_to_fp16, x = input_211_cast_fp16)[name = tensor("input_215_cast_fp16")]; + tensor input_217_cast_fp16 = silu(x = input_215_cast_fp16)[name = tensor("input_217_cast_fp16")]; + tensor var_3113_pad_type_0 = const()[name = tensor("op_3113_pad_type_0"), val = tensor("valid")]; + tensor var_3113_strides_0 = const()[name = tensor("op_3113_strides_0"), val = tensor([1, 1])]; + tensor var_3113_pad_0 = const()[name = tensor("op_3113_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3113_dilations_0 = const()[name = tensor("op_3113_dilations_0"), val = tensor([1, 1])]; + tensor var_3113_groups_0 = const()[name = tensor("op_3113_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(45867136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46063808))), name = tensor("layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_7_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46064000)))]; + tensor var_3113_cast_fp16 = conv(bias = layers_7_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_3113_dilations_0, groups = var_3113_groups_0, pad = var_3113_pad_0, pad_type = var_3113_pad_type_0, strides = var_3113_strides_0, weight = layers_7_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_217_cast_fp16)[name = tensor("op_3113_cast_fp16")]; + tensor var_3119_pad_type_0 = const()[name = tensor("op_3119_pad_type_0"), val = tensor("valid")]; + tensor var_3119_strides_0 = const()[name = tensor("op_3119_strides_0"), val = tensor([1, 1])]; + tensor var_3119_pad_0 = const()[name = tensor("op_3119_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3119_dilations_0 = const()[name = tensor("op_3119_dilations_0"), val = tensor([1, 1])]; + tensor var_3119_groups_0 = const()[name = tensor("op_3119_groups_0"), val = tensor(1)]; + tensor layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46073536))), name = tensor("layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46065088))), shape = tensor([512, 512, 1, 1])]; + tensor var_3119_cast_fp16 = conv(dilations = var_3119_dilations_0, groups = var_3119_groups_0, pad = var_3119_pad_0, pad_type = var_3119_pad_type_0, strides = var_3119_strides_0, weight = layers_7_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_217_cast_fp16)[name = tensor("op_3119_cast_fp16")]; + tensor x_47_cast_fp16 = add(x = var_3113_cast_fp16, y = var_3119_cast_fp16)[name = tensor("x_47_cast_fp16")]; + tensor inputs_77_cast_fp16 = add(x = inputs_75_cast_fp16, y = x_47_cast_fp16)[name = tensor("inputs_77_cast_fp16")]; + tensor out_77_axes_0 = const()[name = tensor("out_77_axes_0"), val = tensor([1])]; + tensor var_3130_to_fp16 = const()[name = tensor("op_3130_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_77_cast_fp16 = layer_norm(axes = out_77_axes_0, epsilon = var_3130_to_fp16, x = inputs_77_cast_fp16)[name = tensor("out_77_cast_fp16")]; + tensor input_219_gamma_0_to_fp16 = const()[name = tensor("input_219_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46106368)))]; + tensor input_219_beta_0_to_fp16 = const()[name = tensor("input_219_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46107456)))]; + tensor input_219_epsilon_0_to_fp16 = const()[name = tensor("input_219_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_219_cast_fp16 = batch_norm(beta = input_219_beta_0_to_fp16, epsilon = input_219_epsilon_0_to_fp16, gamma = input_219_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_77_cast_fp16)[name = tensor("input_219_cast_fp16")]; + tensor var_3150_pad_type_0 = const()[name = tensor("op_3150_pad_type_0"), val = tensor("valid")]; + tensor var_3150_strides_0 = const()[name = tensor("op_3150_strides_0"), val = tensor([1, 1])]; + tensor var_3150_pad_0 = const()[name = tensor("op_3150_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3150_dilations_0 = const()[name = tensor("op_3150_dilations_0"), val = tensor([1, 1])]; + tensor var_3150_groups_0 = const()[name = tensor("op_3150_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46108544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46895040))), name = tensor("layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_7_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46895232)))]; + tensor var_3150_cast_fp16 = conv(bias = layers_7_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_3150_dilations_0, groups = var_3150_groups_0, pad = var_3150_pad_0, pad_type = var_3150_pad_type_0, strides = var_3150_strides_0, weight = layers_7_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_219_cast_fp16)[name = tensor("op_3150_cast_fp16")]; + tensor var_3156_pad_type_0 = const()[name = tensor("op_3156_pad_type_0"), val = tensor("valid")]; + tensor var_3156_strides_0 = const()[name = tensor("op_3156_strides_0"), val = tensor([1, 1])]; + tensor var_3156_pad_0 = const()[name = tensor("op_3156_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3156_dilations_0 = const()[name = tensor("op_3156_dilations_0"), val = tensor([1, 1])]; + tensor var_3156_groups_0 = const()[name = tensor("op_3156_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46930816))), name = tensor("layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(46899392))), shape = tensor([2048, 512, 1, 1])]; + tensor var_3156_cast_fp16 = conv(dilations = var_3156_dilations_0, groups = var_3156_groups_0, pad = var_3156_pad_0, pad_type = var_3156_pad_type_0, strides = var_3156_strides_0, weight = layers_7_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_219_cast_fp16)[name = tensor("op_3156_cast_fp16")]; + tensor input_221_cast_fp16 = add(x = var_3150_cast_fp16, y = var_3156_cast_fp16)[name = tensor("input_221_cast_fp16")]; + tensor input_223_cast_fp16 = silu(x = input_221_cast_fp16)[name = tensor("input_223_cast_fp16")]; + tensor var_3167_pad_type_0 = const()[name = tensor("op_3167_pad_type_0"), val = tensor("valid")]; + tensor var_3167_strides_0 = const()[name = tensor("op_3167_strides_0"), val = tensor([1, 1])]; + tensor var_3167_pad_0 = const()[name = tensor("op_3167_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3167_dilations_0 = const()[name = tensor("op_3167_dilations_0"), val = tensor([1, 1])]; + tensor var_3167_groups_0 = const()[name = tensor("op_3167_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47061952))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47848448))), name = tensor("layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_7_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_7_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47848640)))]; + tensor var_3167_cast_fp16 = conv(bias = layers_7_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_3167_dilations_0, groups = var_3167_groups_0, pad = var_3167_pad_0, pad_type = var_3167_pad_type_0, strides = var_3167_strides_0, weight = layers_7_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_223_cast_fp16)[name = tensor("op_3167_cast_fp16")]; + tensor var_3173_pad_type_0 = const()[name = tensor("op_3173_pad_type_0"), val = tensor("valid")]; + tensor var_3173_strides_0 = const()[name = tensor("op_3173_strides_0"), val = tensor([1, 1])]; + tensor var_3173_pad_0 = const()[name = tensor("op_3173_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3173_dilations_0 = const()[name = tensor("op_3173_dilations_0"), val = tensor([1, 1])]; + tensor var_3173_groups_0 = const()[name = tensor("op_3173_groups_0"), val = tensor(1)]; + tensor layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47885632))), name = tensor("layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(47849728))), shape = tensor([512, 2048, 1, 1])]; + tensor var_3173_cast_fp16 = conv(dilations = var_3173_dilations_0, groups = var_3173_groups_0, pad = var_3173_pad_0, pad_type = var_3173_pad_type_0, strides = var_3173_strides_0, weight = layers_7_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_223_cast_fp16)[name = tensor("op_3173_cast_fp16")]; + tensor x_49_cast_fp16 = add(x = var_3167_cast_fp16, y = var_3173_cast_fp16)[name = tensor("x_49_cast_fp16")]; + tensor var_3175_to_fp16 = const()[name = tensor("op_3175_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3176_cast_fp16 = mul(x = x_49_cast_fp16, y = var_3175_to_fp16)[name = tensor("op_3176_cast_fp16")]; + tensor inputs_79_cast_fp16 = add(x = inputs_77_cast_fp16, y = var_3176_cast_fp16)[name = tensor("inputs_79_cast_fp16")]; + tensor out_79_axes_0 = const()[name = tensor("out_79_axes_0"), val = tensor([1])]; + tensor var_3186_to_fp16 = const()[name = tensor("op_3186_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_79_cast_fp16 = layer_norm(axes = out_79_axes_0, epsilon = var_3186_to_fp16, x = inputs_79_cast_fp16)[name = tensor("out_79_cast_fp16")]; + tensor inputs_81_gamma_0_to_fp16 = const()[name = tensor("inputs_81_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48016768)))]; + tensor inputs_81_beta_0_to_fp16 = const()[name = tensor("inputs_81_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48017856)))]; + tensor inputs_81_epsilon_0_to_fp16 = const()[name = tensor("inputs_81_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_81_cast_fp16 = batch_norm(beta = inputs_81_beta_0_to_fp16, epsilon = inputs_81_epsilon_0_to_fp16, gamma = inputs_81_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_79_cast_fp16)[name = tensor("inputs_81_cast_fp16")]; + tensor var_3200 = const()[name = tensor("op_3200"), val = tensor(3)]; + tensor out_81_axes_0 = const()[name = tensor("out_81_axes_0"), val = tensor([1])]; + tensor var_3231_to_fp16 = const()[name = tensor("op_3231_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_81_cast_fp16 = layer_norm(axes = out_81_axes_0, epsilon = var_3231_to_fp16, x = inputs_81_cast_fp16)[name = tensor("out_81_cast_fp16")]; + tensor input_225_gamma_0_to_fp16 = const()[name = tensor("input_225_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48018944)))]; + tensor input_225_beta_0_to_fp16 = const()[name = tensor("input_225_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48020032)))]; + tensor input_225_epsilon_0_to_fp16 = const()[name = tensor("input_225_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_225_cast_fp16 = batch_norm(beta = input_225_beta_0_to_fp16, epsilon = input_225_epsilon_0_to_fp16, gamma = input_225_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_81_cast_fp16)[name = tensor("input_225_cast_fp16")]; + tensor var_3251_pad_type_0 = const()[name = tensor("op_3251_pad_type_0"), val = tensor("valid")]; + tensor var_3251_strides_0 = const()[name = tensor("op_3251_strides_0"), val = tensor([1, 1])]; + tensor var_3251_pad_0 = const()[name = tensor("op_3251_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3251_dilations_0 = const()[name = tensor("op_3251_dilations_0"), val = tensor([1, 1])]; + tensor var_3251_groups_0 = const()[name = tensor("op_3251_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48021120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48807616))), name = tensor("layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_8_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48807808)))]; + tensor var_3251_cast_fp16 = conv(bias = layers_8_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3251_dilations_0, groups = var_3251_groups_0, pad = var_3251_pad_0, pad_type = var_3251_pad_type_0, strides = var_3251_strides_0, weight = layers_8_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_225_cast_fp16)[name = tensor("op_3251_cast_fp16")]; + tensor var_3257_pad_type_0 = const()[name = tensor("op_3257_pad_type_0"), val = tensor("valid")]; + tensor var_3257_strides_0 = const()[name = tensor("op_3257_strides_0"), val = tensor([1, 1])]; + tensor var_3257_pad_0 = const()[name = tensor("op_3257_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3257_dilations_0 = const()[name = tensor("op_3257_dilations_0"), val = tensor([1, 1])]; + tensor var_3257_groups_0 = const()[name = tensor("op_3257_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48844544))), name = tensor("layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48811968))), shape = tensor([2048, 512, 1, 1])]; + tensor var_3257_cast_fp16 = conv(dilations = var_3257_dilations_0, groups = var_3257_groups_0, pad = var_3257_pad_0, pad_type = var_3257_pad_type_0, strides = var_3257_strides_0, weight = layers_8_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_225_cast_fp16)[name = tensor("op_3257_cast_fp16")]; + tensor input_227_cast_fp16 = add(x = var_3251_cast_fp16, y = var_3257_cast_fp16)[name = tensor("input_227_cast_fp16")]; + tensor input_229_cast_fp16 = silu(x = input_227_cast_fp16)[name = tensor("input_229_cast_fp16")]; + tensor var_3268_pad_type_0 = const()[name = tensor("op_3268_pad_type_0"), val = tensor("valid")]; + tensor var_3268_strides_0 = const()[name = tensor("op_3268_strides_0"), val = tensor([1, 1])]; + tensor var_3268_pad_0 = const()[name = tensor("op_3268_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3268_dilations_0 = const()[name = tensor("op_3268_dilations_0"), val = tensor([1, 1])]; + tensor var_3268_groups_0 = const()[name = tensor("op_3268_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(48975680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49762176))), name = tensor("layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_8_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49762368)))]; + tensor var_3268_cast_fp16 = conv(bias = layers_8_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_3268_dilations_0, groups = var_3268_groups_0, pad = var_3268_pad_0, pad_type = var_3268_pad_type_0, strides = var_3268_strides_0, weight = layers_8_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_229_cast_fp16)[name = tensor("op_3268_cast_fp16")]; + tensor var_3274_pad_type_0 = const()[name = tensor("op_3274_pad_type_0"), val = tensor("valid")]; + tensor var_3274_strides_0 = const()[name = tensor("op_3274_strides_0"), val = tensor([1, 1])]; + tensor var_3274_pad_0 = const()[name = tensor("op_3274_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3274_dilations_0 = const()[name = tensor("op_3274_dilations_0"), val = tensor([1, 1])]; + tensor var_3274_groups_0 = const()[name = tensor("op_3274_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49800128))), name = tensor("layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49763456))), shape = tensor([512, 2048, 1, 1])]; + tensor var_3274_cast_fp16 = conv(dilations = var_3274_dilations_0, groups = var_3274_groups_0, pad = var_3274_pad_0, pad_type = var_3274_pad_type_0, strides = var_3274_strides_0, weight = layers_8_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_229_cast_fp16)[name = tensor("op_3274_cast_fp16")]; + tensor x_51_cast_fp16 = add(x = var_3268_cast_fp16, y = var_3274_cast_fp16)[name = tensor("x_51_cast_fp16")]; + tensor var_3276_to_fp16 = const()[name = tensor("op_3276_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3277_cast_fp16 = mul(x = x_51_cast_fp16, y = var_3276_to_fp16)[name = tensor("op_3277_cast_fp16")]; + tensor inputs_83_cast_fp16 = add(x = inputs_81_cast_fp16, y = var_3277_cast_fp16)[name = tensor("inputs_83_cast_fp16")]; + tensor out_83_axes_0 = const()[name = tensor("out_83_axes_0"), val = tensor([1])]; + tensor var_3287_to_fp16 = const()[name = tensor("op_3287_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_83_cast_fp16 = layer_norm(axes = out_83_axes_0, epsilon = var_3287_to_fp16, x = inputs_83_cast_fp16)[name = tensor("out_83_cast_fp16")]; + tensor obj_35_gamma_0_to_fp16 = const()[name = tensor("obj_35_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49931264)))]; + tensor obj_35_beta_0_to_fp16 = const()[name = tensor("obj_35_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49932352)))]; + tensor obj_35_epsilon_0_to_fp16 = const()[name = tensor("obj_35_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_35_cast_fp16 = batch_norm(beta = obj_35_beta_0_to_fp16, epsilon = obj_35_epsilon_0_to_fp16, gamma = obj_35_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_83_cast_fp16)[name = tensor("obj_35_cast_fp16")]; + tensor var_3312_pad_type_0 = const()[name = tensor("op_3312_pad_type_0"), val = tensor("valid")]; + tensor var_3312_strides_0 = const()[name = tensor("op_3312_strides_0"), val = tensor([1, 1])]; + tensor var_3312_pad_0 = const()[name = tensor("op_3312_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3312_dilations_0 = const()[name = tensor("op_3312_dilations_0"), val = tensor([1, 1])]; + tensor var_3312_groups_0 = const()[name = tensor("op_3312_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(49933440))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50130112))), name = tensor("layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_8_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50130304)))]; + tensor var_3312_cast_fp16 = conv(bias = layers_8_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_3312_dilations_0, groups = var_3312_groups_0, pad = var_3312_pad_0, pad_type = var_3312_pad_type_0, strides = var_3312_strides_0, weight = layers_8_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3312_cast_fp16")]; + tensor var_3318_pad_type_0 = const()[name = tensor("op_3318_pad_type_0"), val = tensor("valid")]; + tensor var_3318_strides_0 = const()[name = tensor("op_3318_strides_0"), val = tensor([1, 1])]; + tensor var_3318_pad_0 = const()[name = tensor("op_3318_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3318_dilations_0 = const()[name = tensor("op_3318_dilations_0"), val = tensor([1, 1])]; + tensor var_3318_groups_0 = const()[name = tensor("op_3318_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50142912))), name = tensor("layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50131392))), shape = tensor([512, 512, 1, 1])]; + tensor var_3318_cast_fp16 = conv(dilations = var_3318_dilations_0, groups = var_3318_groups_0, pad = var_3318_pad_0, pad_type = var_3318_pad_type_0, strides = var_3318_strides_0, weight = layers_8_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3318_cast_fp16")]; + tensor query_33_cast_fp16 = add(x = var_3312_cast_fp16, y = var_3318_cast_fp16)[name = tensor("query_33_cast_fp16")]; + tensor var_3327_pad_type_0 = const()[name = tensor("op_3327_pad_type_0"), val = tensor("valid")]; + tensor var_3327_strides_0 = const()[name = tensor("op_3327_strides_0"), val = tensor([1, 1])]; + tensor var_3327_pad_0 = const()[name = tensor("op_3327_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3327_dilations_0 = const()[name = tensor("op_3327_dilations_0"), val = tensor([1, 1])]; + tensor var_3327_groups_0 = const()[name = tensor("op_3327_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50175744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50372416))), name = tensor("layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_3327_cast_fp16 = conv(dilations = var_3327_dilations_0, groups = var_3327_groups_0, pad = var_3327_pad_0, pad_type = var_3327_pad_type_0, strides = var_3327_strides_0, weight = layers_8_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3327_cast_fp16")]; + tensor var_3333_pad_type_0 = const()[name = tensor("op_3333_pad_type_0"), val = tensor("valid")]; + tensor var_3333_strides_0 = const()[name = tensor("op_3333_strides_0"), val = tensor([1, 1])]; + tensor var_3333_pad_0 = const()[name = tensor("op_3333_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3333_dilations_0 = const()[name = tensor("op_3333_dilations_0"), val = tensor([1, 1])]; + tensor var_3333_groups_0 = const()[name = tensor("op_3333_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50381632))), name = tensor("layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50372608))), shape = tensor([512, 512, 1, 1])]; + tensor var_3333_cast_fp16 = conv(dilations = var_3333_dilations_0, groups = var_3333_groups_0, pad = var_3333_pad_0, pad_type = var_3333_pad_type_0, strides = var_3333_strides_0, weight = layers_8_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3333_cast_fp16")]; + tensor key_17_cast_fp16 = add(x = var_3327_cast_fp16, y = var_3333_cast_fp16)[name = tensor("key_17_cast_fp16")]; + tensor var_3343_pad_type_0 = const()[name = tensor("op_3343_pad_type_0"), val = tensor("valid")]; + tensor var_3343_strides_0 = const()[name = tensor("op_3343_strides_0"), val = tensor([1, 1])]; + tensor var_3343_pad_0 = const()[name = tensor("op_3343_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3343_dilations_0 = const()[name = tensor("op_3343_dilations_0"), val = tensor([1, 1])]; + tensor var_3343_groups_0 = const()[name = tensor("op_3343_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50414464))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50611136))), name = tensor("layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_8_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50611328)))]; + tensor var_3343_cast_fp16 = conv(bias = layers_8_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_3343_dilations_0, groups = var_3343_groups_0, pad = var_3343_pad_0, pad_type = var_3343_pad_type_0, strides = var_3343_strides_0, weight = layers_8_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_35_cast_fp16)[name = tensor("op_3343_cast_fp16")]; + tensor var_3349_pad_type_0 = const()[name = tensor("op_3349_pad_type_0"), val = tensor("valid")]; + tensor var_3349_strides_0 = const()[name = tensor("op_3349_strides_0"), val = tensor([1, 1])]; + tensor var_3349_pad_0 = const()[name = tensor("op_3349_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3349_dilations_0 = const()[name = tensor("op_3349_dilations_0"), val = tensor([1, 1])]; + tensor var_3349_groups_0 = const()[name = tensor("op_3349_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50621184))), name = tensor("layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50612416))), shape = tensor([512, 512, 1, 1])]; + tensor var_3349_cast_fp16 = conv(dilations = var_3349_dilations_0, groups = var_3349_groups_0, pad = var_3349_pad_0, pad_type = var_3349_pad_type_0, strides = var_3349_strides_0, weight = layers_8_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_35_cast_fp16)[name = tensor("op_3349_cast_fp16")]; + tensor value_17_cast_fp16 = add(x = var_3343_cast_fp16, y = var_3349_cast_fp16)[name = tensor("value_17_cast_fp16")]; + tensor var_3352_to_fp16 = const()[name = tensor("op_3352_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50654016)))]; + tensor query_35_cast_fp16 = add(x = query_33_cast_fp16, y = var_3352_to_fp16)[name = tensor("query_35_cast_fp16")]; + tensor var_3355_to_fp16 = const()[name = tensor("op_3355_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50655104)))]; + tensor q_with_bias_v_17_cast_fp16 = add(x = query_33_cast_fp16, y = var_3355_to_fp16)[name = tensor("q_with_bias_v_17_cast_fp16")]; + tensor var_3365_pad_type_0 = const()[name = tensor("op_3365_pad_type_0"), val = tensor("valid")]; + tensor var_3365_strides_0 = const()[name = tensor("op_3365_strides_0"), val = tensor([1, 1])]; + tensor var_3365_pad_0 = const()[name = tensor("op_3365_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3365_dilations_0 = const()[name = tensor("op_3365_dilations_0"), val = tensor([1, 1])]; + tensor var_3365_groups_0 = const()[name = tensor("op_3365_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50656192))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50852864))), name = tensor("layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_3365_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3365_dilations_0, groups = var_3365_groups_0, pad = var_3365_pad_0, pad_type = var_3365_pad_type_0, strides = var_3365_strides_0, weight = layers_8_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_3365_cast_fp16")]; + tensor var_3371_pad_type_0 = const()[name = tensor("op_3371_pad_type_0"), val = tensor("valid")]; + tensor var_3371_strides_0 = const()[name = tensor("op_3371_strides_0"), val = tensor([1, 1])]; + tensor var_3371_pad_0 = const()[name = tensor("op_3371_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3371_dilations_0 = const()[name = tensor("op_3371_dilations_0"), val = tensor([1, 1])]; + tensor var_3371_groups_0 = const()[name = tensor("op_3371_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50867840))), name = tensor("layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50853056))), shape = tensor([512, 512, 1, 1])]; + tensor var_3371_cast_fp16 = conv(dilations = var_3371_dilations_0, groups = var_3371_groups_0, pad = var_3371_pad_0, pad_type = var_3371_pad_type_0, strides = var_3371_strides_0, weight = layers_8_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_3371_cast_fp16")]; + tensor p_17_cast_fp16 = add(x = var_3365_cast_fp16, y = var_3371_cast_fp16)[name = tensor("p_17_cast_fp16")]; + tensor var_3375 = const()[name = tensor("op_3375"), val = tensor([1, 8, 64, 188])]; + tensor var_3376_cast_fp16 = reshape(shape = var_3375, x = q_with_bias_v_17_cast_fp16)[name = tensor("op_3376_cast_fp16")]; + tensor var_3377 = const()[name = tensor("op_3377"), val = tensor([1, 8, 64, -1])]; + tensor var_3378_cast_fp16 = reshape(shape = var_3377, x = p_17_cast_fp16)[name = tensor("op_3378_cast_fp16")]; + tensor matrix_bd_65_transpose_x_0 = const()[name = tensor("matrix_bd_65_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_65_transpose_y_0 = const()[name = tensor("matrix_bd_65_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_65_cast_fp16 = matmul(transpose_x = matrix_bd_65_transpose_x_0, transpose_y = matrix_bd_65_transpose_y_0, x = var_3376_cast_fp16, y = var_3378_cast_fp16)[name = tensor("matrix_bd_65_cast_fp16")]; + tensor matrix_bd_67_pad_0 = const()[name = tensor("matrix_bd_67_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_67_mode_0 = const()[name = tensor("matrix_bd_67_mode_0"), val = tensor("constant")]; + tensor const_98_to_fp16 = const()[name = tensor("const_98_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_67_cast_fp16 = pad(constant_val = const_98_to_fp16, mode = matrix_bd_67_mode_0, pad = matrix_bd_67_pad_0, x = matrix_bd_65_cast_fp16)[name = tensor("matrix_bd_67_cast_fp16")]; + tensor var_3387 = const()[name = tensor("op_3387"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_69_cast_fp16 = reshape(shape = var_3387, x = matrix_bd_67_cast_fp16)[name = tensor("matrix_bd_69_cast_fp16")]; + tensor var_3391_begin_0 = const()[name = tensor("op_3391_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3391_end_0 = const()[name = tensor("op_3391_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_3391_end_mask_0 = const()[name = tensor("op_3391_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3391_cast_fp16 = slice_by_index(begin = var_3391_begin_0, end = var_3391_end_0, end_mask = var_3391_end_mask_0, x = matrix_bd_69_cast_fp16)[name = tensor("op_3391_cast_fp16")]; + tensor var_3392 = const()[name = tensor("op_3392"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_71_cast_fp16 = reshape(shape = var_3392, x = var_3391_cast_fp16)[name = tensor("matrix_bd_71_cast_fp16")]; + tensor var_3397_begin_0 = const()[name = tensor("op_3397_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3397_end_0 = const()[name = tensor("op_3397_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3397_end_mask_0 = const()[name = tensor("op_3397_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3397_cast_fp16 = slice_by_index(begin = var_3397_begin_0, end = var_3397_end_0, end_mask = var_3397_end_mask_0, x = matrix_bd_71_cast_fp16)[name = tensor("op_3397_cast_fp16")]; + tensor var_3398_to_fp16 = const()[name = tensor("op_3398_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_17_cast_fp16 = mul(x = var_3397_cast_fp16, y = var_3398_to_fp16)[name = tensor("qk_mask_17_cast_fp16")]; + tensor var_3402 = const()[name = tensor("op_3402"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_17_cast_fp16 = reshape(shape = var_3402, x = query_35_cast_fp16)[name = tensor("mh_q_17_cast_fp16")]; + tensor var_3404_to_fp16 = const()[name = tensor("op_3404_to_fp16"), val = tensor(0x1p-3)]; + tensor var_3405_cast_fp16 = mul(x = mh_q_17_cast_fp16, y = var_3404_to_fp16)[name = tensor("op_3405_cast_fp16")]; + tensor var_3408 = const()[name = tensor("op_3408"), val = tensor([1, 8, 64, 188])]; + tensor var_3409_cast_fp16 = reshape(shape = var_3408, x = key_17_cast_fp16)[name = tensor("op_3409_cast_fp16")]; + tensor mh_w_33_transpose_x_0 = const()[name = tensor("mh_w_33_transpose_x_0"), val = tensor(true)]; + tensor mh_w_33_transpose_y_0 = const()[name = tensor("mh_w_33_transpose_y_0"), val = tensor(false)]; + tensor mh_w_33_cast_fp16 = matmul(transpose_x = mh_w_33_transpose_x_0, transpose_y = mh_w_33_transpose_y_0, x = var_3405_cast_fp16, y = var_3409_cast_fp16)[name = tensor("mh_w_33_cast_fp16")]; + tensor mh_w_35_cast_fp16 = add(x = mh_w_33_cast_fp16, y = qk_mask_17_cast_fp16)[name = tensor("mh_w_35_cast_fp16")]; + tensor var_3413_cast_fp16 = softmax(axis = var_3200, x = mh_w_35_cast_fp16)[name = tensor("op_3413_cast_fp16")]; + tensor var_3414 = const()[name = tensor("op_3414"), val = tensor([1, 8, 64, 188])]; + tensor var_3415_cast_fp16 = reshape(shape = var_3414, x = value_17_cast_fp16)[name = tensor("op_3415_cast_fp16")]; + tensor attn_17_transpose_x_0 = const()[name = tensor("attn_17_transpose_x_0"), val = tensor(false)]; + tensor attn_17_transpose_y_0 = const()[name = tensor("attn_17_transpose_y_0"), val = tensor(true)]; + tensor attn_17_cast_fp16 = matmul(transpose_x = attn_17_transpose_x_0, transpose_y = attn_17_transpose_y_0, x = var_3415_cast_fp16, y = var_3413_cast_fp16)[name = tensor("attn_17_cast_fp16")]; + tensor var_3418 = const()[name = tensor("op_3418"), val = tensor([1, 512, 1, 188])]; + tensor input_231_cast_fp16 = reshape(shape = var_3418, x = attn_17_cast_fp16)[name = tensor("input_231_cast_fp16")]; + tensor var_3428_pad_type_0 = const()[name = tensor("op_3428_pad_type_0"), val = tensor("valid")]; + tensor var_3428_strides_0 = const()[name = tensor("op_3428_strides_0"), val = tensor([1, 1])]; + tensor var_3428_pad_0 = const()[name = tensor("op_3428_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3428_dilations_0 = const()[name = tensor("op_3428_dilations_0"), val = tensor([1, 1])]; + tensor var_3428_groups_0 = const()[name = tensor("op_3428_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(50900672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51097344))), name = tensor("layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_8_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51097536)))]; + tensor var_3428_cast_fp16 = conv(bias = layers_8_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_3428_dilations_0, groups = var_3428_groups_0, pad = var_3428_pad_0, pad_type = var_3428_pad_type_0, strides = var_3428_strides_0, weight = layers_8_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_231_cast_fp16)[name = tensor("op_3428_cast_fp16")]; + tensor var_3434_pad_type_0 = const()[name = tensor("op_3434_pad_type_0"), val = tensor("valid")]; + tensor var_3434_strides_0 = const()[name = tensor("op_3434_strides_0"), val = tensor([1, 1])]; + tensor var_3434_pad_0 = const()[name = tensor("op_3434_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3434_dilations_0 = const()[name = tensor("op_3434_dilations_0"), val = tensor([1, 1])]; + tensor var_3434_groups_0 = const()[name = tensor("op_3434_groups_0"), val = tensor(1)]; + tensor layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51107392))), name = tensor("layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51098624))), shape = tensor([512, 512, 1, 1])]; + tensor var_3434_cast_fp16 = conv(dilations = var_3434_dilations_0, groups = var_3434_groups_0, pad = var_3434_pad_0, pad_type = var_3434_pad_type_0, strides = var_3434_strides_0, weight = layers_8_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_231_cast_fp16)[name = tensor("op_3434_cast_fp16")]; + tensor obj_37_cast_fp16 = add(x = var_3428_cast_fp16, y = var_3434_cast_fp16)[name = tensor("obj_37_cast_fp16")]; + tensor inputs_85_cast_fp16 = add(x = inputs_83_cast_fp16, y = obj_37_cast_fp16)[name = tensor("inputs_85_cast_fp16")]; + tensor out_85_axes_0 = const()[name = tensor("out_85_axes_0"), val = tensor([1])]; + tensor var_3445_to_fp16 = const()[name = tensor("op_3445_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_85_cast_fp16 = layer_norm(axes = out_85_axes_0, epsilon = var_3445_to_fp16, x = inputs_85_cast_fp16)[name = tensor("out_85_cast_fp16")]; + tensor input_233_gamma_0_to_fp16 = const()[name = tensor("input_233_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51140224)))]; + tensor input_233_beta_0_to_fp16 = const()[name = tensor("input_233_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51141312)))]; + tensor input_233_epsilon_0_to_fp16 = const()[name = tensor("input_233_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_233_cast_fp16 = batch_norm(beta = input_233_beta_0_to_fp16, epsilon = input_233_epsilon_0_to_fp16, gamma = input_233_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_85_cast_fp16)[name = tensor("input_233_cast_fp16")]; + tensor var_3467_pad_type_0 = const()[name = tensor("op_3467_pad_type_0"), val = tensor("valid")]; + tensor var_3467_strides_0 = const()[name = tensor("op_3467_strides_0"), val = tensor([1, 1])]; + tensor var_3467_pad_0 = const()[name = tensor("op_3467_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3467_dilations_0 = const()[name = tensor("op_3467_dilations_0"), val = tensor([1, 1])]; + tensor var_3467_groups_0 = const()[name = tensor("op_3467_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51142400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51535680))), name = tensor("layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_8_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51535872)))]; + tensor var_3467_cast_fp16 = conv(bias = layers_8_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_3467_dilations_0, groups = var_3467_groups_0, pad = var_3467_pad_0, pad_type = var_3467_pad_type_0, strides = var_3467_strides_0, weight = layers_8_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_233_cast_fp16)[name = tensor("op_3467_cast_fp16")]; + tensor var_3473_pad_type_0 = const()[name = tensor("op_3473_pad_type_0"), val = tensor("valid")]; + tensor var_3473_strides_0 = const()[name = tensor("op_3473_strides_0"), val = tensor([1, 1])]; + tensor var_3473_pad_0 = const()[name = tensor("op_3473_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3473_dilations_0 = const()[name = tensor("op_3473_dilations_0"), val = tensor([1, 1])]; + tensor var_3473_groups_0 = const()[name = tensor("op_3473_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51555520))), name = tensor("layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51537984))), shape = tensor([1024, 512, 1, 1])]; + tensor var_3473_cast_fp16 = conv(dilations = var_3473_dilations_0, groups = var_3473_groups_0, pad = var_3473_pad_0, pad_type = var_3473_pad_type_0, strides = var_3473_strides_0, weight = layers_8_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_233_cast_fp16)[name = tensor("op_3473_cast_fp16")]; + tensor input_235_cast_fp16 = add(x = var_3467_cast_fp16, y = var_3473_cast_fp16)[name = tensor("input_235_cast_fp16")]; + tensor input_237_split_num_splits_0 = const()[name = tensor("input_237_split_num_splits_0"), val = tensor(2)]; + tensor input_237_split_axis_0 = const()[name = tensor("input_237_split_axis_0"), val = tensor(1)]; + tensor input_237_split_cast_fp16_0, tensor input_237_split_cast_fp16_1 = split(axis = input_237_split_axis_0, num_splits = input_237_split_num_splits_0, x = input_235_cast_fp16)[name = tensor("input_237_split_cast_fp16")]; + tensor input_237_split_1_sigmoid_cast_fp16 = sigmoid(x = input_237_split_cast_fp16_1)[name = tensor("input_237_split_1_sigmoid_cast_fp16")]; + tensor input_237_cast_fp16 = mul(x = input_237_split_cast_fp16_0, y = input_237_split_1_sigmoid_cast_fp16)[name = tensor("input_237_cast_fp16")]; + tensor input_239_pad_type_0 = const()[name = tensor("input_239_pad_type_0"), val = tensor("custom")]; + tensor input_239_pad_0 = const()[name = tensor("input_239_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_239_groups_0 = const()[name = tensor("input_239_groups_0"), val = tensor(512)]; + tensor input_239_strides_0 = const()[name = tensor("input_239_strides_0"), val = tensor([1, 1])]; + tensor input_239_dilations_0 = const()[name = tensor("input_239_dilations_0"), val = tensor([1, 1])]; + tensor const_207_to_fp16 = const()[name = tensor("const_207_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51621120)))]; + tensor const_208_to_fp16 = const()[name = tensor("const_208_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51630400)))]; + tensor input_241_cast_fp16 = conv(bias = const_208_to_fp16, dilations = input_239_dilations_0, groups = input_239_groups_0, pad = input_239_pad_0, pad_type = input_239_pad_type_0, strides = input_239_strides_0, weight = const_207_to_fp16, x = input_237_cast_fp16)[name = tensor("input_241_cast_fp16")]; + tensor input_243_cast_fp16 = silu(x = input_241_cast_fp16)[name = tensor("input_243_cast_fp16")]; + tensor var_3497_pad_type_0 = const()[name = tensor("op_3497_pad_type_0"), val = tensor("valid")]; + tensor var_3497_strides_0 = const()[name = tensor("op_3497_strides_0"), val = tensor([1, 1])]; + tensor var_3497_pad_0 = const()[name = tensor("op_3497_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3497_dilations_0 = const()[name = tensor("op_3497_dilations_0"), val = tensor([1, 1])]; + tensor var_3497_groups_0 = const()[name = tensor("op_3497_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51631488))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51828160))), name = tensor("layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_8_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51828352)))]; + tensor var_3497_cast_fp16 = conv(bias = layers_8_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_3497_dilations_0, groups = var_3497_groups_0, pad = var_3497_pad_0, pad_type = var_3497_pad_type_0, strides = var_3497_strides_0, weight = layers_8_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_243_cast_fp16)[name = tensor("op_3497_cast_fp16")]; + tensor var_3503_pad_type_0 = const()[name = tensor("op_3503_pad_type_0"), val = tensor("valid")]; + tensor var_3503_strides_0 = const()[name = tensor("op_3503_strides_0"), val = tensor([1, 1])]; + tensor var_3503_pad_0 = const()[name = tensor("op_3503_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3503_dilations_0 = const()[name = tensor("op_3503_dilations_0"), val = tensor([1, 1])]; + tensor var_3503_groups_0 = const()[name = tensor("op_3503_groups_0"), val = tensor(1)]; + tensor layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51837312))), name = tensor("layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51829440))), shape = tensor([512, 512, 1, 1])]; + tensor var_3503_cast_fp16 = conv(dilations = var_3503_dilations_0, groups = var_3503_groups_0, pad = var_3503_pad_0, pad_type = var_3503_pad_type_0, strides = var_3503_strides_0, weight = layers_8_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_243_cast_fp16)[name = tensor("op_3503_cast_fp16")]; + tensor x_53_cast_fp16 = add(x = var_3497_cast_fp16, y = var_3503_cast_fp16)[name = tensor("x_53_cast_fp16")]; + tensor inputs_87_cast_fp16 = add(x = inputs_85_cast_fp16, y = x_53_cast_fp16)[name = tensor("inputs_87_cast_fp16")]; + tensor out_87_axes_0 = const()[name = tensor("out_87_axes_0"), val = tensor([1])]; + tensor var_3514_to_fp16 = const()[name = tensor("op_3514_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_87_cast_fp16 = layer_norm(axes = out_87_axes_0, epsilon = var_3514_to_fp16, x = inputs_87_cast_fp16)[name = tensor("out_87_cast_fp16")]; + tensor input_245_gamma_0_to_fp16 = const()[name = tensor("input_245_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51870144)))]; + tensor input_245_beta_0_to_fp16 = const()[name = tensor("input_245_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51871232)))]; + tensor input_245_epsilon_0_to_fp16 = const()[name = tensor("input_245_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_245_cast_fp16 = batch_norm(beta = input_245_beta_0_to_fp16, epsilon = input_245_epsilon_0_to_fp16, gamma = input_245_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_87_cast_fp16)[name = tensor("input_245_cast_fp16")]; + tensor var_3534_pad_type_0 = const()[name = tensor("op_3534_pad_type_0"), val = tensor("valid")]; + tensor var_3534_strides_0 = const()[name = tensor("op_3534_strides_0"), val = tensor([1, 1])]; + tensor var_3534_pad_0 = const()[name = tensor("op_3534_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3534_dilations_0 = const()[name = tensor("op_3534_dilations_0"), val = tensor([1, 1])]; + tensor var_3534_groups_0 = const()[name = tensor("op_3534_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(51872320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52658816))), name = tensor("layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_8_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52659008)))]; + tensor var_3534_cast_fp16 = conv(bias = layers_8_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_3534_dilations_0, groups = var_3534_groups_0, pad = var_3534_pad_0, pad_type = var_3534_pad_type_0, strides = var_3534_strides_0, weight = layers_8_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_245_cast_fp16)[name = tensor("op_3534_cast_fp16")]; + tensor var_3540_pad_type_0 = const()[name = tensor("op_3540_pad_type_0"), val = tensor("valid")]; + tensor var_3540_strides_0 = const()[name = tensor("op_3540_strides_0"), val = tensor([1, 1])]; + tensor var_3540_pad_0 = const()[name = tensor("op_3540_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3540_dilations_0 = const()[name = tensor("op_3540_dilations_0"), val = tensor([1, 1])]; + tensor var_3540_groups_0 = const()[name = tensor("op_3540_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52694080))), name = tensor("layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52663168))), shape = tensor([2048, 512, 1, 1])]; + tensor var_3540_cast_fp16 = conv(dilations = var_3540_dilations_0, groups = var_3540_groups_0, pad = var_3540_pad_0, pad_type = var_3540_pad_type_0, strides = var_3540_strides_0, weight = layers_8_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_245_cast_fp16)[name = tensor("op_3540_cast_fp16")]; + tensor input_247_cast_fp16 = add(x = var_3534_cast_fp16, y = var_3540_cast_fp16)[name = tensor("input_247_cast_fp16")]; + tensor input_249_cast_fp16 = silu(x = input_247_cast_fp16)[name = tensor("input_249_cast_fp16")]; + tensor var_3551_pad_type_0 = const()[name = tensor("op_3551_pad_type_0"), val = tensor("valid")]; + tensor var_3551_strides_0 = const()[name = tensor("op_3551_strides_0"), val = tensor([1, 1])]; + tensor var_3551_pad_0 = const()[name = tensor("op_3551_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3551_dilations_0 = const()[name = tensor("op_3551_dilations_0"), val = tensor([1, 1])]; + tensor var_3551_groups_0 = const()[name = tensor("op_3551_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(52825216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53611712))), name = tensor("layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_8_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_8_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53611904)))]; + tensor var_3551_cast_fp16 = conv(bias = layers_8_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_3551_dilations_0, groups = var_3551_groups_0, pad = var_3551_pad_0, pad_type = var_3551_pad_type_0, strides = var_3551_strides_0, weight = layers_8_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_249_cast_fp16)[name = tensor("op_3551_cast_fp16")]; + tensor var_3557_pad_type_0 = const()[name = tensor("op_3557_pad_type_0"), val = tensor("valid")]; + tensor var_3557_strides_0 = const()[name = tensor("op_3557_strides_0"), val = tensor([1, 1])]; + tensor var_3557_pad_0 = const()[name = tensor("op_3557_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3557_dilations_0 = const()[name = tensor("op_3557_dilations_0"), val = tensor([1, 1])]; + tensor var_3557_groups_0 = const()[name = tensor("op_3557_groups_0"), val = tensor(1)]; + tensor layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53646336))), name = tensor("layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53612992))), shape = tensor([512, 2048, 1, 1])]; + tensor var_3557_cast_fp16 = conv(dilations = var_3557_dilations_0, groups = var_3557_groups_0, pad = var_3557_pad_0, pad_type = var_3557_pad_type_0, strides = var_3557_strides_0, weight = layers_8_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_249_cast_fp16)[name = tensor("op_3557_cast_fp16")]; + tensor x_55_cast_fp16 = add(x = var_3551_cast_fp16, y = var_3557_cast_fp16)[name = tensor("x_55_cast_fp16")]; + tensor var_3559_to_fp16 = const()[name = tensor("op_3559_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3560_cast_fp16 = mul(x = x_55_cast_fp16, y = var_3559_to_fp16)[name = tensor("op_3560_cast_fp16")]; + tensor inputs_89_cast_fp16 = add(x = inputs_87_cast_fp16, y = var_3560_cast_fp16)[name = tensor("inputs_89_cast_fp16")]; + tensor out_89_axes_0 = const()[name = tensor("out_89_axes_0"), val = tensor([1])]; + tensor var_3570_to_fp16 = const()[name = tensor("op_3570_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_89_cast_fp16 = layer_norm(axes = out_89_axes_0, epsilon = var_3570_to_fp16, x = inputs_89_cast_fp16)[name = tensor("out_89_cast_fp16")]; + tensor inputs_91_gamma_0_to_fp16 = const()[name = tensor("inputs_91_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53777472)))]; + tensor inputs_91_beta_0_to_fp16 = const()[name = tensor("inputs_91_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53778560)))]; + tensor inputs_91_epsilon_0_to_fp16 = const()[name = tensor("inputs_91_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_91_cast_fp16 = batch_norm(beta = inputs_91_beta_0_to_fp16, epsilon = inputs_91_epsilon_0_to_fp16, gamma = inputs_91_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_89_cast_fp16)[name = tensor("inputs_91_cast_fp16")]; + tensor var_3584 = const()[name = tensor("op_3584"), val = tensor(3)]; + tensor out_91_axes_0 = const()[name = tensor("out_91_axes_0"), val = tensor([1])]; + tensor var_3615_to_fp16 = const()[name = tensor("op_3615_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_91_cast_fp16 = layer_norm(axes = out_91_axes_0, epsilon = var_3615_to_fp16, x = inputs_91_cast_fp16)[name = tensor("out_91_cast_fp16")]; + tensor input_251_gamma_0_to_fp16 = const()[name = tensor("input_251_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53779648)))]; + tensor input_251_beta_0_to_fp16 = const()[name = tensor("input_251_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53780736)))]; + tensor input_251_epsilon_0_to_fp16 = const()[name = tensor("input_251_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_251_cast_fp16 = batch_norm(beta = input_251_beta_0_to_fp16, epsilon = input_251_epsilon_0_to_fp16, gamma = input_251_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_91_cast_fp16)[name = tensor("input_251_cast_fp16")]; + tensor var_3635_pad_type_0 = const()[name = tensor("op_3635_pad_type_0"), val = tensor("valid")]; + tensor var_3635_strides_0 = const()[name = tensor("op_3635_strides_0"), val = tensor([1, 1])]; + tensor var_3635_pad_0 = const()[name = tensor("op_3635_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3635_dilations_0 = const()[name = tensor("op_3635_dilations_0"), val = tensor([1, 1])]; + tensor var_3635_groups_0 = const()[name = tensor("op_3635_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(53781824))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54568320))), name = tensor("layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_9_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54568512)))]; + tensor var_3635_cast_fp16 = conv(bias = layers_9_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_3635_dilations_0, groups = var_3635_groups_0, pad = var_3635_pad_0, pad_type = var_3635_pad_type_0, strides = var_3635_strides_0, weight = layers_9_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_251_cast_fp16)[name = tensor("op_3635_cast_fp16")]; + tensor var_3641_pad_type_0 = const()[name = tensor("op_3641_pad_type_0"), val = tensor("valid")]; + tensor var_3641_strides_0 = const()[name = tensor("op_3641_strides_0"), val = tensor([1, 1])]; + tensor var_3641_pad_0 = const()[name = tensor("op_3641_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3641_dilations_0 = const()[name = tensor("op_3641_dilations_0"), val = tensor([1, 1])]; + tensor var_3641_groups_0 = const()[name = tensor("op_3641_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54604800))), name = tensor("layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54572672))), shape = tensor([2048, 512, 1, 1])]; + tensor var_3641_cast_fp16 = conv(dilations = var_3641_dilations_0, groups = var_3641_groups_0, pad = var_3641_pad_0, pad_type = var_3641_pad_type_0, strides = var_3641_strides_0, weight = layers_9_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_251_cast_fp16)[name = tensor("op_3641_cast_fp16")]; + tensor input_253_cast_fp16 = add(x = var_3635_cast_fp16, y = var_3641_cast_fp16)[name = tensor("input_253_cast_fp16")]; + tensor input_255_cast_fp16 = silu(x = input_253_cast_fp16)[name = tensor("input_255_cast_fp16")]; + tensor var_3652_pad_type_0 = const()[name = tensor("op_3652_pad_type_0"), val = tensor("valid")]; + tensor var_3652_strides_0 = const()[name = tensor("op_3652_strides_0"), val = tensor([1, 1])]; + tensor var_3652_pad_0 = const()[name = tensor("op_3652_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3652_dilations_0 = const()[name = tensor("op_3652_dilations_0"), val = tensor([1, 1])]; + tensor var_3652_groups_0 = const()[name = tensor("op_3652_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(54735936))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55522432))), name = tensor("layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_9_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55522624)))]; + tensor var_3652_cast_fp16 = conv(bias = layers_9_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_3652_dilations_0, groups = var_3652_groups_0, pad = var_3652_pad_0, pad_type = var_3652_pad_type_0, strides = var_3652_strides_0, weight = layers_9_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_255_cast_fp16)[name = tensor("op_3652_cast_fp16")]; + tensor var_3658_pad_type_0 = const()[name = tensor("op_3658_pad_type_0"), val = tensor("valid")]; + tensor var_3658_strides_0 = const()[name = tensor("op_3658_strides_0"), val = tensor([1, 1])]; + tensor var_3658_pad_0 = const()[name = tensor("op_3658_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3658_dilations_0 = const()[name = tensor("op_3658_dilations_0"), val = tensor([1, 1])]; + tensor var_3658_groups_0 = const()[name = tensor("op_3658_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55559424))), name = tensor("layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55523712))), shape = tensor([512, 2048, 1, 1])]; + tensor var_3658_cast_fp16 = conv(dilations = var_3658_dilations_0, groups = var_3658_groups_0, pad = var_3658_pad_0, pad_type = var_3658_pad_type_0, strides = var_3658_strides_0, weight = layers_9_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_255_cast_fp16)[name = tensor("op_3658_cast_fp16")]; + tensor x_57_cast_fp16 = add(x = var_3652_cast_fp16, y = var_3658_cast_fp16)[name = tensor("x_57_cast_fp16")]; + tensor var_3660_to_fp16 = const()[name = tensor("op_3660_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3661_cast_fp16 = mul(x = x_57_cast_fp16, y = var_3660_to_fp16)[name = tensor("op_3661_cast_fp16")]; + tensor inputs_93_cast_fp16 = add(x = inputs_91_cast_fp16, y = var_3661_cast_fp16)[name = tensor("inputs_93_cast_fp16")]; + tensor out_93_axes_0 = const()[name = tensor("out_93_axes_0"), val = tensor([1])]; + tensor var_3671_to_fp16 = const()[name = tensor("op_3671_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_93_cast_fp16 = layer_norm(axes = out_93_axes_0, epsilon = var_3671_to_fp16, x = inputs_93_cast_fp16)[name = tensor("out_93_cast_fp16")]; + tensor obj_39_gamma_0_to_fp16 = const()[name = tensor("obj_39_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55690560)))]; + tensor obj_39_beta_0_to_fp16 = const()[name = tensor("obj_39_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55691648)))]; + tensor obj_39_epsilon_0_to_fp16 = const()[name = tensor("obj_39_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_39_cast_fp16 = batch_norm(beta = obj_39_beta_0_to_fp16, epsilon = obj_39_epsilon_0_to_fp16, gamma = obj_39_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_93_cast_fp16)[name = tensor("obj_39_cast_fp16")]; + tensor var_3696_pad_type_0 = const()[name = tensor("op_3696_pad_type_0"), val = tensor("valid")]; + tensor var_3696_strides_0 = const()[name = tensor("op_3696_strides_0"), val = tensor([1, 1])]; + tensor var_3696_pad_0 = const()[name = tensor("op_3696_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3696_dilations_0 = const()[name = tensor("op_3696_dilations_0"), val = tensor([1, 1])]; + tensor var_3696_groups_0 = const()[name = tensor("op_3696_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55692736))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55889408))), name = tensor("layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_9_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55889600)))]; + tensor var_3696_cast_fp16 = conv(bias = layers_9_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_3696_dilations_0, groups = var_3696_groups_0, pad = var_3696_pad_0, pad_type = var_3696_pad_type_0, strides = var_3696_strides_0, weight = layers_9_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3696_cast_fp16")]; + tensor var_3702_pad_type_0 = const()[name = tensor("op_3702_pad_type_0"), val = tensor("valid")]; + tensor var_3702_strides_0 = const()[name = tensor("op_3702_strides_0"), val = tensor([1, 1])]; + tensor var_3702_pad_0 = const()[name = tensor("op_3702_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3702_dilations_0 = const()[name = tensor("op_3702_dilations_0"), val = tensor([1, 1])]; + tensor var_3702_groups_0 = const()[name = tensor("op_3702_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55900672))), name = tensor("layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55890688))), shape = tensor([512, 512, 1, 1])]; + tensor var_3702_cast_fp16 = conv(dilations = var_3702_dilations_0, groups = var_3702_groups_0, pad = var_3702_pad_0, pad_type = var_3702_pad_type_0, strides = var_3702_strides_0, weight = layers_9_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3702_cast_fp16")]; + tensor query_37_cast_fp16 = add(x = var_3696_cast_fp16, y = var_3702_cast_fp16)[name = tensor("query_37_cast_fp16")]; + tensor var_3711_pad_type_0 = const()[name = tensor("op_3711_pad_type_0"), val = tensor("valid")]; + tensor var_3711_strides_0 = const()[name = tensor("op_3711_strides_0"), val = tensor([1, 1])]; + tensor var_3711_pad_0 = const()[name = tensor("op_3711_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3711_dilations_0 = const()[name = tensor("op_3711_dilations_0"), val = tensor([1, 1])]; + tensor var_3711_groups_0 = const()[name = tensor("op_3711_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(55933504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56130176))), name = tensor("layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_3711_cast_fp16 = conv(dilations = var_3711_dilations_0, groups = var_3711_groups_0, pad = var_3711_pad_0, pad_type = var_3711_pad_type_0, strides = var_3711_strides_0, weight = layers_9_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3711_cast_fp16")]; + tensor var_3717_pad_type_0 = const()[name = tensor("op_3717_pad_type_0"), val = tensor("valid")]; + tensor var_3717_strides_0 = const()[name = tensor("op_3717_strides_0"), val = tensor([1, 1])]; + tensor var_3717_pad_0 = const()[name = tensor("op_3717_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3717_dilations_0 = const()[name = tensor("op_3717_dilations_0"), val = tensor([1, 1])]; + tensor var_3717_groups_0 = const()[name = tensor("op_3717_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56138432))), name = tensor("layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56130368))), shape = tensor([512, 512, 1, 1])]; + tensor var_3717_cast_fp16 = conv(dilations = var_3717_dilations_0, groups = var_3717_groups_0, pad = var_3717_pad_0, pad_type = var_3717_pad_type_0, strides = var_3717_strides_0, weight = layers_9_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3717_cast_fp16")]; + tensor key_19_cast_fp16 = add(x = var_3711_cast_fp16, y = var_3717_cast_fp16)[name = tensor("key_19_cast_fp16")]; + tensor var_3727_pad_type_0 = const()[name = tensor("op_3727_pad_type_0"), val = tensor("valid")]; + tensor var_3727_strides_0 = const()[name = tensor("op_3727_strides_0"), val = tensor([1, 1])]; + tensor var_3727_pad_0 = const()[name = tensor("op_3727_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3727_dilations_0 = const()[name = tensor("op_3727_dilations_0"), val = tensor([1, 1])]; + tensor var_3727_groups_0 = const()[name = tensor("op_3727_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56171264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56367936))), name = tensor("layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_9_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56368128)))]; + tensor var_3727_cast_fp16 = conv(bias = layers_9_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_3727_dilations_0, groups = var_3727_groups_0, pad = var_3727_pad_0, pad_type = var_3727_pad_type_0, strides = var_3727_strides_0, weight = layers_9_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_39_cast_fp16)[name = tensor("op_3727_cast_fp16")]; + tensor var_3733_pad_type_0 = const()[name = tensor("op_3733_pad_type_0"), val = tensor("valid")]; + tensor var_3733_strides_0 = const()[name = tensor("op_3733_strides_0"), val = tensor([1, 1])]; + tensor var_3733_pad_0 = const()[name = tensor("op_3733_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3733_dilations_0 = const()[name = tensor("op_3733_dilations_0"), val = tensor([1, 1])]; + tensor var_3733_groups_0 = const()[name = tensor("op_3733_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56376768))), name = tensor("layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56369216))), shape = tensor([512, 512, 1, 1])]; + tensor var_3733_cast_fp16 = conv(dilations = var_3733_dilations_0, groups = var_3733_groups_0, pad = var_3733_pad_0, pad_type = var_3733_pad_type_0, strides = var_3733_strides_0, weight = layers_9_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_39_cast_fp16)[name = tensor("op_3733_cast_fp16")]; + tensor value_19_cast_fp16 = add(x = var_3727_cast_fp16, y = var_3733_cast_fp16)[name = tensor("value_19_cast_fp16")]; + tensor var_3736_to_fp16 = const()[name = tensor("op_3736_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56409600)))]; + tensor query_39_cast_fp16 = add(x = query_37_cast_fp16, y = var_3736_to_fp16)[name = tensor("query_39_cast_fp16")]; + tensor var_3739_to_fp16 = const()[name = tensor("op_3739_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56410688)))]; + tensor q_with_bias_v_19_cast_fp16 = add(x = query_37_cast_fp16, y = var_3739_to_fp16)[name = tensor("q_with_bias_v_19_cast_fp16")]; + tensor var_3749_pad_type_0 = const()[name = tensor("op_3749_pad_type_0"), val = tensor("valid")]; + tensor var_3749_strides_0 = const()[name = tensor("op_3749_strides_0"), val = tensor([1, 1])]; + tensor var_3749_pad_0 = const()[name = tensor("op_3749_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3749_dilations_0 = const()[name = tensor("op_3749_dilations_0"), val = tensor([1, 1])]; + tensor var_3749_groups_0 = const()[name = tensor("op_3749_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56411776))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56608448))), name = tensor("layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_3749_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_3749_dilations_0, groups = var_3749_groups_0, pad = var_3749_pad_0, pad_type = var_3749_pad_type_0, strides = var_3749_strides_0, weight = layers_9_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_3749_cast_fp16")]; + tensor var_3755_pad_type_0 = const()[name = tensor("op_3755_pad_type_0"), val = tensor("valid")]; + tensor var_3755_strides_0 = const()[name = tensor("op_3755_strides_0"), val = tensor([1, 1])]; + tensor var_3755_pad_0 = const()[name = tensor("op_3755_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3755_dilations_0 = const()[name = tensor("op_3755_dilations_0"), val = tensor([1, 1])]; + tensor var_3755_groups_0 = const()[name = tensor("op_3755_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56624256))), name = tensor("layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56608640))), shape = tensor([512, 512, 1, 1])]; + tensor var_3755_cast_fp16 = conv(dilations = var_3755_dilations_0, groups = var_3755_groups_0, pad = var_3755_pad_0, pad_type = var_3755_pad_type_0, strides = var_3755_strides_0, weight = layers_9_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_3755_cast_fp16")]; + tensor p_19_cast_fp16 = add(x = var_3749_cast_fp16, y = var_3755_cast_fp16)[name = tensor("p_19_cast_fp16")]; + tensor var_3759 = const()[name = tensor("op_3759"), val = tensor([1, 8, 64, 188])]; + tensor var_3760_cast_fp16 = reshape(shape = var_3759, x = q_with_bias_v_19_cast_fp16)[name = tensor("op_3760_cast_fp16")]; + tensor var_3761 = const()[name = tensor("op_3761"), val = tensor([1, 8, 64, -1])]; + tensor var_3762_cast_fp16 = reshape(shape = var_3761, x = p_19_cast_fp16)[name = tensor("op_3762_cast_fp16")]; + tensor matrix_bd_73_transpose_x_0 = const()[name = tensor("matrix_bd_73_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_73_transpose_y_0 = const()[name = tensor("matrix_bd_73_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_73_cast_fp16 = matmul(transpose_x = matrix_bd_73_transpose_x_0, transpose_y = matrix_bd_73_transpose_y_0, x = var_3760_cast_fp16, y = var_3762_cast_fp16)[name = tensor("matrix_bd_73_cast_fp16")]; + tensor matrix_bd_75_pad_0 = const()[name = tensor("matrix_bd_75_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_75_mode_0 = const()[name = tensor("matrix_bd_75_mode_0"), val = tensor("constant")]; + tensor const_109_to_fp16 = const()[name = tensor("const_109_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_75_cast_fp16 = pad(constant_val = const_109_to_fp16, mode = matrix_bd_75_mode_0, pad = matrix_bd_75_pad_0, x = matrix_bd_73_cast_fp16)[name = tensor("matrix_bd_75_cast_fp16")]; + tensor var_3771 = const()[name = tensor("op_3771"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_77_cast_fp16 = reshape(shape = var_3771, x = matrix_bd_75_cast_fp16)[name = tensor("matrix_bd_77_cast_fp16")]; + tensor var_3775_begin_0 = const()[name = tensor("op_3775_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_3775_end_0 = const()[name = tensor("op_3775_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_3775_end_mask_0 = const()[name = tensor("op_3775_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_3775_cast_fp16 = slice_by_index(begin = var_3775_begin_0, end = var_3775_end_0, end_mask = var_3775_end_mask_0, x = matrix_bd_77_cast_fp16)[name = tensor("op_3775_cast_fp16")]; + tensor var_3776 = const()[name = tensor("op_3776"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_79_cast_fp16 = reshape(shape = var_3776, x = var_3775_cast_fp16)[name = tensor("matrix_bd_79_cast_fp16")]; + tensor var_3781_begin_0 = const()[name = tensor("op_3781_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3781_end_0 = const()[name = tensor("op_3781_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_3781_end_mask_0 = const()[name = tensor("op_3781_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_3781_cast_fp16 = slice_by_index(begin = var_3781_begin_0, end = var_3781_end_0, end_mask = var_3781_end_mask_0, x = matrix_bd_79_cast_fp16)[name = tensor("op_3781_cast_fp16")]; + tensor var_3782_to_fp16 = const()[name = tensor("op_3782_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_19_cast_fp16 = mul(x = var_3781_cast_fp16, y = var_3782_to_fp16)[name = tensor("qk_mask_19_cast_fp16")]; + tensor var_3786 = const()[name = tensor("op_3786"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_19_cast_fp16 = reshape(shape = var_3786, x = query_39_cast_fp16)[name = tensor("mh_q_19_cast_fp16")]; + tensor var_3788_to_fp16 = const()[name = tensor("op_3788_to_fp16"), val = tensor(0x1p-3)]; + tensor var_3789_cast_fp16 = mul(x = mh_q_19_cast_fp16, y = var_3788_to_fp16)[name = tensor("op_3789_cast_fp16")]; + tensor var_3792 = const()[name = tensor("op_3792"), val = tensor([1, 8, 64, 188])]; + tensor var_3793_cast_fp16 = reshape(shape = var_3792, x = key_19_cast_fp16)[name = tensor("op_3793_cast_fp16")]; + tensor mh_w_37_transpose_x_0 = const()[name = tensor("mh_w_37_transpose_x_0"), val = tensor(true)]; + tensor mh_w_37_transpose_y_0 = const()[name = tensor("mh_w_37_transpose_y_0"), val = tensor(false)]; + tensor mh_w_37_cast_fp16 = matmul(transpose_x = mh_w_37_transpose_x_0, transpose_y = mh_w_37_transpose_y_0, x = var_3789_cast_fp16, y = var_3793_cast_fp16)[name = tensor("mh_w_37_cast_fp16")]; + tensor mh_w_39_cast_fp16 = add(x = mh_w_37_cast_fp16, y = qk_mask_19_cast_fp16)[name = tensor("mh_w_39_cast_fp16")]; + tensor var_3797_cast_fp16 = softmax(axis = var_3584, x = mh_w_39_cast_fp16)[name = tensor("op_3797_cast_fp16")]; + tensor var_3798 = const()[name = tensor("op_3798"), val = tensor([1, 8, 64, 188])]; + tensor var_3799_cast_fp16 = reshape(shape = var_3798, x = value_19_cast_fp16)[name = tensor("op_3799_cast_fp16")]; + tensor attn_19_transpose_x_0 = const()[name = tensor("attn_19_transpose_x_0"), val = tensor(false)]; + tensor attn_19_transpose_y_0 = const()[name = tensor("attn_19_transpose_y_0"), val = tensor(true)]; + tensor attn_19_cast_fp16 = matmul(transpose_x = attn_19_transpose_x_0, transpose_y = attn_19_transpose_y_0, x = var_3799_cast_fp16, y = var_3797_cast_fp16)[name = tensor("attn_19_cast_fp16")]; + tensor var_3802 = const()[name = tensor("op_3802"), val = tensor([1, 512, 1, 188])]; + tensor input_257_cast_fp16 = reshape(shape = var_3802, x = attn_19_cast_fp16)[name = tensor("input_257_cast_fp16")]; + tensor var_3812_pad_type_0 = const()[name = tensor("op_3812_pad_type_0"), val = tensor("valid")]; + tensor var_3812_strides_0 = const()[name = tensor("op_3812_strides_0"), val = tensor([1, 1])]; + tensor var_3812_pad_0 = const()[name = tensor("op_3812_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3812_dilations_0 = const()[name = tensor("op_3812_dilations_0"), val = tensor([1, 1])]; + tensor var_3812_groups_0 = const()[name = tensor("op_3812_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56657088))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56853760))), name = tensor("layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_9_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56853952)))]; + tensor var_3812_cast_fp16 = conv(bias = layers_9_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_3812_dilations_0, groups = var_3812_groups_0, pad = var_3812_pad_0, pad_type = var_3812_pad_type_0, strides = var_3812_strides_0, weight = layers_9_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_257_cast_fp16)[name = tensor("op_3812_cast_fp16")]; + tensor var_3818_pad_type_0 = const()[name = tensor("op_3818_pad_type_0"), val = tensor("valid")]; + tensor var_3818_strides_0 = const()[name = tensor("op_3818_strides_0"), val = tensor([1, 1])]; + tensor var_3818_pad_0 = const()[name = tensor("op_3818_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3818_dilations_0 = const()[name = tensor("op_3818_dilations_0"), val = tensor([1, 1])]; + tensor var_3818_groups_0 = const()[name = tensor("op_3818_groups_0"), val = tensor(1)]; + tensor layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56863616))), name = tensor("layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56855040))), shape = tensor([512, 512, 1, 1])]; + tensor var_3818_cast_fp16 = conv(dilations = var_3818_dilations_0, groups = var_3818_groups_0, pad = var_3818_pad_0, pad_type = var_3818_pad_type_0, strides = var_3818_strides_0, weight = layers_9_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_257_cast_fp16)[name = tensor("op_3818_cast_fp16")]; + tensor obj_41_cast_fp16 = add(x = var_3812_cast_fp16, y = var_3818_cast_fp16)[name = tensor("obj_41_cast_fp16")]; + tensor inputs_95_cast_fp16 = add(x = inputs_93_cast_fp16, y = obj_41_cast_fp16)[name = tensor("inputs_95_cast_fp16")]; + tensor out_95_axes_0 = const()[name = tensor("out_95_axes_0"), val = tensor([1])]; + tensor var_3829_to_fp16 = const()[name = tensor("op_3829_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_95_cast_fp16 = layer_norm(axes = out_95_axes_0, epsilon = var_3829_to_fp16, x = inputs_95_cast_fp16)[name = tensor("out_95_cast_fp16")]; + tensor input_259_gamma_0_to_fp16 = const()[name = tensor("input_259_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56896448)))]; + tensor input_259_beta_0_to_fp16 = const()[name = tensor("input_259_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56897536)))]; + tensor input_259_epsilon_0_to_fp16 = const()[name = tensor("input_259_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_259_cast_fp16 = batch_norm(beta = input_259_beta_0_to_fp16, epsilon = input_259_epsilon_0_to_fp16, gamma = input_259_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_95_cast_fp16)[name = tensor("input_259_cast_fp16")]; + tensor var_3851_pad_type_0 = const()[name = tensor("op_3851_pad_type_0"), val = tensor("valid")]; + tensor var_3851_strides_0 = const()[name = tensor("op_3851_strides_0"), val = tensor([1, 1])]; + tensor var_3851_pad_0 = const()[name = tensor("op_3851_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3851_dilations_0 = const()[name = tensor("op_3851_dilations_0"), val = tensor([1, 1])]; + tensor var_3851_groups_0 = const()[name = tensor("op_3851_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(56898624))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57291904))), name = tensor("layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_9_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57292096)))]; + tensor var_3851_cast_fp16 = conv(bias = layers_9_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_3851_dilations_0, groups = var_3851_groups_0, pad = var_3851_pad_0, pad_type = var_3851_pad_type_0, strides = var_3851_strides_0, weight = layers_9_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_259_cast_fp16)[name = tensor("op_3851_cast_fp16")]; + tensor var_3857_pad_type_0 = const()[name = tensor("op_3857_pad_type_0"), val = tensor("valid")]; + tensor var_3857_strides_0 = const()[name = tensor("op_3857_strides_0"), val = tensor([1, 1])]; + tensor var_3857_pad_0 = const()[name = tensor("op_3857_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3857_dilations_0 = const()[name = tensor("op_3857_dilations_0"), val = tensor([1, 1])]; + tensor var_3857_groups_0 = const()[name = tensor("op_3857_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57311296))), name = tensor("layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57294208))), shape = tensor([1024, 512, 1, 1])]; + tensor var_3857_cast_fp16 = conv(dilations = var_3857_dilations_0, groups = var_3857_groups_0, pad = var_3857_pad_0, pad_type = var_3857_pad_type_0, strides = var_3857_strides_0, weight = layers_9_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_259_cast_fp16)[name = tensor("op_3857_cast_fp16")]; + tensor input_261_cast_fp16 = add(x = var_3851_cast_fp16, y = var_3857_cast_fp16)[name = tensor("input_261_cast_fp16")]; + tensor input_263_split_num_splits_0 = const()[name = tensor("input_263_split_num_splits_0"), val = tensor(2)]; + tensor input_263_split_axis_0 = const()[name = tensor("input_263_split_axis_0"), val = tensor(1)]; + tensor input_263_split_cast_fp16_0, tensor input_263_split_cast_fp16_1 = split(axis = input_263_split_axis_0, num_splits = input_263_split_num_splits_0, x = input_261_cast_fp16)[name = tensor("input_263_split_cast_fp16")]; + tensor input_263_split_1_sigmoid_cast_fp16 = sigmoid(x = input_263_split_cast_fp16_1)[name = tensor("input_263_split_1_sigmoid_cast_fp16")]; + tensor input_263_cast_fp16 = mul(x = input_263_split_cast_fp16_0, y = input_263_split_1_sigmoid_cast_fp16)[name = tensor("input_263_cast_fp16")]; + tensor input_265_pad_type_0 = const()[name = tensor("input_265_pad_type_0"), val = tensor("custom")]; + tensor input_265_pad_0 = const()[name = tensor("input_265_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_265_groups_0 = const()[name = tensor("input_265_groups_0"), val = tensor(512)]; + tensor input_265_strides_0 = const()[name = tensor("input_265_strides_0"), val = tensor([1, 1])]; + tensor input_265_dilations_0 = const()[name = tensor("input_265_dilations_0"), val = tensor([1, 1])]; + tensor const_209_to_fp16 = const()[name = tensor("const_209_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57376896)))]; + tensor const_210_to_fp16 = const()[name = tensor("const_210_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57386176)))]; + tensor input_267_cast_fp16 = conv(bias = const_210_to_fp16, dilations = input_265_dilations_0, groups = input_265_groups_0, pad = input_265_pad_0, pad_type = input_265_pad_type_0, strides = input_265_strides_0, weight = const_209_to_fp16, x = input_263_cast_fp16)[name = tensor("input_267_cast_fp16")]; + tensor input_269_cast_fp16 = silu(x = input_267_cast_fp16)[name = tensor("input_269_cast_fp16")]; + tensor var_3881_pad_type_0 = const()[name = tensor("op_3881_pad_type_0"), val = tensor("valid")]; + tensor var_3881_strides_0 = const()[name = tensor("op_3881_strides_0"), val = tensor([1, 1])]; + tensor var_3881_pad_0 = const()[name = tensor("op_3881_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3881_dilations_0 = const()[name = tensor("op_3881_dilations_0"), val = tensor([1, 1])]; + tensor var_3881_groups_0 = const()[name = tensor("op_3881_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57387264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57583936))), name = tensor("layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_9_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57584128)))]; + tensor var_3881_cast_fp16 = conv(bias = layers_9_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_3881_dilations_0, groups = var_3881_groups_0, pad = var_3881_pad_0, pad_type = var_3881_pad_type_0, strides = var_3881_strides_0, weight = layers_9_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_269_cast_fp16)[name = tensor("op_3881_cast_fp16")]; + tensor var_3887_pad_type_0 = const()[name = tensor("op_3887_pad_type_0"), val = tensor("valid")]; + tensor var_3887_strides_0 = const()[name = tensor("op_3887_strides_0"), val = tensor([1, 1])]; + tensor var_3887_pad_0 = const()[name = tensor("op_3887_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3887_dilations_0 = const()[name = tensor("op_3887_dilations_0"), val = tensor([1, 1])]; + tensor var_3887_groups_0 = const()[name = tensor("op_3887_groups_0"), val = tensor(1)]; + tensor layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57593408))), name = tensor("layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57585216))), shape = tensor([512, 512, 1, 1])]; + tensor var_3887_cast_fp16 = conv(dilations = var_3887_dilations_0, groups = var_3887_groups_0, pad = var_3887_pad_0, pad_type = var_3887_pad_type_0, strides = var_3887_strides_0, weight = layers_9_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_269_cast_fp16)[name = tensor("op_3887_cast_fp16")]; + tensor x_59_cast_fp16 = add(x = var_3881_cast_fp16, y = var_3887_cast_fp16)[name = tensor("x_59_cast_fp16")]; + tensor inputs_97_cast_fp16 = add(x = inputs_95_cast_fp16, y = x_59_cast_fp16)[name = tensor("inputs_97_cast_fp16")]; + tensor out_97_axes_0 = const()[name = tensor("out_97_axes_0"), val = tensor([1])]; + tensor var_3898_to_fp16 = const()[name = tensor("op_3898_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_97_cast_fp16 = layer_norm(axes = out_97_axes_0, epsilon = var_3898_to_fp16, x = inputs_97_cast_fp16)[name = tensor("out_97_cast_fp16")]; + tensor input_271_gamma_0_to_fp16 = const()[name = tensor("input_271_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57626240)))]; + tensor input_271_beta_0_to_fp16 = const()[name = tensor("input_271_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57627328)))]; + tensor input_271_epsilon_0_to_fp16 = const()[name = tensor("input_271_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_271_cast_fp16 = batch_norm(beta = input_271_beta_0_to_fp16, epsilon = input_271_epsilon_0_to_fp16, gamma = input_271_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_97_cast_fp16)[name = tensor("input_271_cast_fp16")]; + tensor var_3918_pad_type_0 = const()[name = tensor("op_3918_pad_type_0"), val = tensor("valid")]; + tensor var_3918_strides_0 = const()[name = tensor("op_3918_strides_0"), val = tensor([1, 1])]; + tensor var_3918_pad_0 = const()[name = tensor("op_3918_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3918_dilations_0 = const()[name = tensor("op_3918_dilations_0"), val = tensor([1, 1])]; + tensor var_3918_groups_0 = const()[name = tensor("op_3918_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(57628416))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58414912))), name = tensor("layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_9_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58415104)))]; + tensor var_3918_cast_fp16 = conv(bias = layers_9_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_3918_dilations_0, groups = var_3918_groups_0, pad = var_3918_pad_0, pad_type = var_3918_pad_type_0, strides = var_3918_strides_0, weight = layers_9_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_271_cast_fp16)[name = tensor("op_3918_cast_fp16")]; + tensor var_3924_pad_type_0 = const()[name = tensor("op_3924_pad_type_0"), val = tensor("valid")]; + tensor var_3924_strides_0 = const()[name = tensor("op_3924_strides_0"), val = tensor([1, 1])]; + tensor var_3924_pad_0 = const()[name = tensor("op_3924_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3924_dilations_0 = const()[name = tensor("op_3924_dilations_0"), val = tensor([1, 1])]; + tensor var_3924_groups_0 = const()[name = tensor("op_3924_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58451456))), name = tensor("layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58419264))), shape = tensor([2048, 512, 1, 1])]; + tensor var_3924_cast_fp16 = conv(dilations = var_3924_dilations_0, groups = var_3924_groups_0, pad = var_3924_pad_0, pad_type = var_3924_pad_type_0, strides = var_3924_strides_0, weight = layers_9_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_271_cast_fp16)[name = tensor("op_3924_cast_fp16")]; + tensor input_273_cast_fp16 = add(x = var_3918_cast_fp16, y = var_3924_cast_fp16)[name = tensor("input_273_cast_fp16")]; + tensor input_275_cast_fp16 = silu(x = input_273_cast_fp16)[name = tensor("input_275_cast_fp16")]; + tensor var_3935_pad_type_0 = const()[name = tensor("op_3935_pad_type_0"), val = tensor("valid")]; + tensor var_3935_strides_0 = const()[name = tensor("op_3935_strides_0"), val = tensor([1, 1])]; + tensor var_3935_pad_0 = const()[name = tensor("op_3935_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3935_dilations_0 = const()[name = tensor("op_3935_dilations_0"), val = tensor([1, 1])]; + tensor var_3935_groups_0 = const()[name = tensor("op_3935_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(58582592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59369088))), name = tensor("layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_9_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_9_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59369280)))]; + tensor var_3935_cast_fp16 = conv(bias = layers_9_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_3935_dilations_0, groups = var_3935_groups_0, pad = var_3935_pad_0, pad_type = var_3935_pad_type_0, strides = var_3935_strides_0, weight = layers_9_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_275_cast_fp16)[name = tensor("op_3935_cast_fp16")]; + tensor var_3941_pad_type_0 = const()[name = tensor("op_3941_pad_type_0"), val = tensor("valid")]; + tensor var_3941_strides_0 = const()[name = tensor("op_3941_strides_0"), val = tensor([1, 1])]; + tensor var_3941_pad_0 = const()[name = tensor("op_3941_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_3941_dilations_0 = const()[name = tensor("op_3941_dilations_0"), val = tensor([1, 1])]; + tensor var_3941_groups_0 = const()[name = tensor("op_3941_groups_0"), val = tensor(1)]; + tensor layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59408640))), name = tensor("layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59370368))), shape = tensor([512, 2048, 1, 1])]; + tensor var_3941_cast_fp16 = conv(dilations = var_3941_dilations_0, groups = var_3941_groups_0, pad = var_3941_pad_0, pad_type = var_3941_pad_type_0, strides = var_3941_strides_0, weight = layers_9_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_275_cast_fp16)[name = tensor("op_3941_cast_fp16")]; + tensor x_61_cast_fp16 = add(x = var_3935_cast_fp16, y = var_3941_cast_fp16)[name = tensor("x_61_cast_fp16")]; + tensor var_3943_to_fp16 = const()[name = tensor("op_3943_to_fp16"), val = tensor(0x1p-1)]; + tensor var_3944_cast_fp16 = mul(x = x_61_cast_fp16, y = var_3943_to_fp16)[name = tensor("op_3944_cast_fp16")]; + tensor inputs_99_cast_fp16 = add(x = inputs_97_cast_fp16, y = var_3944_cast_fp16)[name = tensor("inputs_99_cast_fp16")]; + tensor out_99_axes_0 = const()[name = tensor("out_99_axes_0"), val = tensor([1])]; + tensor var_3954_to_fp16 = const()[name = tensor("op_3954_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_99_cast_fp16 = layer_norm(axes = out_99_axes_0, epsilon = var_3954_to_fp16, x = inputs_99_cast_fp16)[name = tensor("out_99_cast_fp16")]; + tensor inputs_101_gamma_0_to_fp16 = const()[name = tensor("inputs_101_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59539776)))]; + tensor inputs_101_beta_0_to_fp16 = const()[name = tensor("inputs_101_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59540864)))]; + tensor inputs_101_epsilon_0_to_fp16 = const()[name = tensor("inputs_101_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_101_cast_fp16 = batch_norm(beta = inputs_101_beta_0_to_fp16, epsilon = inputs_101_epsilon_0_to_fp16, gamma = inputs_101_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_99_cast_fp16)[name = tensor("inputs_101_cast_fp16")]; + tensor var_3968 = const()[name = tensor("op_3968"), val = tensor(3)]; + tensor out_101_axes_0 = const()[name = tensor("out_101_axes_0"), val = tensor([1])]; + tensor var_3999_to_fp16 = const()[name = tensor("op_3999_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_101_cast_fp16 = layer_norm(axes = out_101_axes_0, epsilon = var_3999_to_fp16, x = inputs_101_cast_fp16)[name = tensor("out_101_cast_fp16")]; + tensor input_277_gamma_0_to_fp16 = const()[name = tensor("input_277_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59541952)))]; + tensor input_277_beta_0_to_fp16 = const()[name = tensor("input_277_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59543040)))]; + tensor input_277_epsilon_0_to_fp16 = const()[name = tensor("input_277_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_277_cast_fp16 = batch_norm(beta = input_277_beta_0_to_fp16, epsilon = input_277_epsilon_0_to_fp16, gamma = input_277_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_101_cast_fp16)[name = tensor("input_277_cast_fp16")]; + tensor var_4019_pad_type_0 = const()[name = tensor("op_4019_pad_type_0"), val = tensor("valid")]; + tensor var_4019_strides_0 = const()[name = tensor("op_4019_strides_0"), val = tensor([1, 1])]; + tensor var_4019_pad_0 = const()[name = tensor("op_4019_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4019_dilations_0 = const()[name = tensor("op_4019_dilations_0"), val = tensor([1, 1])]; + tensor var_4019_groups_0 = const()[name = tensor("op_4019_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(59544128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60330624))), name = tensor("layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_10_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60330816)))]; + tensor var_4019_cast_fp16 = conv(bias = layers_10_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4019_dilations_0, groups = var_4019_groups_0, pad = var_4019_pad_0, pad_type = var_4019_pad_type_0, strides = var_4019_strides_0, weight = layers_10_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_277_cast_fp16)[name = tensor("op_4019_cast_fp16")]; + tensor var_4025_pad_type_0 = const()[name = tensor("op_4025_pad_type_0"), val = tensor("valid")]; + tensor var_4025_strides_0 = const()[name = tensor("op_4025_strides_0"), val = tensor([1, 1])]; + tensor var_4025_pad_0 = const()[name = tensor("op_4025_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4025_dilations_0 = const()[name = tensor("op_4025_dilations_0"), val = tensor([1, 1])]; + tensor var_4025_groups_0 = const()[name = tensor("op_4025_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60369216))), name = tensor("layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60334976))), shape = tensor([2048, 512, 1, 1])]; + tensor var_4025_cast_fp16 = conv(dilations = var_4025_dilations_0, groups = var_4025_groups_0, pad = var_4025_pad_0, pad_type = var_4025_pad_type_0, strides = var_4025_strides_0, weight = layers_10_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_277_cast_fp16)[name = tensor("op_4025_cast_fp16")]; + tensor input_279_cast_fp16 = add(x = var_4019_cast_fp16, y = var_4025_cast_fp16)[name = tensor("input_279_cast_fp16")]; + tensor input_281_cast_fp16 = silu(x = input_279_cast_fp16)[name = tensor("input_281_cast_fp16")]; + tensor var_4036_pad_type_0 = const()[name = tensor("op_4036_pad_type_0"), val = tensor("valid")]; + tensor var_4036_strides_0 = const()[name = tensor("op_4036_strides_0"), val = tensor([1, 1])]; + tensor var_4036_pad_0 = const()[name = tensor("op_4036_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4036_dilations_0 = const()[name = tensor("op_4036_dilations_0"), val = tensor([1, 1])]; + tensor var_4036_groups_0 = const()[name = tensor("op_4036_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(60500352))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61286848))), name = tensor("layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_10_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61287040)))]; + tensor var_4036_cast_fp16 = conv(bias = layers_10_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_4036_dilations_0, groups = var_4036_groups_0, pad = var_4036_pad_0, pad_type = var_4036_pad_type_0, strides = var_4036_strides_0, weight = layers_10_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_281_cast_fp16)[name = tensor("op_4036_cast_fp16")]; + tensor var_4042_pad_type_0 = const()[name = tensor("op_4042_pad_type_0"), val = tensor("valid")]; + tensor var_4042_strides_0 = const()[name = tensor("op_4042_strides_0"), val = tensor([1, 1])]; + tensor var_4042_pad_0 = const()[name = tensor("op_4042_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4042_dilations_0 = const()[name = tensor("op_4042_dilations_0"), val = tensor([1, 1])]; + tensor var_4042_groups_0 = const()[name = tensor("op_4042_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61328448))), name = tensor("layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61288128))), shape = tensor([512, 2048, 1, 1])]; + tensor var_4042_cast_fp16 = conv(dilations = var_4042_dilations_0, groups = var_4042_groups_0, pad = var_4042_pad_0, pad_type = var_4042_pad_type_0, strides = var_4042_strides_0, weight = layers_10_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_281_cast_fp16)[name = tensor("op_4042_cast_fp16")]; + tensor x_63_cast_fp16 = add(x = var_4036_cast_fp16, y = var_4042_cast_fp16)[name = tensor("x_63_cast_fp16")]; + tensor var_4044_to_fp16 = const()[name = tensor("op_4044_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4045_cast_fp16 = mul(x = x_63_cast_fp16, y = var_4044_to_fp16)[name = tensor("op_4045_cast_fp16")]; + tensor inputs_103_cast_fp16 = add(x = inputs_101_cast_fp16, y = var_4045_cast_fp16)[name = tensor("inputs_103_cast_fp16")]; + tensor out_103_axes_0 = const()[name = tensor("out_103_axes_0"), val = tensor([1])]; + tensor var_4055_to_fp16 = const()[name = tensor("op_4055_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_103_cast_fp16 = layer_norm(axes = out_103_axes_0, epsilon = var_4055_to_fp16, x = inputs_103_cast_fp16)[name = tensor("out_103_cast_fp16")]; + tensor obj_43_gamma_0_to_fp16 = const()[name = tensor("obj_43_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61459584)))]; + tensor obj_43_beta_0_to_fp16 = const()[name = tensor("obj_43_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61460672)))]; + tensor obj_43_epsilon_0_to_fp16 = const()[name = tensor("obj_43_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_43_cast_fp16 = batch_norm(beta = obj_43_beta_0_to_fp16, epsilon = obj_43_epsilon_0_to_fp16, gamma = obj_43_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_103_cast_fp16)[name = tensor("obj_43_cast_fp16")]; + tensor var_4080_pad_type_0 = const()[name = tensor("op_4080_pad_type_0"), val = tensor("valid")]; + tensor var_4080_strides_0 = const()[name = tensor("op_4080_strides_0"), val = tensor([1, 1])]; + tensor var_4080_pad_0 = const()[name = tensor("op_4080_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4080_dilations_0 = const()[name = tensor("op_4080_dilations_0"), val = tensor([1, 1])]; + tensor var_4080_groups_0 = const()[name = tensor("op_4080_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61461760))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61658432))), name = tensor("layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_10_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61658624)))]; + tensor var_4080_cast_fp16 = conv(bias = layers_10_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_4080_dilations_0, groups = var_4080_groups_0, pad = var_4080_pad_0, pad_type = var_4080_pad_type_0, strides = var_4080_strides_0, weight = layers_10_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4080_cast_fp16")]; + tensor var_4086_pad_type_0 = const()[name = tensor("op_4086_pad_type_0"), val = tensor("valid")]; + tensor var_4086_strides_0 = const()[name = tensor("op_4086_strides_0"), val = tensor([1, 1])]; + tensor var_4086_pad_0 = const()[name = tensor("op_4086_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4086_dilations_0 = const()[name = tensor("op_4086_dilations_0"), val = tensor([1, 1])]; + tensor var_4086_groups_0 = const()[name = tensor("op_4086_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61669056))), name = tensor("layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61659712))), shape = tensor([512, 512, 1, 1])]; + tensor var_4086_cast_fp16 = conv(dilations = var_4086_dilations_0, groups = var_4086_groups_0, pad = var_4086_pad_0, pad_type = var_4086_pad_type_0, strides = var_4086_strides_0, weight = layers_10_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4086_cast_fp16")]; + tensor query_41_cast_fp16 = add(x = var_4080_cast_fp16, y = var_4086_cast_fp16)[name = tensor("query_41_cast_fp16")]; + tensor var_4095_pad_type_0 = const()[name = tensor("op_4095_pad_type_0"), val = tensor("valid")]; + tensor var_4095_strides_0 = const()[name = tensor("op_4095_strides_0"), val = tensor([1, 1])]; + tensor var_4095_pad_0 = const()[name = tensor("op_4095_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4095_dilations_0 = const()[name = tensor("op_4095_dilations_0"), val = tensor([1, 1])]; + tensor var_4095_groups_0 = const()[name = tensor("op_4095_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61701888))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61898560))), name = tensor("layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4095_cast_fp16 = conv(dilations = var_4095_dilations_0, groups = var_4095_groups_0, pad = var_4095_pad_0, pad_type = var_4095_pad_type_0, strides = var_4095_strides_0, weight = layers_10_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4095_cast_fp16")]; + tensor var_4101_pad_type_0 = const()[name = tensor("op_4101_pad_type_0"), val = tensor("valid")]; + tensor var_4101_strides_0 = const()[name = tensor("op_4101_strides_0"), val = tensor([1, 1])]; + tensor var_4101_pad_0 = const()[name = tensor("op_4101_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4101_dilations_0 = const()[name = tensor("op_4101_dilations_0"), val = tensor([1, 1])]; + tensor var_4101_groups_0 = const()[name = tensor("op_4101_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61906368))), name = tensor("layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61898752))), shape = tensor([512, 512, 1, 1])]; + tensor var_4101_cast_fp16 = conv(dilations = var_4101_dilations_0, groups = var_4101_groups_0, pad = var_4101_pad_0, pad_type = var_4101_pad_type_0, strides = var_4101_strides_0, weight = layers_10_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4101_cast_fp16")]; + tensor key_21_cast_fp16 = add(x = var_4095_cast_fp16, y = var_4101_cast_fp16)[name = tensor("key_21_cast_fp16")]; + tensor var_4111_pad_type_0 = const()[name = tensor("op_4111_pad_type_0"), val = tensor("valid")]; + tensor var_4111_strides_0 = const()[name = tensor("op_4111_strides_0"), val = tensor([1, 1])]; + tensor var_4111_pad_0 = const()[name = tensor("op_4111_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4111_dilations_0 = const()[name = tensor("op_4111_dilations_0"), val = tensor([1, 1])]; + tensor var_4111_groups_0 = const()[name = tensor("op_4111_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(61939200))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62135872))), name = tensor("layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_10_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62136064)))]; + tensor var_4111_cast_fp16 = conv(bias = layers_10_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_4111_dilations_0, groups = var_4111_groups_0, pad = var_4111_pad_0, pad_type = var_4111_pad_type_0, strides = var_4111_strides_0, weight = layers_10_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_43_cast_fp16)[name = tensor("op_4111_cast_fp16")]; + tensor var_4117_pad_type_0 = const()[name = tensor("op_4117_pad_type_0"), val = tensor("valid")]; + tensor var_4117_strides_0 = const()[name = tensor("op_4117_strides_0"), val = tensor([1, 1])]; + tensor var_4117_pad_0 = const()[name = tensor("op_4117_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4117_dilations_0 = const()[name = tensor("op_4117_dilations_0"), val = tensor([1, 1])]; + tensor var_4117_groups_0 = const()[name = tensor("op_4117_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62145024))), name = tensor("layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62137152))), shape = tensor([512, 512, 1, 1])]; + tensor var_4117_cast_fp16 = conv(dilations = var_4117_dilations_0, groups = var_4117_groups_0, pad = var_4117_pad_0, pad_type = var_4117_pad_type_0, strides = var_4117_strides_0, weight = layers_10_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_43_cast_fp16)[name = tensor("op_4117_cast_fp16")]; + tensor value_21_cast_fp16 = add(x = var_4111_cast_fp16, y = var_4117_cast_fp16)[name = tensor("value_21_cast_fp16")]; + tensor var_4120_to_fp16 = const()[name = tensor("op_4120_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62177856)))]; + tensor query_43_cast_fp16 = add(x = query_41_cast_fp16, y = var_4120_to_fp16)[name = tensor("query_43_cast_fp16")]; + tensor var_4123_to_fp16 = const()[name = tensor("op_4123_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62178944)))]; + tensor q_with_bias_v_21_cast_fp16 = add(x = query_41_cast_fp16, y = var_4123_to_fp16)[name = tensor("q_with_bias_v_21_cast_fp16")]; + tensor var_4133_pad_type_0 = const()[name = tensor("op_4133_pad_type_0"), val = tensor("valid")]; + tensor var_4133_strides_0 = const()[name = tensor("op_4133_strides_0"), val = tensor([1, 1])]; + tensor var_4133_pad_0 = const()[name = tensor("op_4133_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4133_dilations_0 = const()[name = tensor("op_4133_dilations_0"), val = tensor([1, 1])]; + tensor var_4133_groups_0 = const()[name = tensor("op_4133_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62180032))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62376704))), name = tensor("layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4133_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4133_dilations_0, groups = var_4133_groups_0, pad = var_4133_pad_0, pad_type = var_4133_pad_type_0, strides = var_4133_strides_0, weight = layers_10_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4133_cast_fp16")]; + tensor var_4139_pad_type_0 = const()[name = tensor("op_4139_pad_type_0"), val = tensor("valid")]; + tensor var_4139_strides_0 = const()[name = tensor("op_4139_strides_0"), val = tensor([1, 1])]; + tensor var_4139_pad_0 = const()[name = tensor("op_4139_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4139_dilations_0 = const()[name = tensor("op_4139_dilations_0"), val = tensor([1, 1])]; + tensor var_4139_groups_0 = const()[name = tensor("op_4139_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62393344))), name = tensor("layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62376896))), shape = tensor([512, 512, 1, 1])]; + tensor var_4139_cast_fp16 = conv(dilations = var_4139_dilations_0, groups = var_4139_groups_0, pad = var_4139_pad_0, pad_type = var_4139_pad_type_0, strides = var_4139_strides_0, weight = layers_10_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4139_cast_fp16")]; + tensor p_21_cast_fp16 = add(x = var_4133_cast_fp16, y = var_4139_cast_fp16)[name = tensor("p_21_cast_fp16")]; + tensor var_4143 = const()[name = tensor("op_4143"), val = tensor([1, 8, 64, 188])]; + tensor var_4144_cast_fp16 = reshape(shape = var_4143, x = q_with_bias_v_21_cast_fp16)[name = tensor("op_4144_cast_fp16")]; + tensor var_4145 = const()[name = tensor("op_4145"), val = tensor([1, 8, 64, -1])]; + tensor var_4146_cast_fp16 = reshape(shape = var_4145, x = p_21_cast_fp16)[name = tensor("op_4146_cast_fp16")]; + tensor matrix_bd_81_transpose_x_0 = const()[name = tensor("matrix_bd_81_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_81_transpose_y_0 = const()[name = tensor("matrix_bd_81_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_81_cast_fp16 = matmul(transpose_x = matrix_bd_81_transpose_x_0, transpose_y = matrix_bd_81_transpose_y_0, x = var_4144_cast_fp16, y = var_4146_cast_fp16)[name = tensor("matrix_bd_81_cast_fp16")]; + tensor matrix_bd_83_pad_0 = const()[name = tensor("matrix_bd_83_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_83_mode_0 = const()[name = tensor("matrix_bd_83_mode_0"), val = tensor("constant")]; + tensor const_120_to_fp16 = const()[name = tensor("const_120_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_83_cast_fp16 = pad(constant_val = const_120_to_fp16, mode = matrix_bd_83_mode_0, pad = matrix_bd_83_pad_0, x = matrix_bd_81_cast_fp16)[name = tensor("matrix_bd_83_cast_fp16")]; + tensor var_4155 = const()[name = tensor("op_4155"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_85_cast_fp16 = reshape(shape = var_4155, x = matrix_bd_83_cast_fp16)[name = tensor("matrix_bd_85_cast_fp16")]; + tensor var_4159_begin_0 = const()[name = tensor("op_4159_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4159_end_0 = const()[name = tensor("op_4159_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4159_end_mask_0 = const()[name = tensor("op_4159_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4159_cast_fp16 = slice_by_index(begin = var_4159_begin_0, end = var_4159_end_0, end_mask = var_4159_end_mask_0, x = matrix_bd_85_cast_fp16)[name = tensor("op_4159_cast_fp16")]; + tensor var_4160 = const()[name = tensor("op_4160"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_87_cast_fp16 = reshape(shape = var_4160, x = var_4159_cast_fp16)[name = tensor("matrix_bd_87_cast_fp16")]; + tensor var_4165_begin_0 = const()[name = tensor("op_4165_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4165_end_0 = const()[name = tensor("op_4165_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4165_end_mask_0 = const()[name = tensor("op_4165_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4165_cast_fp16 = slice_by_index(begin = var_4165_begin_0, end = var_4165_end_0, end_mask = var_4165_end_mask_0, x = matrix_bd_87_cast_fp16)[name = tensor("op_4165_cast_fp16")]; + tensor var_4166_to_fp16 = const()[name = tensor("op_4166_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_21_cast_fp16 = mul(x = var_4165_cast_fp16, y = var_4166_to_fp16)[name = tensor("qk_mask_21_cast_fp16")]; + tensor var_4170 = const()[name = tensor("op_4170"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_21_cast_fp16 = reshape(shape = var_4170, x = query_43_cast_fp16)[name = tensor("mh_q_21_cast_fp16")]; + tensor var_4172_to_fp16 = const()[name = tensor("op_4172_to_fp16"), val = tensor(0x1p-3)]; + tensor var_4173_cast_fp16 = mul(x = mh_q_21_cast_fp16, y = var_4172_to_fp16)[name = tensor("op_4173_cast_fp16")]; + tensor var_4176 = const()[name = tensor("op_4176"), val = tensor([1, 8, 64, 188])]; + tensor var_4177_cast_fp16 = reshape(shape = var_4176, x = key_21_cast_fp16)[name = tensor("op_4177_cast_fp16")]; + tensor mh_w_41_transpose_x_0 = const()[name = tensor("mh_w_41_transpose_x_0"), val = tensor(true)]; + tensor mh_w_41_transpose_y_0 = const()[name = tensor("mh_w_41_transpose_y_0"), val = tensor(false)]; + tensor mh_w_41_cast_fp16 = matmul(transpose_x = mh_w_41_transpose_x_0, transpose_y = mh_w_41_transpose_y_0, x = var_4173_cast_fp16, y = var_4177_cast_fp16)[name = tensor("mh_w_41_cast_fp16")]; + tensor mh_w_43_cast_fp16 = add(x = mh_w_41_cast_fp16, y = qk_mask_21_cast_fp16)[name = tensor("mh_w_43_cast_fp16")]; + tensor var_4181_cast_fp16 = softmax(axis = var_3968, x = mh_w_43_cast_fp16)[name = tensor("op_4181_cast_fp16")]; + tensor var_4182 = const()[name = tensor("op_4182"), val = tensor([1, 8, 64, 188])]; + tensor var_4183_cast_fp16 = reshape(shape = var_4182, x = value_21_cast_fp16)[name = tensor("op_4183_cast_fp16")]; + tensor attn_21_transpose_x_0 = const()[name = tensor("attn_21_transpose_x_0"), val = tensor(false)]; + tensor attn_21_transpose_y_0 = const()[name = tensor("attn_21_transpose_y_0"), val = tensor(true)]; + tensor attn_21_cast_fp16 = matmul(transpose_x = attn_21_transpose_x_0, transpose_y = attn_21_transpose_y_0, x = var_4183_cast_fp16, y = var_4181_cast_fp16)[name = tensor("attn_21_cast_fp16")]; + tensor var_4186 = const()[name = tensor("op_4186"), val = tensor([1, 512, 1, 188])]; + tensor input_283_cast_fp16 = reshape(shape = var_4186, x = attn_21_cast_fp16)[name = tensor("input_283_cast_fp16")]; + tensor var_4196_pad_type_0 = const()[name = tensor("op_4196_pad_type_0"), val = tensor("valid")]; + tensor var_4196_strides_0 = const()[name = tensor("op_4196_strides_0"), val = tensor([1, 1])]; + tensor var_4196_pad_0 = const()[name = tensor("op_4196_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4196_dilations_0 = const()[name = tensor("op_4196_dilations_0"), val = tensor([1, 1])]; + tensor var_4196_groups_0 = const()[name = tensor("op_4196_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62426176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62622848))), name = tensor("layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_10_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62623040)))]; + tensor var_4196_cast_fp16 = conv(bias = layers_10_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_4196_dilations_0, groups = var_4196_groups_0, pad = var_4196_pad_0, pad_type = var_4196_pad_type_0, strides = var_4196_strides_0, weight = layers_10_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_283_cast_fp16)[name = tensor("op_4196_cast_fp16")]; + tensor var_4202_pad_type_0 = const()[name = tensor("op_4202_pad_type_0"), val = tensor("valid")]; + tensor var_4202_strides_0 = const()[name = tensor("op_4202_strides_0"), val = tensor([1, 1])]; + tensor var_4202_pad_0 = const()[name = tensor("op_4202_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4202_dilations_0 = const()[name = tensor("op_4202_dilations_0"), val = tensor([1, 1])]; + tensor var_4202_groups_0 = const()[name = tensor("op_4202_groups_0"), val = tensor(1)]; + tensor layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62632832))), name = tensor("layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62624128))), shape = tensor([512, 512, 1, 1])]; + tensor var_4202_cast_fp16 = conv(dilations = var_4202_dilations_0, groups = var_4202_groups_0, pad = var_4202_pad_0, pad_type = var_4202_pad_type_0, strides = var_4202_strides_0, weight = layers_10_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_283_cast_fp16)[name = tensor("op_4202_cast_fp16")]; + tensor obj_45_cast_fp16 = add(x = var_4196_cast_fp16, y = var_4202_cast_fp16)[name = tensor("obj_45_cast_fp16")]; + tensor inputs_105_cast_fp16 = add(x = inputs_103_cast_fp16, y = obj_45_cast_fp16)[name = tensor("inputs_105_cast_fp16")]; + tensor out_105_axes_0 = const()[name = tensor("out_105_axes_0"), val = tensor([1])]; + tensor var_4213_to_fp16 = const()[name = tensor("op_4213_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_105_cast_fp16 = layer_norm(axes = out_105_axes_0, epsilon = var_4213_to_fp16, x = inputs_105_cast_fp16)[name = tensor("out_105_cast_fp16")]; + tensor input_285_gamma_0_to_fp16 = const()[name = tensor("input_285_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62665664)))]; + tensor input_285_beta_0_to_fp16 = const()[name = tensor("input_285_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62666752)))]; + tensor input_285_epsilon_0_to_fp16 = const()[name = tensor("input_285_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_285_cast_fp16 = batch_norm(beta = input_285_beta_0_to_fp16, epsilon = input_285_epsilon_0_to_fp16, gamma = input_285_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_105_cast_fp16)[name = tensor("input_285_cast_fp16")]; + tensor var_4235_pad_type_0 = const()[name = tensor("op_4235_pad_type_0"), val = tensor("valid")]; + tensor var_4235_strides_0 = const()[name = tensor("op_4235_strides_0"), val = tensor([1, 1])]; + tensor var_4235_pad_0 = const()[name = tensor("op_4235_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4235_dilations_0 = const()[name = tensor("op_4235_dilations_0"), val = tensor([1, 1])]; + tensor var_4235_groups_0 = const()[name = tensor("op_4235_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(62667840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63061120))), name = tensor("layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_10_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63061312)))]; + tensor var_4235_cast_fp16 = conv(bias = layers_10_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_4235_dilations_0, groups = var_4235_groups_0, pad = var_4235_pad_0, pad_type = var_4235_pad_type_0, strides = var_4235_strides_0, weight = layers_10_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_285_cast_fp16)[name = tensor("op_4235_cast_fp16")]; + tensor var_4241_pad_type_0 = const()[name = tensor("op_4241_pad_type_0"), val = tensor("valid")]; + tensor var_4241_strides_0 = const()[name = tensor("op_4241_strides_0"), val = tensor([1, 1])]; + tensor var_4241_pad_0 = const()[name = tensor("op_4241_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4241_dilations_0 = const()[name = tensor("op_4241_dilations_0"), val = tensor([1, 1])]; + tensor var_4241_groups_0 = const()[name = tensor("op_4241_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63080960))), name = tensor("layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63063424))), shape = tensor([1024, 512, 1, 1])]; + tensor var_4241_cast_fp16 = conv(dilations = var_4241_dilations_0, groups = var_4241_groups_0, pad = var_4241_pad_0, pad_type = var_4241_pad_type_0, strides = var_4241_strides_0, weight = layers_10_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_285_cast_fp16)[name = tensor("op_4241_cast_fp16")]; + tensor input_287_cast_fp16 = add(x = var_4235_cast_fp16, y = var_4241_cast_fp16)[name = tensor("input_287_cast_fp16")]; + tensor input_289_split_num_splits_0 = const()[name = tensor("input_289_split_num_splits_0"), val = tensor(2)]; + tensor input_289_split_axis_0 = const()[name = tensor("input_289_split_axis_0"), val = tensor(1)]; + tensor input_289_split_cast_fp16_0, tensor input_289_split_cast_fp16_1 = split(axis = input_289_split_axis_0, num_splits = input_289_split_num_splits_0, x = input_287_cast_fp16)[name = tensor("input_289_split_cast_fp16")]; + tensor input_289_split_1_sigmoid_cast_fp16 = sigmoid(x = input_289_split_cast_fp16_1)[name = tensor("input_289_split_1_sigmoid_cast_fp16")]; + tensor input_289_cast_fp16 = mul(x = input_289_split_cast_fp16_0, y = input_289_split_1_sigmoid_cast_fp16)[name = tensor("input_289_cast_fp16")]; + tensor input_291_pad_type_0 = const()[name = tensor("input_291_pad_type_0"), val = tensor("custom")]; + tensor input_291_pad_0 = const()[name = tensor("input_291_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_291_groups_0 = const()[name = tensor("input_291_groups_0"), val = tensor(512)]; + tensor input_291_strides_0 = const()[name = tensor("input_291_strides_0"), val = tensor([1, 1])]; + tensor input_291_dilations_0 = const()[name = tensor("input_291_dilations_0"), val = tensor([1, 1])]; + tensor const_211_to_fp16 = const()[name = tensor("const_211_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63146560)))]; + tensor const_212_to_fp16 = const()[name = tensor("const_212_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63155840)))]; + tensor input_293_cast_fp16 = conv(bias = const_212_to_fp16, dilations = input_291_dilations_0, groups = input_291_groups_0, pad = input_291_pad_0, pad_type = input_291_pad_type_0, strides = input_291_strides_0, weight = const_211_to_fp16, x = input_289_cast_fp16)[name = tensor("input_293_cast_fp16")]; + tensor input_295_cast_fp16 = silu(x = input_293_cast_fp16)[name = tensor("input_295_cast_fp16")]; + tensor var_4265_pad_type_0 = const()[name = tensor("op_4265_pad_type_0"), val = tensor("valid")]; + tensor var_4265_strides_0 = const()[name = tensor("op_4265_strides_0"), val = tensor([1, 1])]; + tensor var_4265_pad_0 = const()[name = tensor("op_4265_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4265_dilations_0 = const()[name = tensor("op_4265_dilations_0"), val = tensor([1, 1])]; + tensor var_4265_groups_0 = const()[name = tensor("op_4265_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63156928))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63353600))), name = tensor("layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_10_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63353792)))]; + tensor var_4265_cast_fp16 = conv(bias = layers_10_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_4265_dilations_0, groups = var_4265_groups_0, pad = var_4265_pad_0, pad_type = var_4265_pad_type_0, strides = var_4265_strides_0, weight = layers_10_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_295_cast_fp16)[name = tensor("op_4265_cast_fp16")]; + tensor var_4271_pad_type_0 = const()[name = tensor("op_4271_pad_type_0"), val = tensor("valid")]; + tensor var_4271_strides_0 = const()[name = tensor("op_4271_strides_0"), val = tensor([1, 1])]; + tensor var_4271_pad_0 = const()[name = tensor("op_4271_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4271_dilations_0 = const()[name = tensor("op_4271_dilations_0"), val = tensor([1, 1])]; + tensor var_4271_groups_0 = const()[name = tensor("op_4271_groups_0"), val = tensor(1)]; + tensor layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63363712))), name = tensor("layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63354880))), shape = tensor([512, 512, 1, 1])]; + tensor var_4271_cast_fp16 = conv(dilations = var_4271_dilations_0, groups = var_4271_groups_0, pad = var_4271_pad_0, pad_type = var_4271_pad_type_0, strides = var_4271_strides_0, weight = layers_10_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_295_cast_fp16)[name = tensor("op_4271_cast_fp16")]; + tensor x_65_cast_fp16 = add(x = var_4265_cast_fp16, y = var_4271_cast_fp16)[name = tensor("x_65_cast_fp16")]; + tensor inputs_107_cast_fp16 = add(x = inputs_105_cast_fp16, y = x_65_cast_fp16)[name = tensor("inputs_107_cast_fp16")]; + tensor out_107_axes_0 = const()[name = tensor("out_107_axes_0"), val = tensor([1])]; + tensor var_4282_to_fp16 = const()[name = tensor("op_4282_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_107_cast_fp16 = layer_norm(axes = out_107_axes_0, epsilon = var_4282_to_fp16, x = inputs_107_cast_fp16)[name = tensor("out_107_cast_fp16")]; + tensor input_297_gamma_0_to_fp16 = const()[name = tensor("input_297_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63396544)))]; + tensor input_297_beta_0_to_fp16 = const()[name = tensor("input_297_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63397632)))]; + tensor input_297_epsilon_0_to_fp16 = const()[name = tensor("input_297_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_297_cast_fp16 = batch_norm(beta = input_297_beta_0_to_fp16, epsilon = input_297_epsilon_0_to_fp16, gamma = input_297_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_107_cast_fp16)[name = tensor("input_297_cast_fp16")]; + tensor var_4302_pad_type_0 = const()[name = tensor("op_4302_pad_type_0"), val = tensor("valid")]; + tensor var_4302_strides_0 = const()[name = tensor("op_4302_strides_0"), val = tensor([1, 1])]; + tensor var_4302_pad_0 = const()[name = tensor("op_4302_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4302_dilations_0 = const()[name = tensor("op_4302_dilations_0"), val = tensor([1, 1])]; + tensor var_4302_groups_0 = const()[name = tensor("op_4302_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(63398720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64185216))), name = tensor("layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_10_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64185408)))]; + tensor var_4302_cast_fp16 = conv(bias = layers_10_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_4302_dilations_0, groups = var_4302_groups_0, pad = var_4302_pad_0, pad_type = var_4302_pad_type_0, strides = var_4302_strides_0, weight = layers_10_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_297_cast_fp16)[name = tensor("op_4302_cast_fp16")]; + tensor var_4308_pad_type_0 = const()[name = tensor("op_4308_pad_type_0"), val = tensor("valid")]; + tensor var_4308_strides_0 = const()[name = tensor("op_4308_strides_0"), val = tensor([1, 1])]; + tensor var_4308_pad_0 = const()[name = tensor("op_4308_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4308_dilations_0 = const()[name = tensor("op_4308_dilations_0"), val = tensor([1, 1])]; + tensor var_4308_groups_0 = const()[name = tensor("op_4308_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64224256))), name = tensor("layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64189568))), shape = tensor([2048, 512, 1, 1])]; + tensor var_4308_cast_fp16 = conv(dilations = var_4308_dilations_0, groups = var_4308_groups_0, pad = var_4308_pad_0, pad_type = var_4308_pad_type_0, strides = var_4308_strides_0, weight = layers_10_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_297_cast_fp16)[name = tensor("op_4308_cast_fp16")]; + tensor input_299_cast_fp16 = add(x = var_4302_cast_fp16, y = var_4308_cast_fp16)[name = tensor("input_299_cast_fp16")]; + tensor input_301_cast_fp16 = silu(x = input_299_cast_fp16)[name = tensor("input_301_cast_fp16")]; + tensor var_4319_pad_type_0 = const()[name = tensor("op_4319_pad_type_0"), val = tensor("valid")]; + tensor var_4319_strides_0 = const()[name = tensor("op_4319_strides_0"), val = tensor([1, 1])]; + tensor var_4319_pad_0 = const()[name = tensor("op_4319_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4319_dilations_0 = const()[name = tensor("op_4319_dilations_0"), val = tensor([1, 1])]; + tensor var_4319_groups_0 = const()[name = tensor("op_4319_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(64355392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65141888))), name = tensor("layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_10_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_10_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65142080)))]; + tensor var_4319_cast_fp16 = conv(bias = layers_10_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_4319_dilations_0, groups = var_4319_groups_0, pad = var_4319_pad_0, pad_type = var_4319_pad_type_0, strides = var_4319_strides_0, weight = layers_10_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_301_cast_fp16)[name = tensor("op_4319_cast_fp16")]; + tensor var_4325_pad_type_0 = const()[name = tensor("op_4325_pad_type_0"), val = tensor("valid")]; + tensor var_4325_strides_0 = const()[name = tensor("op_4325_strides_0"), val = tensor([1, 1])]; + tensor var_4325_pad_0 = const()[name = tensor("op_4325_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4325_dilations_0 = const()[name = tensor("op_4325_dilations_0"), val = tensor([1, 1])]; + tensor var_4325_groups_0 = const()[name = tensor("op_4325_groups_0"), val = tensor(1)]; + tensor layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65185344))), name = tensor("layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65143168))), shape = tensor([512, 2048, 1, 1])]; + tensor var_4325_cast_fp16 = conv(dilations = var_4325_dilations_0, groups = var_4325_groups_0, pad = var_4325_pad_0, pad_type = var_4325_pad_type_0, strides = var_4325_strides_0, weight = layers_10_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_301_cast_fp16)[name = tensor("op_4325_cast_fp16")]; + tensor x_67_cast_fp16 = add(x = var_4319_cast_fp16, y = var_4325_cast_fp16)[name = tensor("x_67_cast_fp16")]; + tensor var_4327_to_fp16 = const()[name = tensor("op_4327_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4328_cast_fp16 = mul(x = x_67_cast_fp16, y = var_4327_to_fp16)[name = tensor("op_4328_cast_fp16")]; + tensor inputs_109_cast_fp16 = add(x = inputs_107_cast_fp16, y = var_4328_cast_fp16)[name = tensor("inputs_109_cast_fp16")]; + tensor out_109_axes_0 = const()[name = tensor("out_109_axes_0"), val = tensor([1])]; + tensor var_4338_to_fp16 = const()[name = tensor("op_4338_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_109_cast_fp16 = layer_norm(axes = out_109_axes_0, epsilon = var_4338_to_fp16, x = inputs_109_cast_fp16)[name = tensor("out_109_cast_fp16")]; + tensor inputs_111_gamma_0_to_fp16 = const()[name = tensor("inputs_111_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65316480)))]; + tensor inputs_111_beta_0_to_fp16 = const()[name = tensor("inputs_111_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65317568)))]; + tensor inputs_111_epsilon_0_to_fp16 = const()[name = tensor("inputs_111_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_111_cast_fp16 = batch_norm(beta = inputs_111_beta_0_to_fp16, epsilon = inputs_111_epsilon_0_to_fp16, gamma = inputs_111_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_109_cast_fp16)[name = tensor("inputs_111_cast_fp16")]; + tensor var_4352 = const()[name = tensor("op_4352"), val = tensor(3)]; + tensor out_111_axes_0 = const()[name = tensor("out_111_axes_0"), val = tensor([1])]; + tensor var_4383_to_fp16 = const()[name = tensor("op_4383_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_111_cast_fp16 = layer_norm(axes = out_111_axes_0, epsilon = var_4383_to_fp16, x = inputs_111_cast_fp16)[name = tensor("out_111_cast_fp16")]; + tensor input_303_gamma_0_to_fp16 = const()[name = tensor("input_303_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65318656)))]; + tensor input_303_beta_0_to_fp16 = const()[name = tensor("input_303_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65319744)))]; + tensor input_303_epsilon_0_to_fp16 = const()[name = tensor("input_303_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_303_cast_fp16 = batch_norm(beta = input_303_beta_0_to_fp16, epsilon = input_303_epsilon_0_to_fp16, gamma = input_303_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_111_cast_fp16)[name = tensor("input_303_cast_fp16")]; + tensor var_4403_pad_type_0 = const()[name = tensor("op_4403_pad_type_0"), val = tensor("valid")]; + tensor var_4403_strides_0 = const()[name = tensor("op_4403_strides_0"), val = tensor([1, 1])]; + tensor var_4403_pad_0 = const()[name = tensor("op_4403_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4403_dilations_0 = const()[name = tensor("op_4403_dilations_0"), val = tensor([1, 1])]; + tensor var_4403_groups_0 = const()[name = tensor("op_4403_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(65320832))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66107328))), name = tensor("layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_11_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66107520)))]; + tensor var_4403_cast_fp16 = conv(bias = layers_11_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4403_dilations_0, groups = var_4403_groups_0, pad = var_4403_pad_0, pad_type = var_4403_pad_type_0, strides = var_4403_strides_0, weight = layers_11_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_303_cast_fp16)[name = tensor("op_4403_cast_fp16")]; + tensor var_4409_pad_type_0 = const()[name = tensor("op_4409_pad_type_0"), val = tensor("valid")]; + tensor var_4409_strides_0 = const()[name = tensor("op_4409_strides_0"), val = tensor([1, 1])]; + tensor var_4409_pad_0 = const()[name = tensor("op_4409_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4409_dilations_0 = const()[name = tensor("op_4409_dilations_0"), val = tensor([1, 1])]; + tensor var_4409_groups_0 = const()[name = tensor("op_4409_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66147264))), name = tensor("layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66111680))), shape = tensor([2048, 512, 1, 1])]; + tensor var_4409_cast_fp16 = conv(dilations = var_4409_dilations_0, groups = var_4409_groups_0, pad = var_4409_pad_0, pad_type = var_4409_pad_type_0, strides = var_4409_strides_0, weight = layers_11_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_303_cast_fp16)[name = tensor("op_4409_cast_fp16")]; + tensor input_305_cast_fp16 = add(x = var_4403_cast_fp16, y = var_4409_cast_fp16)[name = tensor("input_305_cast_fp16")]; + tensor input_307_cast_fp16 = silu(x = input_305_cast_fp16)[name = tensor("input_307_cast_fp16")]; + tensor var_4420_pad_type_0 = const()[name = tensor("op_4420_pad_type_0"), val = tensor("valid")]; + tensor var_4420_strides_0 = const()[name = tensor("op_4420_strides_0"), val = tensor([1, 1])]; + tensor var_4420_pad_0 = const()[name = tensor("op_4420_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4420_dilations_0 = const()[name = tensor("op_4420_dilations_0"), val = tensor([1, 1])]; + tensor var_4420_groups_0 = const()[name = tensor("op_4420_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(66278400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67064896))), name = tensor("layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_11_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67065088)))]; + tensor var_4420_cast_fp16 = conv(bias = layers_11_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_4420_dilations_0, groups = var_4420_groups_0, pad = var_4420_pad_0, pad_type = var_4420_pad_type_0, strides = var_4420_strides_0, weight = layers_11_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_307_cast_fp16)[name = tensor("op_4420_cast_fp16")]; + tensor var_4426_pad_type_0 = const()[name = tensor("op_4426_pad_type_0"), val = tensor("valid")]; + tensor var_4426_strides_0 = const()[name = tensor("op_4426_strides_0"), val = tensor([1, 1])]; + tensor var_4426_pad_0 = const()[name = tensor("op_4426_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4426_dilations_0 = const()[name = tensor("op_4426_dilations_0"), val = tensor([1, 1])]; + tensor var_4426_groups_0 = const()[name = tensor("op_4426_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67105920))), name = tensor("layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67066176))), shape = tensor([512, 2048, 1, 1])]; + tensor var_4426_cast_fp16 = conv(dilations = var_4426_dilations_0, groups = var_4426_groups_0, pad = var_4426_pad_0, pad_type = var_4426_pad_type_0, strides = var_4426_strides_0, weight = layers_11_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_307_cast_fp16)[name = tensor("op_4426_cast_fp16")]; + tensor x_69_cast_fp16 = add(x = var_4420_cast_fp16, y = var_4426_cast_fp16)[name = tensor("x_69_cast_fp16")]; + tensor var_4428_to_fp16 = const()[name = tensor("op_4428_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4429_cast_fp16 = mul(x = x_69_cast_fp16, y = var_4428_to_fp16)[name = tensor("op_4429_cast_fp16")]; + tensor inputs_113_cast_fp16 = add(x = inputs_111_cast_fp16, y = var_4429_cast_fp16)[name = tensor("inputs_113_cast_fp16")]; + tensor out_113_axes_0 = const()[name = tensor("out_113_axes_0"), val = tensor([1])]; + tensor var_4439_to_fp16 = const()[name = tensor("op_4439_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_113_cast_fp16 = layer_norm(axes = out_113_axes_0, epsilon = var_4439_to_fp16, x = inputs_113_cast_fp16)[name = tensor("out_113_cast_fp16")]; + tensor obj_47_gamma_0_to_fp16 = const()[name = tensor("obj_47_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67237056)))]; + tensor obj_47_beta_0_to_fp16 = const()[name = tensor("obj_47_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67238144)))]; + tensor obj_47_epsilon_0_to_fp16 = const()[name = tensor("obj_47_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_47_cast_fp16 = batch_norm(beta = obj_47_beta_0_to_fp16, epsilon = obj_47_epsilon_0_to_fp16, gamma = obj_47_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_113_cast_fp16)[name = tensor("obj_47_cast_fp16")]; + tensor var_4464_pad_type_0 = const()[name = tensor("op_4464_pad_type_0"), val = tensor("valid")]; + tensor var_4464_strides_0 = const()[name = tensor("op_4464_strides_0"), val = tensor([1, 1])]; + tensor var_4464_pad_0 = const()[name = tensor("op_4464_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4464_dilations_0 = const()[name = tensor("op_4464_dilations_0"), val = tensor([1, 1])]; + tensor var_4464_groups_0 = const()[name = tensor("op_4464_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67239232))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67435904))), name = tensor("layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_11_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67436096)))]; + tensor var_4464_cast_fp16 = conv(bias = layers_11_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_4464_dilations_0, groups = var_4464_groups_0, pad = var_4464_pad_0, pad_type = var_4464_pad_type_0, strides = var_4464_strides_0, weight = layers_11_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4464_cast_fp16")]; + tensor var_4470_pad_type_0 = const()[name = tensor("op_4470_pad_type_0"), val = tensor("valid")]; + tensor var_4470_strides_0 = const()[name = tensor("op_4470_strides_0"), val = tensor([1, 1])]; + tensor var_4470_pad_0 = const()[name = tensor("op_4470_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4470_dilations_0 = const()[name = tensor("op_4470_dilations_0"), val = tensor([1, 1])]; + tensor var_4470_groups_0 = const()[name = tensor("op_4470_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67446528))), name = tensor("layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67437184))), shape = tensor([512, 512, 1, 1])]; + tensor var_4470_cast_fp16 = conv(dilations = var_4470_dilations_0, groups = var_4470_groups_0, pad = var_4470_pad_0, pad_type = var_4470_pad_type_0, strides = var_4470_strides_0, weight = layers_11_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4470_cast_fp16")]; + tensor query_45_cast_fp16 = add(x = var_4464_cast_fp16, y = var_4470_cast_fp16)[name = tensor("query_45_cast_fp16")]; + tensor var_4479_pad_type_0 = const()[name = tensor("op_4479_pad_type_0"), val = tensor("valid")]; + tensor var_4479_strides_0 = const()[name = tensor("op_4479_strides_0"), val = tensor([1, 1])]; + tensor var_4479_pad_0 = const()[name = tensor("op_4479_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4479_dilations_0 = const()[name = tensor("op_4479_dilations_0"), val = tensor([1, 1])]; + tensor var_4479_groups_0 = const()[name = tensor("op_4479_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67479360))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67676032))), name = tensor("layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4479_cast_fp16 = conv(dilations = var_4479_dilations_0, groups = var_4479_groups_0, pad = var_4479_pad_0, pad_type = var_4479_pad_type_0, strides = var_4479_strides_0, weight = layers_11_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4479_cast_fp16")]; + tensor var_4485_pad_type_0 = const()[name = tensor("op_4485_pad_type_0"), val = tensor("valid")]; + tensor var_4485_strides_0 = const()[name = tensor("op_4485_strides_0"), val = tensor([1, 1])]; + tensor var_4485_pad_0 = const()[name = tensor("op_4485_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4485_dilations_0 = const()[name = tensor("op_4485_dilations_0"), val = tensor([1, 1])]; + tensor var_4485_groups_0 = const()[name = tensor("op_4485_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67685952))), name = tensor("layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67676224))), shape = tensor([512, 512, 1, 1])]; + tensor var_4485_cast_fp16 = conv(dilations = var_4485_dilations_0, groups = var_4485_groups_0, pad = var_4485_pad_0, pad_type = var_4485_pad_type_0, strides = var_4485_strides_0, weight = layers_11_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4485_cast_fp16")]; + tensor key_23_cast_fp16 = add(x = var_4479_cast_fp16, y = var_4485_cast_fp16)[name = tensor("key_23_cast_fp16")]; + tensor var_4495_pad_type_0 = const()[name = tensor("op_4495_pad_type_0"), val = tensor("valid")]; + tensor var_4495_strides_0 = const()[name = tensor("op_4495_strides_0"), val = tensor([1, 1])]; + tensor var_4495_pad_0 = const()[name = tensor("op_4495_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4495_dilations_0 = const()[name = tensor("op_4495_dilations_0"), val = tensor([1, 1])]; + tensor var_4495_groups_0 = const()[name = tensor("op_4495_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67718784))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67915456))), name = tensor("layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_11_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67915648)))]; + tensor var_4495_cast_fp16 = conv(bias = layers_11_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_4495_dilations_0, groups = var_4495_groups_0, pad = var_4495_pad_0, pad_type = var_4495_pad_type_0, strides = var_4495_strides_0, weight = layers_11_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_47_cast_fp16)[name = tensor("op_4495_cast_fp16")]; + tensor var_4501_pad_type_0 = const()[name = tensor("op_4501_pad_type_0"), val = tensor("valid")]; + tensor var_4501_strides_0 = const()[name = tensor("op_4501_strides_0"), val = tensor([1, 1])]; + tensor var_4501_pad_0 = const()[name = tensor("op_4501_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4501_dilations_0 = const()[name = tensor("op_4501_dilations_0"), val = tensor([1, 1])]; + tensor var_4501_groups_0 = const()[name = tensor("op_4501_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67925056))), name = tensor("layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67916736))), shape = tensor([512, 512, 1, 1])]; + tensor var_4501_cast_fp16 = conv(dilations = var_4501_dilations_0, groups = var_4501_groups_0, pad = var_4501_pad_0, pad_type = var_4501_pad_type_0, strides = var_4501_strides_0, weight = layers_11_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_47_cast_fp16)[name = tensor("op_4501_cast_fp16")]; + tensor value_23_cast_fp16 = add(x = var_4495_cast_fp16, y = var_4501_cast_fp16)[name = tensor("value_23_cast_fp16")]; + tensor var_4504_to_fp16 = const()[name = tensor("op_4504_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67957888)))]; + tensor query_47_cast_fp16 = add(x = query_45_cast_fp16, y = var_4504_to_fp16)[name = tensor("query_47_cast_fp16")]; + tensor var_4507_to_fp16 = const()[name = tensor("op_4507_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67958976)))]; + tensor q_with_bias_v_23_cast_fp16 = add(x = query_45_cast_fp16, y = var_4507_to_fp16)[name = tensor("q_with_bias_v_23_cast_fp16")]; + tensor var_4517_pad_type_0 = const()[name = tensor("op_4517_pad_type_0"), val = tensor("valid")]; + tensor var_4517_strides_0 = const()[name = tensor("op_4517_strides_0"), val = tensor([1, 1])]; + tensor var_4517_pad_0 = const()[name = tensor("op_4517_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4517_dilations_0 = const()[name = tensor("op_4517_dilations_0"), val = tensor([1, 1])]; + tensor var_4517_groups_0 = const()[name = tensor("op_4517_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(67960064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68156736))), name = tensor("layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4517_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4517_dilations_0, groups = var_4517_groups_0, pad = var_4517_pad_0, pad_type = var_4517_pad_type_0, strides = var_4517_strides_0, weight = layers_11_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4517_cast_fp16")]; + tensor var_4523_pad_type_0 = const()[name = tensor("op_4523_pad_type_0"), val = tensor("valid")]; + tensor var_4523_strides_0 = const()[name = tensor("op_4523_strides_0"), val = tensor([1, 1])]; + tensor var_4523_pad_0 = const()[name = tensor("op_4523_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4523_dilations_0 = const()[name = tensor("op_4523_dilations_0"), val = tensor([1, 1])]; + tensor var_4523_groups_0 = const()[name = tensor("op_4523_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68178304))), name = tensor("layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68156928))), shape = tensor([512, 512, 1, 1])]; + tensor var_4523_cast_fp16 = conv(dilations = var_4523_dilations_0, groups = var_4523_groups_0, pad = var_4523_pad_0, pad_type = var_4523_pad_type_0, strides = var_4523_strides_0, weight = layers_11_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4523_cast_fp16")]; + tensor p_23_cast_fp16 = add(x = var_4517_cast_fp16, y = var_4523_cast_fp16)[name = tensor("p_23_cast_fp16")]; + tensor var_4527 = const()[name = tensor("op_4527"), val = tensor([1, 8, 64, 188])]; + tensor var_4528_cast_fp16 = reshape(shape = var_4527, x = q_with_bias_v_23_cast_fp16)[name = tensor("op_4528_cast_fp16")]; + tensor var_4529 = const()[name = tensor("op_4529"), val = tensor([1, 8, 64, -1])]; + tensor var_4530_cast_fp16 = reshape(shape = var_4529, x = p_23_cast_fp16)[name = tensor("op_4530_cast_fp16")]; + tensor matrix_bd_89_transpose_x_0 = const()[name = tensor("matrix_bd_89_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_89_transpose_y_0 = const()[name = tensor("matrix_bd_89_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_89_cast_fp16 = matmul(transpose_x = matrix_bd_89_transpose_x_0, transpose_y = matrix_bd_89_transpose_y_0, x = var_4528_cast_fp16, y = var_4530_cast_fp16)[name = tensor("matrix_bd_89_cast_fp16")]; + tensor matrix_bd_91_pad_0 = const()[name = tensor("matrix_bd_91_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_91_mode_0 = const()[name = tensor("matrix_bd_91_mode_0"), val = tensor("constant")]; + tensor const_131_to_fp16 = const()[name = tensor("const_131_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_91_cast_fp16 = pad(constant_val = const_131_to_fp16, mode = matrix_bd_91_mode_0, pad = matrix_bd_91_pad_0, x = matrix_bd_89_cast_fp16)[name = tensor("matrix_bd_91_cast_fp16")]; + tensor var_4539 = const()[name = tensor("op_4539"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_93_cast_fp16 = reshape(shape = var_4539, x = matrix_bd_91_cast_fp16)[name = tensor("matrix_bd_93_cast_fp16")]; + tensor var_4543_begin_0 = const()[name = tensor("op_4543_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4543_end_0 = const()[name = tensor("op_4543_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4543_end_mask_0 = const()[name = tensor("op_4543_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4543_cast_fp16 = slice_by_index(begin = var_4543_begin_0, end = var_4543_end_0, end_mask = var_4543_end_mask_0, x = matrix_bd_93_cast_fp16)[name = tensor("op_4543_cast_fp16")]; + tensor var_4544 = const()[name = tensor("op_4544"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_95_cast_fp16 = reshape(shape = var_4544, x = var_4543_cast_fp16)[name = tensor("matrix_bd_95_cast_fp16")]; + tensor var_4549_begin_0 = const()[name = tensor("op_4549_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4549_end_0 = const()[name = tensor("op_4549_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4549_end_mask_0 = const()[name = tensor("op_4549_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4549_cast_fp16 = slice_by_index(begin = var_4549_begin_0, end = var_4549_end_0, end_mask = var_4549_end_mask_0, x = matrix_bd_95_cast_fp16)[name = tensor("op_4549_cast_fp16")]; + tensor var_4550_to_fp16 = const()[name = tensor("op_4550_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_23_cast_fp16 = mul(x = var_4549_cast_fp16, y = var_4550_to_fp16)[name = tensor("qk_mask_23_cast_fp16")]; + tensor var_4554 = const()[name = tensor("op_4554"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_23_cast_fp16 = reshape(shape = var_4554, x = query_47_cast_fp16)[name = tensor("mh_q_23_cast_fp16")]; + tensor var_4556_to_fp16 = const()[name = tensor("op_4556_to_fp16"), val = tensor(0x1p-3)]; + tensor var_4557_cast_fp16 = mul(x = mh_q_23_cast_fp16, y = var_4556_to_fp16)[name = tensor("op_4557_cast_fp16")]; + tensor var_4560 = const()[name = tensor("op_4560"), val = tensor([1, 8, 64, 188])]; + tensor var_4561_cast_fp16 = reshape(shape = var_4560, x = key_23_cast_fp16)[name = tensor("op_4561_cast_fp16")]; + tensor mh_w_45_transpose_x_0 = const()[name = tensor("mh_w_45_transpose_x_0"), val = tensor(true)]; + tensor mh_w_45_transpose_y_0 = const()[name = tensor("mh_w_45_transpose_y_0"), val = tensor(false)]; + tensor mh_w_45_cast_fp16 = matmul(transpose_x = mh_w_45_transpose_x_0, transpose_y = mh_w_45_transpose_y_0, x = var_4557_cast_fp16, y = var_4561_cast_fp16)[name = tensor("mh_w_45_cast_fp16")]; + tensor mh_w_47_cast_fp16 = add(x = mh_w_45_cast_fp16, y = qk_mask_23_cast_fp16)[name = tensor("mh_w_47_cast_fp16")]; + tensor var_4565_cast_fp16 = softmax(axis = var_4352, x = mh_w_47_cast_fp16)[name = tensor("op_4565_cast_fp16")]; + tensor var_4566 = const()[name = tensor("op_4566"), val = tensor([1, 8, 64, 188])]; + tensor var_4567_cast_fp16 = reshape(shape = var_4566, x = value_23_cast_fp16)[name = tensor("op_4567_cast_fp16")]; + tensor attn_23_transpose_x_0 = const()[name = tensor("attn_23_transpose_x_0"), val = tensor(false)]; + tensor attn_23_transpose_y_0 = const()[name = tensor("attn_23_transpose_y_0"), val = tensor(true)]; + tensor attn_23_cast_fp16 = matmul(transpose_x = attn_23_transpose_x_0, transpose_y = attn_23_transpose_y_0, x = var_4567_cast_fp16, y = var_4565_cast_fp16)[name = tensor("attn_23_cast_fp16")]; + tensor var_4570 = const()[name = tensor("op_4570"), val = tensor([1, 512, 1, 188])]; + tensor input_309_cast_fp16 = reshape(shape = var_4570, x = attn_23_cast_fp16)[name = tensor("input_309_cast_fp16")]; + tensor var_4580_pad_type_0 = const()[name = tensor("op_4580_pad_type_0"), val = tensor("valid")]; + tensor var_4580_strides_0 = const()[name = tensor("op_4580_strides_0"), val = tensor([1, 1])]; + tensor var_4580_pad_0 = const()[name = tensor("op_4580_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4580_dilations_0 = const()[name = tensor("op_4580_dilations_0"), val = tensor([1, 1])]; + tensor var_4580_groups_0 = const()[name = tensor("op_4580_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68211136))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68407808))), name = tensor("layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_11_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68408000)))]; + tensor var_4580_cast_fp16 = conv(bias = layers_11_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_4580_dilations_0, groups = var_4580_groups_0, pad = var_4580_pad_0, pad_type = var_4580_pad_type_0, strides = var_4580_strides_0, weight = layers_11_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_309_cast_fp16)[name = tensor("op_4580_cast_fp16")]; + tensor var_4586_pad_type_0 = const()[name = tensor("op_4586_pad_type_0"), val = tensor("valid")]; + tensor var_4586_strides_0 = const()[name = tensor("op_4586_strides_0"), val = tensor([1, 1])]; + tensor var_4586_pad_0 = const()[name = tensor("op_4586_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4586_dilations_0 = const()[name = tensor("op_4586_dilations_0"), val = tensor([1, 1])]; + tensor var_4586_groups_0 = const()[name = tensor("op_4586_groups_0"), val = tensor(1)]; + tensor layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68418048))), name = tensor("layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68409088))), shape = tensor([512, 512, 1, 1])]; + tensor var_4586_cast_fp16 = conv(dilations = var_4586_dilations_0, groups = var_4586_groups_0, pad = var_4586_pad_0, pad_type = var_4586_pad_type_0, strides = var_4586_strides_0, weight = layers_11_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_309_cast_fp16)[name = tensor("op_4586_cast_fp16")]; + tensor obj_49_cast_fp16 = add(x = var_4580_cast_fp16, y = var_4586_cast_fp16)[name = tensor("obj_49_cast_fp16")]; + tensor inputs_115_cast_fp16 = add(x = inputs_113_cast_fp16, y = obj_49_cast_fp16)[name = tensor("inputs_115_cast_fp16")]; + tensor out_115_axes_0 = const()[name = tensor("out_115_axes_0"), val = tensor([1])]; + tensor var_4597_to_fp16 = const()[name = tensor("op_4597_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_115_cast_fp16 = layer_norm(axes = out_115_axes_0, epsilon = var_4597_to_fp16, x = inputs_115_cast_fp16)[name = tensor("out_115_cast_fp16")]; + tensor input_311_gamma_0_to_fp16 = const()[name = tensor("input_311_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68450880)))]; + tensor input_311_beta_0_to_fp16 = const()[name = tensor("input_311_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68451968)))]; + tensor input_311_epsilon_0_to_fp16 = const()[name = tensor("input_311_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_311_cast_fp16 = batch_norm(beta = input_311_beta_0_to_fp16, epsilon = input_311_epsilon_0_to_fp16, gamma = input_311_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_115_cast_fp16)[name = tensor("input_311_cast_fp16")]; + tensor var_4619_pad_type_0 = const()[name = tensor("op_4619_pad_type_0"), val = tensor("valid")]; + tensor var_4619_strides_0 = const()[name = tensor("op_4619_strides_0"), val = tensor([1, 1])]; + tensor var_4619_pad_0 = const()[name = tensor("op_4619_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4619_dilations_0 = const()[name = tensor("op_4619_dilations_0"), val = tensor([1, 1])]; + tensor var_4619_groups_0 = const()[name = tensor("op_4619_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68453056))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68846336))), name = tensor("layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_11_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68846528)))]; + tensor var_4619_cast_fp16 = conv(bias = layers_11_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_4619_dilations_0, groups = var_4619_groups_0, pad = var_4619_pad_0, pad_type = var_4619_pad_type_0, strides = var_4619_strides_0, weight = layers_11_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_311_cast_fp16)[name = tensor("op_4619_cast_fp16")]; + tensor var_4625_pad_type_0 = const()[name = tensor("op_4625_pad_type_0"), val = tensor("valid")]; + tensor var_4625_strides_0 = const()[name = tensor("op_4625_strides_0"), val = tensor([1, 1])]; + tensor var_4625_pad_0 = const()[name = tensor("op_4625_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4625_dilations_0 = const()[name = tensor("op_4625_dilations_0"), val = tensor([1, 1])]; + tensor var_4625_groups_0 = const()[name = tensor("op_4625_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68866112))), name = tensor("layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68848640))), shape = tensor([1024, 512, 1, 1])]; + tensor var_4625_cast_fp16 = conv(dilations = var_4625_dilations_0, groups = var_4625_groups_0, pad = var_4625_pad_0, pad_type = var_4625_pad_type_0, strides = var_4625_strides_0, weight = layers_11_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_311_cast_fp16)[name = tensor("op_4625_cast_fp16")]; + tensor input_313_cast_fp16 = add(x = var_4619_cast_fp16, y = var_4625_cast_fp16)[name = tensor("input_313_cast_fp16")]; + tensor input_315_split_num_splits_0 = const()[name = tensor("input_315_split_num_splits_0"), val = tensor(2)]; + tensor input_315_split_axis_0 = const()[name = tensor("input_315_split_axis_0"), val = tensor(1)]; + tensor input_315_split_cast_fp16_0, tensor input_315_split_cast_fp16_1 = split(axis = input_315_split_axis_0, num_splits = input_315_split_num_splits_0, x = input_313_cast_fp16)[name = tensor("input_315_split_cast_fp16")]; + tensor input_315_split_1_sigmoid_cast_fp16 = sigmoid(x = input_315_split_cast_fp16_1)[name = tensor("input_315_split_1_sigmoid_cast_fp16")]; + tensor input_315_cast_fp16 = mul(x = input_315_split_cast_fp16_0, y = input_315_split_1_sigmoid_cast_fp16)[name = tensor("input_315_cast_fp16")]; + tensor input_317_pad_type_0 = const()[name = tensor("input_317_pad_type_0"), val = tensor("custom")]; + tensor input_317_pad_0 = const()[name = tensor("input_317_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_317_groups_0 = const()[name = tensor("input_317_groups_0"), val = tensor(512)]; + tensor input_317_strides_0 = const()[name = tensor("input_317_strides_0"), val = tensor([1, 1])]; + tensor input_317_dilations_0 = const()[name = tensor("input_317_dilations_0"), val = tensor([1, 1])]; + tensor const_213_to_fp16 = const()[name = tensor("const_213_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68931712)))]; + tensor const_214_to_fp16 = const()[name = tensor("const_214_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68940992)))]; + tensor input_319_cast_fp16 = conv(bias = const_214_to_fp16, dilations = input_317_dilations_0, groups = input_317_groups_0, pad = input_317_pad_0, pad_type = input_317_pad_type_0, strides = input_317_strides_0, weight = const_213_to_fp16, x = input_315_cast_fp16)[name = tensor("input_319_cast_fp16")]; + tensor input_321_cast_fp16 = silu(x = input_319_cast_fp16)[name = tensor("input_321_cast_fp16")]; + tensor var_4649_pad_type_0 = const()[name = tensor("op_4649_pad_type_0"), val = tensor("valid")]; + tensor var_4649_strides_0 = const()[name = tensor("op_4649_strides_0"), val = tensor([1, 1])]; + tensor var_4649_pad_0 = const()[name = tensor("op_4649_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4649_dilations_0 = const()[name = tensor("op_4649_dilations_0"), val = tensor([1, 1])]; + tensor var_4649_groups_0 = const()[name = tensor("op_4649_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(68942080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69138752))), name = tensor("layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_11_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69138944)))]; + tensor var_4649_cast_fp16 = conv(bias = layers_11_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_4649_dilations_0, groups = var_4649_groups_0, pad = var_4649_pad_0, pad_type = var_4649_pad_type_0, strides = var_4649_strides_0, weight = layers_11_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_321_cast_fp16)[name = tensor("op_4649_cast_fp16")]; + tensor var_4655_pad_type_0 = const()[name = tensor("op_4655_pad_type_0"), val = tensor("valid")]; + tensor var_4655_strides_0 = const()[name = tensor("op_4655_strides_0"), val = tensor([1, 1])]; + tensor var_4655_pad_0 = const()[name = tensor("op_4655_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4655_dilations_0 = const()[name = tensor("op_4655_dilations_0"), val = tensor([1, 1])]; + tensor var_4655_groups_0 = const()[name = tensor("op_4655_groups_0"), val = tensor(1)]; + tensor layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69149120))), name = tensor("layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69140032))), shape = tensor([512, 512, 1, 1])]; + tensor var_4655_cast_fp16 = conv(dilations = var_4655_dilations_0, groups = var_4655_groups_0, pad = var_4655_pad_0, pad_type = var_4655_pad_type_0, strides = var_4655_strides_0, weight = layers_11_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_321_cast_fp16)[name = tensor("op_4655_cast_fp16")]; + tensor x_71_cast_fp16 = add(x = var_4649_cast_fp16, y = var_4655_cast_fp16)[name = tensor("x_71_cast_fp16")]; + tensor inputs_117_cast_fp16 = add(x = inputs_115_cast_fp16, y = x_71_cast_fp16)[name = tensor("inputs_117_cast_fp16")]; + tensor out_117_axes_0 = const()[name = tensor("out_117_axes_0"), val = tensor([1])]; + tensor var_4666_to_fp16 = const()[name = tensor("op_4666_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_117_cast_fp16 = layer_norm(axes = out_117_axes_0, epsilon = var_4666_to_fp16, x = inputs_117_cast_fp16)[name = tensor("out_117_cast_fp16")]; + tensor input_323_gamma_0_to_fp16 = const()[name = tensor("input_323_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69181952)))]; + tensor input_323_beta_0_to_fp16 = const()[name = tensor("input_323_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69183040)))]; + tensor input_323_epsilon_0_to_fp16 = const()[name = tensor("input_323_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_323_cast_fp16 = batch_norm(beta = input_323_beta_0_to_fp16, epsilon = input_323_epsilon_0_to_fp16, gamma = input_323_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_117_cast_fp16)[name = tensor("input_323_cast_fp16")]; + tensor var_4686_pad_type_0 = const()[name = tensor("op_4686_pad_type_0"), val = tensor("valid")]; + tensor var_4686_strides_0 = const()[name = tensor("op_4686_strides_0"), val = tensor([1, 1])]; + tensor var_4686_pad_0 = const()[name = tensor("op_4686_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4686_dilations_0 = const()[name = tensor("op_4686_dilations_0"), val = tensor([1, 1])]; + tensor var_4686_groups_0 = const()[name = tensor("op_4686_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69184128))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69970624))), name = tensor("layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_11_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69970816)))]; + tensor var_4686_cast_fp16 = conv(bias = layers_11_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_4686_dilations_0, groups = var_4686_groups_0, pad = var_4686_pad_0, pad_type = var_4686_pad_type_0, strides = var_4686_strides_0, weight = layers_11_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_323_cast_fp16)[name = tensor("op_4686_cast_fp16")]; + tensor var_4692_pad_type_0 = const()[name = tensor("op_4692_pad_type_0"), val = tensor("valid")]; + tensor var_4692_strides_0 = const()[name = tensor("op_4692_strides_0"), val = tensor([1, 1])]; + tensor var_4692_pad_0 = const()[name = tensor("op_4692_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4692_dilations_0 = const()[name = tensor("op_4692_dilations_0"), val = tensor([1, 1])]; + tensor var_4692_groups_0 = const()[name = tensor("op_4692_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70009856))), name = tensor("layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(69974976))), shape = tensor([2048, 512, 1, 1])]; + tensor var_4692_cast_fp16 = conv(dilations = var_4692_dilations_0, groups = var_4692_groups_0, pad = var_4692_pad_0, pad_type = var_4692_pad_type_0, strides = var_4692_strides_0, weight = layers_11_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_323_cast_fp16)[name = tensor("op_4692_cast_fp16")]; + tensor input_325_cast_fp16 = add(x = var_4686_cast_fp16, y = var_4692_cast_fp16)[name = tensor("input_325_cast_fp16")]; + tensor input_327_cast_fp16 = silu(x = input_325_cast_fp16)[name = tensor("input_327_cast_fp16")]; + tensor var_4703_pad_type_0 = const()[name = tensor("op_4703_pad_type_0"), val = tensor("valid")]; + tensor var_4703_strides_0 = const()[name = tensor("op_4703_strides_0"), val = tensor([1, 1])]; + tensor var_4703_pad_0 = const()[name = tensor("op_4703_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4703_dilations_0 = const()[name = tensor("op_4703_dilations_0"), val = tensor([1, 1])]; + tensor var_4703_groups_0 = const()[name = tensor("op_4703_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70140992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70927488))), name = tensor("layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_11_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_11_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70927680)))]; + tensor var_4703_cast_fp16 = conv(bias = layers_11_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_4703_dilations_0, groups = var_4703_groups_0, pad = var_4703_pad_0, pad_type = var_4703_pad_type_0, strides = var_4703_strides_0, weight = layers_11_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_327_cast_fp16)[name = tensor("op_4703_cast_fp16")]; + tensor var_4709_pad_type_0 = const()[name = tensor("op_4709_pad_type_0"), val = tensor("valid")]; + tensor var_4709_strides_0 = const()[name = tensor("op_4709_strides_0"), val = tensor([1, 1])]; + tensor var_4709_pad_0 = const()[name = tensor("op_4709_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4709_dilations_0 = const()[name = tensor("op_4709_dilations_0"), val = tensor([1, 1])]; + tensor var_4709_groups_0 = const()[name = tensor("op_4709_groups_0"), val = tensor(1)]; + tensor layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70970688))), name = tensor("layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(70928768))), shape = tensor([512, 2048, 1, 1])]; + tensor var_4709_cast_fp16 = conv(dilations = var_4709_dilations_0, groups = var_4709_groups_0, pad = var_4709_pad_0, pad_type = var_4709_pad_type_0, strides = var_4709_strides_0, weight = layers_11_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_327_cast_fp16)[name = tensor("op_4709_cast_fp16")]; + tensor x_73_cast_fp16 = add(x = var_4703_cast_fp16, y = var_4709_cast_fp16)[name = tensor("x_73_cast_fp16")]; + tensor var_4711_to_fp16 = const()[name = tensor("op_4711_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4712_cast_fp16 = mul(x = x_73_cast_fp16, y = var_4711_to_fp16)[name = tensor("op_4712_cast_fp16")]; + tensor inputs_119_cast_fp16 = add(x = inputs_117_cast_fp16, y = var_4712_cast_fp16)[name = tensor("inputs_119_cast_fp16")]; + tensor out_119_axes_0 = const()[name = tensor("out_119_axes_0"), val = tensor([1])]; + tensor var_4722_to_fp16 = const()[name = tensor("op_4722_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_119_cast_fp16 = layer_norm(axes = out_119_axes_0, epsilon = var_4722_to_fp16, x = inputs_119_cast_fp16)[name = tensor("out_119_cast_fp16")]; + tensor inputs_121_gamma_0_to_fp16 = const()[name = tensor("inputs_121_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71101824)))]; + tensor inputs_121_beta_0_to_fp16 = const()[name = tensor("inputs_121_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71102912)))]; + tensor inputs_121_epsilon_0_to_fp16 = const()[name = tensor("inputs_121_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_121_cast_fp16 = batch_norm(beta = inputs_121_beta_0_to_fp16, epsilon = inputs_121_epsilon_0_to_fp16, gamma = inputs_121_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_119_cast_fp16)[name = tensor("inputs_121_cast_fp16")]; + tensor var_4736 = const()[name = tensor("op_4736"), val = tensor(3)]; + tensor out_121_axes_0 = const()[name = tensor("out_121_axes_0"), val = tensor([1])]; + tensor var_4767_to_fp16 = const()[name = tensor("op_4767_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_121_cast_fp16 = layer_norm(axes = out_121_axes_0, epsilon = var_4767_to_fp16, x = inputs_121_cast_fp16)[name = tensor("out_121_cast_fp16")]; + tensor input_329_gamma_0_to_fp16 = const()[name = tensor("input_329_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71104000)))]; + tensor input_329_beta_0_to_fp16 = const()[name = tensor("input_329_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71105088)))]; + tensor input_329_epsilon_0_to_fp16 = const()[name = tensor("input_329_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_329_cast_fp16 = batch_norm(beta = input_329_beta_0_to_fp16, epsilon = input_329_epsilon_0_to_fp16, gamma = input_329_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_121_cast_fp16)[name = tensor("input_329_cast_fp16")]; + tensor var_4787_pad_type_0 = const()[name = tensor("op_4787_pad_type_0"), val = tensor("valid")]; + tensor var_4787_strides_0 = const()[name = tensor("op_4787_strides_0"), val = tensor([1, 1])]; + tensor var_4787_pad_0 = const()[name = tensor("op_4787_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4787_dilations_0 = const()[name = tensor("op_4787_dilations_0"), val = tensor([1, 1])]; + tensor var_4787_groups_0 = const()[name = tensor("op_4787_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71106176))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71892672))), name = tensor("layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_12_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71892864)))]; + tensor var_4787_cast_fp16 = conv(bias = layers_12_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_4787_dilations_0, groups = var_4787_groups_0, pad = var_4787_pad_0, pad_type = var_4787_pad_type_0, strides = var_4787_strides_0, weight = layers_12_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_329_cast_fp16)[name = tensor("op_4787_cast_fp16")]; + tensor var_4793_pad_type_0 = const()[name = tensor("op_4793_pad_type_0"), val = tensor("valid")]; + tensor var_4793_strides_0 = const()[name = tensor("op_4793_strides_0"), val = tensor([1, 1])]; + tensor var_4793_pad_0 = const()[name = tensor("op_4793_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4793_dilations_0 = const()[name = tensor("op_4793_dilations_0"), val = tensor([1, 1])]; + tensor var_4793_groups_0 = const()[name = tensor("op_4793_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71931776))), name = tensor("layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(71897024))), shape = tensor([2048, 512, 1, 1])]; + tensor var_4793_cast_fp16 = conv(dilations = var_4793_dilations_0, groups = var_4793_groups_0, pad = var_4793_pad_0, pad_type = var_4793_pad_type_0, strides = var_4793_strides_0, weight = layers_12_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_329_cast_fp16)[name = tensor("op_4793_cast_fp16")]; + tensor input_331_cast_fp16 = add(x = var_4787_cast_fp16, y = var_4793_cast_fp16)[name = tensor("input_331_cast_fp16")]; + tensor input_333_cast_fp16 = silu(x = input_331_cast_fp16)[name = tensor("input_333_cast_fp16")]; + tensor var_4804_pad_type_0 = const()[name = tensor("op_4804_pad_type_0"), val = tensor("valid")]; + tensor var_4804_strides_0 = const()[name = tensor("op_4804_strides_0"), val = tensor([1, 1])]; + tensor var_4804_pad_0 = const()[name = tensor("op_4804_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4804_dilations_0 = const()[name = tensor("op_4804_dilations_0"), val = tensor([1, 1])]; + tensor var_4804_groups_0 = const()[name = tensor("op_4804_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72062912))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72849408))), name = tensor("layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_12_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72849600)))]; + tensor var_4804_cast_fp16 = conv(bias = layers_12_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_4804_dilations_0, groups = var_4804_groups_0, pad = var_4804_pad_0, pad_type = var_4804_pad_type_0, strides = var_4804_strides_0, weight = layers_12_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_333_cast_fp16)[name = tensor("op_4804_cast_fp16")]; + tensor var_4810_pad_type_0 = const()[name = tensor("op_4810_pad_type_0"), val = tensor("valid")]; + tensor var_4810_strides_0 = const()[name = tensor("op_4810_strides_0"), val = tensor([1, 1])]; + tensor var_4810_pad_0 = const()[name = tensor("op_4810_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4810_dilations_0 = const()[name = tensor("op_4810_dilations_0"), val = tensor([1, 1])]; + tensor var_4810_groups_0 = const()[name = tensor("op_4810_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72890432))), name = tensor("layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(72850688))), shape = tensor([512, 2048, 1, 1])]; + tensor var_4810_cast_fp16 = conv(dilations = var_4810_dilations_0, groups = var_4810_groups_0, pad = var_4810_pad_0, pad_type = var_4810_pad_type_0, strides = var_4810_strides_0, weight = layers_12_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_333_cast_fp16)[name = tensor("op_4810_cast_fp16")]; + tensor x_75_cast_fp16 = add(x = var_4804_cast_fp16, y = var_4810_cast_fp16)[name = tensor("x_75_cast_fp16")]; + tensor var_4812_to_fp16 = const()[name = tensor("op_4812_to_fp16"), val = tensor(0x1p-1)]; + tensor var_4813_cast_fp16 = mul(x = x_75_cast_fp16, y = var_4812_to_fp16)[name = tensor("op_4813_cast_fp16")]; + tensor inputs_123_cast_fp16 = add(x = inputs_121_cast_fp16, y = var_4813_cast_fp16)[name = tensor("inputs_123_cast_fp16")]; + tensor out_123_axes_0 = const()[name = tensor("out_123_axes_0"), val = tensor([1])]; + tensor var_4823_to_fp16 = const()[name = tensor("op_4823_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_123_cast_fp16 = layer_norm(axes = out_123_axes_0, epsilon = var_4823_to_fp16, x = inputs_123_cast_fp16)[name = tensor("out_123_cast_fp16")]; + tensor obj_51_gamma_0_to_fp16 = const()[name = tensor("obj_51_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73021568)))]; + tensor obj_51_beta_0_to_fp16 = const()[name = tensor("obj_51_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73022656)))]; + tensor obj_51_epsilon_0_to_fp16 = const()[name = tensor("obj_51_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_51_cast_fp16 = batch_norm(beta = obj_51_beta_0_to_fp16, epsilon = obj_51_epsilon_0_to_fp16, gamma = obj_51_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_123_cast_fp16)[name = tensor("obj_51_cast_fp16")]; + tensor var_4848_pad_type_0 = const()[name = tensor("op_4848_pad_type_0"), val = tensor("valid")]; + tensor var_4848_strides_0 = const()[name = tensor("op_4848_strides_0"), val = tensor([1, 1])]; + tensor var_4848_pad_0 = const()[name = tensor("op_4848_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4848_dilations_0 = const()[name = tensor("op_4848_dilations_0"), val = tensor([1, 1])]; + tensor var_4848_groups_0 = const()[name = tensor("op_4848_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73023744))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73220416))), name = tensor("layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_12_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73220608)))]; + tensor var_4848_cast_fp16 = conv(bias = layers_12_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_4848_dilations_0, groups = var_4848_groups_0, pad = var_4848_pad_0, pad_type = var_4848_pad_type_0, strides = var_4848_strides_0, weight = layers_12_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4848_cast_fp16")]; + tensor var_4854_pad_type_0 = const()[name = tensor("op_4854_pad_type_0"), val = tensor("valid")]; + tensor var_4854_strides_0 = const()[name = tensor("op_4854_strides_0"), val = tensor([1, 1])]; + tensor var_4854_pad_0 = const()[name = tensor("op_4854_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4854_dilations_0 = const()[name = tensor("op_4854_dilations_0"), val = tensor([1, 1])]; + tensor var_4854_groups_0 = const()[name = tensor("op_4854_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73230784))), name = tensor("layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73221696))), shape = tensor([512, 512, 1, 1])]; + tensor var_4854_cast_fp16 = conv(dilations = var_4854_dilations_0, groups = var_4854_groups_0, pad = var_4854_pad_0, pad_type = var_4854_pad_type_0, strides = var_4854_strides_0, weight = layers_12_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4854_cast_fp16")]; + tensor query_49_cast_fp16 = add(x = var_4848_cast_fp16, y = var_4854_cast_fp16)[name = tensor("query_49_cast_fp16")]; + tensor var_4863_pad_type_0 = const()[name = tensor("op_4863_pad_type_0"), val = tensor("valid")]; + tensor var_4863_strides_0 = const()[name = tensor("op_4863_strides_0"), val = tensor([1, 1])]; + tensor var_4863_pad_0 = const()[name = tensor("op_4863_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4863_dilations_0 = const()[name = tensor("op_4863_dilations_0"), val = tensor([1, 1])]; + tensor var_4863_groups_0 = const()[name = tensor("op_4863_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73263616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73460288))), name = tensor("layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4863_cast_fp16 = conv(dilations = var_4863_dilations_0, groups = var_4863_groups_0, pad = var_4863_pad_0, pad_type = var_4863_pad_type_0, strides = var_4863_strides_0, weight = layers_12_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4863_cast_fp16")]; + tensor var_4869_pad_type_0 = const()[name = tensor("op_4869_pad_type_0"), val = tensor("valid")]; + tensor var_4869_strides_0 = const()[name = tensor("op_4869_strides_0"), val = tensor([1, 1])]; + tensor var_4869_pad_0 = const()[name = tensor("op_4869_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4869_dilations_0 = const()[name = tensor("op_4869_dilations_0"), val = tensor([1, 1])]; + tensor var_4869_groups_0 = const()[name = tensor("op_4869_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73468160))), name = tensor("layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73460480))), shape = tensor([512, 512, 1, 1])]; + tensor var_4869_cast_fp16 = conv(dilations = var_4869_dilations_0, groups = var_4869_groups_0, pad = var_4869_pad_0, pad_type = var_4869_pad_type_0, strides = var_4869_strides_0, weight = layers_12_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4869_cast_fp16")]; + tensor key_25_cast_fp16 = add(x = var_4863_cast_fp16, y = var_4869_cast_fp16)[name = tensor("key_25_cast_fp16")]; + tensor var_4879_pad_type_0 = const()[name = tensor("op_4879_pad_type_0"), val = tensor("valid")]; + tensor var_4879_strides_0 = const()[name = tensor("op_4879_strides_0"), val = tensor([1, 1])]; + tensor var_4879_pad_0 = const()[name = tensor("op_4879_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4879_dilations_0 = const()[name = tensor("op_4879_dilations_0"), val = tensor([1, 1])]; + tensor var_4879_groups_0 = const()[name = tensor("op_4879_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73500992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73697664))), name = tensor("layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_12_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73697856)))]; + tensor var_4879_cast_fp16 = conv(bias = layers_12_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_4879_dilations_0, groups = var_4879_groups_0, pad = var_4879_pad_0, pad_type = var_4879_pad_type_0, strides = var_4879_strides_0, weight = layers_12_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_51_cast_fp16)[name = tensor("op_4879_cast_fp16")]; + tensor var_4885_pad_type_0 = const()[name = tensor("op_4885_pad_type_0"), val = tensor("valid")]; + tensor var_4885_strides_0 = const()[name = tensor("op_4885_strides_0"), val = tensor([1, 1])]; + tensor var_4885_pad_0 = const()[name = tensor("op_4885_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4885_dilations_0 = const()[name = tensor("op_4885_dilations_0"), val = tensor([1, 1])]; + tensor var_4885_groups_0 = const()[name = tensor("op_4885_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73707712))), name = tensor("layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73698944))), shape = tensor([512, 512, 1, 1])]; + tensor var_4885_cast_fp16 = conv(dilations = var_4885_dilations_0, groups = var_4885_groups_0, pad = var_4885_pad_0, pad_type = var_4885_pad_type_0, strides = var_4885_strides_0, weight = layers_12_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_51_cast_fp16)[name = tensor("op_4885_cast_fp16")]; + tensor value_25_cast_fp16 = add(x = var_4879_cast_fp16, y = var_4885_cast_fp16)[name = tensor("value_25_cast_fp16")]; + tensor var_4888_to_fp16 = const()[name = tensor("op_4888_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73740544)))]; + tensor query_51_cast_fp16 = add(x = query_49_cast_fp16, y = var_4888_to_fp16)[name = tensor("query_51_cast_fp16")]; + tensor var_4891_to_fp16 = const()[name = tensor("op_4891_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73741632)))]; + tensor q_with_bias_v_25_cast_fp16 = add(x = query_49_cast_fp16, y = var_4891_to_fp16)[name = tensor("q_with_bias_v_25_cast_fp16")]; + tensor var_4901_pad_type_0 = const()[name = tensor("op_4901_pad_type_0"), val = tensor("valid")]; + tensor var_4901_strides_0 = const()[name = tensor("op_4901_strides_0"), val = tensor([1, 1])]; + tensor var_4901_pad_0 = const()[name = tensor("op_4901_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4901_dilations_0 = const()[name = tensor("op_4901_dilations_0"), val = tensor([1, 1])]; + tensor var_4901_groups_0 = const()[name = tensor("op_4901_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73742720))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73939392))), name = tensor("layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_4901_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_4901_dilations_0, groups = var_4901_groups_0, pad = var_4901_pad_0, pad_type = var_4901_pad_type_0, strides = var_4901_strides_0, weight = layers_12_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_4901_cast_fp16")]; + tensor var_4907_pad_type_0 = const()[name = tensor("op_4907_pad_type_0"), val = tensor("valid")]; + tensor var_4907_strides_0 = const()[name = tensor("op_4907_strides_0"), val = tensor([1, 1])]; + tensor var_4907_pad_0 = const()[name = tensor("op_4907_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4907_dilations_0 = const()[name = tensor("op_4907_dilations_0"), val = tensor([1, 1])]; + tensor var_4907_groups_0 = const()[name = tensor("op_4907_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73961280))), name = tensor("layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73939584))), shape = tensor([512, 512, 1, 1])]; + tensor var_4907_cast_fp16 = conv(dilations = var_4907_dilations_0, groups = var_4907_groups_0, pad = var_4907_pad_0, pad_type = var_4907_pad_type_0, strides = var_4907_strides_0, weight = layers_12_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_4907_cast_fp16")]; + tensor p_25_cast_fp16 = add(x = var_4901_cast_fp16, y = var_4907_cast_fp16)[name = tensor("p_25_cast_fp16")]; + tensor var_4911 = const()[name = tensor("op_4911"), val = tensor([1, 8, 64, 188])]; + tensor var_4912_cast_fp16 = reshape(shape = var_4911, x = q_with_bias_v_25_cast_fp16)[name = tensor("op_4912_cast_fp16")]; + tensor var_4913 = const()[name = tensor("op_4913"), val = tensor([1, 8, 64, -1])]; + tensor var_4914_cast_fp16 = reshape(shape = var_4913, x = p_25_cast_fp16)[name = tensor("op_4914_cast_fp16")]; + tensor matrix_bd_97_transpose_x_0 = const()[name = tensor("matrix_bd_97_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_97_transpose_y_0 = const()[name = tensor("matrix_bd_97_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_97_cast_fp16 = matmul(transpose_x = matrix_bd_97_transpose_x_0, transpose_y = matrix_bd_97_transpose_y_0, x = var_4912_cast_fp16, y = var_4914_cast_fp16)[name = tensor("matrix_bd_97_cast_fp16")]; + tensor matrix_bd_99_pad_0 = const()[name = tensor("matrix_bd_99_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_99_mode_0 = const()[name = tensor("matrix_bd_99_mode_0"), val = tensor("constant")]; + tensor const_142_to_fp16 = const()[name = tensor("const_142_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_99_cast_fp16 = pad(constant_val = const_142_to_fp16, mode = matrix_bd_99_mode_0, pad = matrix_bd_99_pad_0, x = matrix_bd_97_cast_fp16)[name = tensor("matrix_bd_99_cast_fp16")]; + tensor var_4923 = const()[name = tensor("op_4923"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_101_cast_fp16 = reshape(shape = var_4923, x = matrix_bd_99_cast_fp16)[name = tensor("matrix_bd_101_cast_fp16")]; + tensor var_4927_begin_0 = const()[name = tensor("op_4927_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_4927_end_0 = const()[name = tensor("op_4927_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_4927_end_mask_0 = const()[name = tensor("op_4927_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_4927_cast_fp16 = slice_by_index(begin = var_4927_begin_0, end = var_4927_end_0, end_mask = var_4927_end_mask_0, x = matrix_bd_101_cast_fp16)[name = tensor("op_4927_cast_fp16")]; + tensor var_4928 = const()[name = tensor("op_4928"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_103_cast_fp16 = reshape(shape = var_4928, x = var_4927_cast_fp16)[name = tensor("matrix_bd_103_cast_fp16")]; + tensor var_4933_begin_0 = const()[name = tensor("op_4933_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4933_end_0 = const()[name = tensor("op_4933_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_4933_end_mask_0 = const()[name = tensor("op_4933_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_4933_cast_fp16 = slice_by_index(begin = var_4933_begin_0, end = var_4933_end_0, end_mask = var_4933_end_mask_0, x = matrix_bd_103_cast_fp16)[name = tensor("op_4933_cast_fp16")]; + tensor var_4934_to_fp16 = const()[name = tensor("op_4934_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_25_cast_fp16 = mul(x = var_4933_cast_fp16, y = var_4934_to_fp16)[name = tensor("qk_mask_25_cast_fp16")]; + tensor var_4938 = const()[name = tensor("op_4938"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_25_cast_fp16 = reshape(shape = var_4938, x = query_51_cast_fp16)[name = tensor("mh_q_25_cast_fp16")]; + tensor var_4940_to_fp16 = const()[name = tensor("op_4940_to_fp16"), val = tensor(0x1p-3)]; + tensor var_4941_cast_fp16 = mul(x = mh_q_25_cast_fp16, y = var_4940_to_fp16)[name = tensor("op_4941_cast_fp16")]; + tensor var_4944 = const()[name = tensor("op_4944"), val = tensor([1, 8, 64, 188])]; + tensor var_4945_cast_fp16 = reshape(shape = var_4944, x = key_25_cast_fp16)[name = tensor("op_4945_cast_fp16")]; + tensor mh_w_49_transpose_x_0 = const()[name = tensor("mh_w_49_transpose_x_0"), val = tensor(true)]; + tensor mh_w_49_transpose_y_0 = const()[name = tensor("mh_w_49_transpose_y_0"), val = tensor(false)]; + tensor mh_w_49_cast_fp16 = matmul(transpose_x = mh_w_49_transpose_x_0, transpose_y = mh_w_49_transpose_y_0, x = var_4941_cast_fp16, y = var_4945_cast_fp16)[name = tensor("mh_w_49_cast_fp16")]; + tensor mh_w_51_cast_fp16 = add(x = mh_w_49_cast_fp16, y = qk_mask_25_cast_fp16)[name = tensor("mh_w_51_cast_fp16")]; + tensor var_4949_cast_fp16 = softmax(axis = var_4736, x = mh_w_51_cast_fp16)[name = tensor("op_4949_cast_fp16")]; + tensor var_4950 = const()[name = tensor("op_4950"), val = tensor([1, 8, 64, 188])]; + tensor var_4951_cast_fp16 = reshape(shape = var_4950, x = value_25_cast_fp16)[name = tensor("op_4951_cast_fp16")]; + tensor attn_25_transpose_x_0 = const()[name = tensor("attn_25_transpose_x_0"), val = tensor(false)]; + tensor attn_25_transpose_y_0 = const()[name = tensor("attn_25_transpose_y_0"), val = tensor(true)]; + tensor attn_25_cast_fp16 = matmul(transpose_x = attn_25_transpose_x_0, transpose_y = attn_25_transpose_y_0, x = var_4951_cast_fp16, y = var_4949_cast_fp16)[name = tensor("attn_25_cast_fp16")]; + tensor var_4954 = const()[name = tensor("op_4954"), val = tensor([1, 512, 1, 188])]; + tensor input_335_cast_fp16 = reshape(shape = var_4954, x = attn_25_cast_fp16)[name = tensor("input_335_cast_fp16")]; + tensor var_4964_pad_type_0 = const()[name = tensor("op_4964_pad_type_0"), val = tensor("valid")]; + tensor var_4964_strides_0 = const()[name = tensor("op_4964_strides_0"), val = tensor([1, 1])]; + tensor var_4964_pad_0 = const()[name = tensor("op_4964_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4964_dilations_0 = const()[name = tensor("op_4964_dilations_0"), val = tensor([1, 1])]; + tensor var_4964_groups_0 = const()[name = tensor("op_4964_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(73994112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74190784))), name = tensor("layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_12_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74190976)))]; + tensor var_4964_cast_fp16 = conv(bias = layers_12_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_4964_dilations_0, groups = var_4964_groups_0, pad = var_4964_pad_0, pad_type = var_4964_pad_type_0, strides = var_4964_strides_0, weight = layers_12_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_335_cast_fp16)[name = tensor("op_4964_cast_fp16")]; + tensor var_4970_pad_type_0 = const()[name = tensor("op_4970_pad_type_0"), val = tensor("valid")]; + tensor var_4970_strides_0 = const()[name = tensor("op_4970_strides_0"), val = tensor([1, 1])]; + tensor var_4970_pad_0 = const()[name = tensor("op_4970_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_4970_dilations_0 = const()[name = tensor("op_4970_dilations_0"), val = tensor([1, 1])]; + tensor var_4970_groups_0 = const()[name = tensor("op_4970_groups_0"), val = tensor(1)]; + tensor layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74201984))), name = tensor("layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74192064))), shape = tensor([512, 512, 1, 1])]; + tensor var_4970_cast_fp16 = conv(dilations = var_4970_dilations_0, groups = var_4970_groups_0, pad = var_4970_pad_0, pad_type = var_4970_pad_type_0, strides = var_4970_strides_0, weight = layers_12_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_335_cast_fp16)[name = tensor("op_4970_cast_fp16")]; + tensor obj_53_cast_fp16 = add(x = var_4964_cast_fp16, y = var_4970_cast_fp16)[name = tensor("obj_53_cast_fp16")]; + tensor inputs_125_cast_fp16 = add(x = inputs_123_cast_fp16, y = obj_53_cast_fp16)[name = tensor("inputs_125_cast_fp16")]; + tensor out_125_axes_0 = const()[name = tensor("out_125_axes_0"), val = tensor([1])]; + tensor var_4981_to_fp16 = const()[name = tensor("op_4981_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_125_cast_fp16 = layer_norm(axes = out_125_axes_0, epsilon = var_4981_to_fp16, x = inputs_125_cast_fp16)[name = tensor("out_125_cast_fp16")]; + tensor input_337_gamma_0_to_fp16 = const()[name = tensor("input_337_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74234816)))]; + tensor input_337_beta_0_to_fp16 = const()[name = tensor("input_337_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74235904)))]; + tensor input_337_epsilon_0_to_fp16 = const()[name = tensor("input_337_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_337_cast_fp16 = batch_norm(beta = input_337_beta_0_to_fp16, epsilon = input_337_epsilon_0_to_fp16, gamma = input_337_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_125_cast_fp16)[name = tensor("input_337_cast_fp16")]; + tensor var_5003_pad_type_0 = const()[name = tensor("op_5003_pad_type_0"), val = tensor("valid")]; + tensor var_5003_strides_0 = const()[name = tensor("op_5003_strides_0"), val = tensor([1, 1])]; + tensor var_5003_pad_0 = const()[name = tensor("op_5003_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5003_dilations_0 = const()[name = tensor("op_5003_dilations_0"), val = tensor([1, 1])]; + tensor var_5003_groups_0 = const()[name = tensor("op_5003_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74236992))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74630272))), name = tensor("layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_12_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74630464)))]; + tensor var_5003_cast_fp16 = conv(bias = layers_12_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_5003_dilations_0, groups = var_5003_groups_0, pad = var_5003_pad_0, pad_type = var_5003_pad_type_0, strides = var_5003_strides_0, weight = layers_12_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_337_cast_fp16)[name = tensor("op_5003_cast_fp16")]; + tensor var_5009_pad_type_0 = const()[name = tensor("op_5009_pad_type_0"), val = tensor("valid")]; + tensor var_5009_strides_0 = const()[name = tensor("op_5009_strides_0"), val = tensor([1, 1])]; + tensor var_5009_pad_0 = const()[name = tensor("op_5009_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5009_dilations_0 = const()[name = tensor("op_5009_dilations_0"), val = tensor([1, 1])]; + tensor var_5009_groups_0 = const()[name = tensor("op_5009_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74651328))), name = tensor("layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74632576))), shape = tensor([1024, 512, 1, 1])]; + tensor var_5009_cast_fp16 = conv(dilations = var_5009_dilations_0, groups = var_5009_groups_0, pad = var_5009_pad_0, pad_type = var_5009_pad_type_0, strides = var_5009_strides_0, weight = layers_12_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_337_cast_fp16)[name = tensor("op_5009_cast_fp16")]; + tensor input_339_cast_fp16 = add(x = var_5003_cast_fp16, y = var_5009_cast_fp16)[name = tensor("input_339_cast_fp16")]; + tensor input_341_split_num_splits_0 = const()[name = tensor("input_341_split_num_splits_0"), val = tensor(2)]; + tensor input_341_split_axis_0 = const()[name = tensor("input_341_split_axis_0"), val = tensor(1)]; + tensor input_341_split_cast_fp16_0, tensor input_341_split_cast_fp16_1 = split(axis = input_341_split_axis_0, num_splits = input_341_split_num_splits_0, x = input_339_cast_fp16)[name = tensor("input_341_split_cast_fp16")]; + tensor input_341_split_1_sigmoid_cast_fp16 = sigmoid(x = input_341_split_cast_fp16_1)[name = tensor("input_341_split_1_sigmoid_cast_fp16")]; + tensor input_341_cast_fp16 = mul(x = input_341_split_cast_fp16_0, y = input_341_split_1_sigmoid_cast_fp16)[name = tensor("input_341_cast_fp16")]; + tensor input_343_pad_type_0 = const()[name = tensor("input_343_pad_type_0"), val = tensor("custom")]; + tensor input_343_pad_0 = const()[name = tensor("input_343_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_343_groups_0 = const()[name = tensor("input_343_groups_0"), val = tensor(512)]; + tensor input_343_strides_0 = const()[name = tensor("input_343_strides_0"), val = tensor([1, 1])]; + tensor input_343_dilations_0 = const()[name = tensor("input_343_dilations_0"), val = tensor([1, 1])]; + tensor const_215_to_fp16 = const()[name = tensor("const_215_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74716928)))]; + tensor const_216_to_fp16 = const()[name = tensor("const_216_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74726208)))]; + tensor input_345_cast_fp16 = conv(bias = const_216_to_fp16, dilations = input_343_dilations_0, groups = input_343_groups_0, pad = input_343_pad_0, pad_type = input_343_pad_type_0, strides = input_343_strides_0, weight = const_215_to_fp16, x = input_341_cast_fp16)[name = tensor("input_345_cast_fp16")]; + tensor input_347_cast_fp16 = silu(x = input_345_cast_fp16)[name = tensor("input_347_cast_fp16")]; + tensor var_5033_pad_type_0 = const()[name = tensor("op_5033_pad_type_0"), val = tensor("valid")]; + tensor var_5033_strides_0 = const()[name = tensor("op_5033_strides_0"), val = tensor([1, 1])]; + tensor var_5033_pad_0 = const()[name = tensor("op_5033_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5033_dilations_0 = const()[name = tensor("op_5033_dilations_0"), val = tensor([1, 1])]; + tensor var_5033_groups_0 = const()[name = tensor("op_5033_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74727296))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74923968))), name = tensor("layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_12_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74924160)))]; + tensor var_5033_cast_fp16 = conv(bias = layers_12_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_5033_dilations_0, groups = var_5033_groups_0, pad = var_5033_pad_0, pad_type = var_5033_pad_type_0, strides = var_5033_strides_0, weight = layers_12_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_347_cast_fp16)[name = tensor("op_5033_cast_fp16")]; + tensor var_5039_pad_type_0 = const()[name = tensor("op_5039_pad_type_0"), val = tensor("valid")]; + tensor var_5039_strides_0 = const()[name = tensor("op_5039_strides_0"), val = tensor([1, 1])]; + tensor var_5039_pad_0 = const()[name = tensor("op_5039_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5039_dilations_0 = const()[name = tensor("op_5039_dilations_0"), val = tensor([1, 1])]; + tensor var_5039_groups_0 = const()[name = tensor("op_5039_groups_0"), val = tensor(1)]; + tensor layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74934784))), name = tensor("layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74925248))), shape = tensor([512, 512, 1, 1])]; + tensor var_5039_cast_fp16 = conv(dilations = var_5039_dilations_0, groups = var_5039_groups_0, pad = var_5039_pad_0, pad_type = var_5039_pad_type_0, strides = var_5039_strides_0, weight = layers_12_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_347_cast_fp16)[name = tensor("op_5039_cast_fp16")]; + tensor x_77_cast_fp16 = add(x = var_5033_cast_fp16, y = var_5039_cast_fp16)[name = tensor("x_77_cast_fp16")]; + tensor inputs_127_cast_fp16 = add(x = inputs_125_cast_fp16, y = x_77_cast_fp16)[name = tensor("inputs_127_cast_fp16")]; + tensor out_127_axes_0 = const()[name = tensor("out_127_axes_0"), val = tensor([1])]; + tensor var_5050_to_fp16 = const()[name = tensor("op_5050_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_127_cast_fp16 = layer_norm(axes = out_127_axes_0, epsilon = var_5050_to_fp16, x = inputs_127_cast_fp16)[name = tensor("out_127_cast_fp16")]; + tensor input_349_gamma_0_to_fp16 = const()[name = tensor("input_349_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74967616)))]; + tensor input_349_beta_0_to_fp16 = const()[name = tensor("input_349_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74968704)))]; + tensor input_349_epsilon_0_to_fp16 = const()[name = tensor("input_349_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_349_cast_fp16 = batch_norm(beta = input_349_beta_0_to_fp16, epsilon = input_349_epsilon_0_to_fp16, gamma = input_349_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_127_cast_fp16)[name = tensor("input_349_cast_fp16")]; + tensor var_5070_pad_type_0 = const()[name = tensor("op_5070_pad_type_0"), val = tensor("valid")]; + tensor var_5070_strides_0 = const()[name = tensor("op_5070_strides_0"), val = tensor([1, 1])]; + tensor var_5070_pad_0 = const()[name = tensor("op_5070_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5070_dilations_0 = const()[name = tensor("op_5070_dilations_0"), val = tensor([1, 1])]; + tensor var_5070_groups_0 = const()[name = tensor("op_5070_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(74969792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75756288))), name = tensor("layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_12_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75756480)))]; + tensor var_5070_cast_fp16 = conv(bias = layers_12_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_5070_dilations_0, groups = var_5070_groups_0, pad = var_5070_pad_0, pad_type = var_5070_pad_type_0, strides = var_5070_strides_0, weight = layers_12_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_349_cast_fp16)[name = tensor("op_5070_cast_fp16")]; + tensor var_5076_pad_type_0 = const()[name = tensor("op_5076_pad_type_0"), val = tensor("valid")]; + tensor var_5076_strides_0 = const()[name = tensor("op_5076_strides_0"), val = tensor([1, 1])]; + tensor var_5076_pad_0 = const()[name = tensor("op_5076_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5076_dilations_0 = const()[name = tensor("op_5076_dilations_0"), val = tensor([1, 1])]; + tensor var_5076_groups_0 = const()[name = tensor("op_5076_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75797184))), name = tensor("layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75760640))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5076_cast_fp16 = conv(dilations = var_5076_dilations_0, groups = var_5076_groups_0, pad = var_5076_pad_0, pad_type = var_5076_pad_type_0, strides = var_5076_strides_0, weight = layers_12_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_349_cast_fp16)[name = tensor("op_5076_cast_fp16")]; + tensor input_351_cast_fp16 = add(x = var_5070_cast_fp16, y = var_5076_cast_fp16)[name = tensor("input_351_cast_fp16")]; + tensor input_353_cast_fp16 = silu(x = input_351_cast_fp16)[name = tensor("input_353_cast_fp16")]; + tensor var_5087_pad_type_0 = const()[name = tensor("op_5087_pad_type_0"), val = tensor("valid")]; + tensor var_5087_strides_0 = const()[name = tensor("op_5087_strides_0"), val = tensor([1, 1])]; + tensor var_5087_pad_0 = const()[name = tensor("op_5087_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5087_dilations_0 = const()[name = tensor("op_5087_dilations_0"), val = tensor([1, 1])]; + tensor var_5087_groups_0 = const()[name = tensor("op_5087_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(75928320))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76714816))), name = tensor("layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_12_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_12_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76715008)))]; + tensor var_5087_cast_fp16 = conv(bias = layers_12_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_5087_dilations_0, groups = var_5087_groups_0, pad = var_5087_pad_0, pad_type = var_5087_pad_type_0, strides = var_5087_strides_0, weight = layers_12_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_353_cast_fp16)[name = tensor("op_5087_cast_fp16")]; + tensor var_5093_pad_type_0 = const()[name = tensor("op_5093_pad_type_0"), val = tensor("valid")]; + tensor var_5093_strides_0 = const()[name = tensor("op_5093_strides_0"), val = tensor([1, 1])]; + tensor var_5093_pad_0 = const()[name = tensor("op_5093_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5093_dilations_0 = const()[name = tensor("op_5093_dilations_0"), val = tensor([1, 1])]; + tensor var_5093_groups_0 = const()[name = tensor("op_5093_groups_0"), val = tensor(1)]; + tensor layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76758592))), name = tensor("layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76716096))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5093_cast_fp16 = conv(dilations = var_5093_dilations_0, groups = var_5093_groups_0, pad = var_5093_pad_0, pad_type = var_5093_pad_type_0, strides = var_5093_strides_0, weight = layers_12_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_353_cast_fp16)[name = tensor("op_5093_cast_fp16")]; + tensor x_79_cast_fp16 = add(x = var_5087_cast_fp16, y = var_5093_cast_fp16)[name = tensor("x_79_cast_fp16")]; + tensor var_5095_to_fp16 = const()[name = tensor("op_5095_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5096_cast_fp16 = mul(x = x_79_cast_fp16, y = var_5095_to_fp16)[name = tensor("op_5096_cast_fp16")]; + tensor inputs_129_cast_fp16 = add(x = inputs_127_cast_fp16, y = var_5096_cast_fp16)[name = tensor("inputs_129_cast_fp16")]; + tensor out_129_axes_0 = const()[name = tensor("out_129_axes_0"), val = tensor([1])]; + tensor var_5106_to_fp16 = const()[name = tensor("op_5106_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_129_cast_fp16 = layer_norm(axes = out_129_axes_0, epsilon = var_5106_to_fp16, x = inputs_129_cast_fp16)[name = tensor("out_129_cast_fp16")]; + tensor inputs_131_gamma_0_to_fp16 = const()[name = tensor("inputs_131_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76889728)))]; + tensor inputs_131_beta_0_to_fp16 = const()[name = tensor("inputs_131_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76890816)))]; + tensor inputs_131_epsilon_0_to_fp16 = const()[name = tensor("inputs_131_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_131_cast_fp16 = batch_norm(beta = inputs_131_beta_0_to_fp16, epsilon = inputs_131_epsilon_0_to_fp16, gamma = inputs_131_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_129_cast_fp16)[name = tensor("inputs_131_cast_fp16")]; + tensor var_5120 = const()[name = tensor("op_5120"), val = tensor(3)]; + tensor out_131_axes_0 = const()[name = tensor("out_131_axes_0"), val = tensor([1])]; + tensor var_5151_to_fp16 = const()[name = tensor("op_5151_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_131_cast_fp16 = layer_norm(axes = out_131_axes_0, epsilon = var_5151_to_fp16, x = inputs_131_cast_fp16)[name = tensor("out_131_cast_fp16")]; + tensor input_355_gamma_0_to_fp16 = const()[name = tensor("input_355_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76891904)))]; + tensor input_355_beta_0_to_fp16 = const()[name = tensor("input_355_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76892992)))]; + tensor input_355_epsilon_0_to_fp16 = const()[name = tensor("input_355_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_355_cast_fp16 = batch_norm(beta = input_355_beta_0_to_fp16, epsilon = input_355_epsilon_0_to_fp16, gamma = input_355_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_131_cast_fp16)[name = tensor("input_355_cast_fp16")]; + tensor var_5171_pad_type_0 = const()[name = tensor("op_5171_pad_type_0"), val = tensor("valid")]; + tensor var_5171_strides_0 = const()[name = tensor("op_5171_strides_0"), val = tensor([1, 1])]; + tensor var_5171_pad_0 = const()[name = tensor("op_5171_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5171_dilations_0 = const()[name = tensor("op_5171_dilations_0"), val = tensor([1, 1])]; + tensor var_5171_groups_0 = const()[name = tensor("op_5171_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(76894080))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77680576))), name = tensor("layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_13_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77680768)))]; + tensor var_5171_cast_fp16 = conv(bias = layers_13_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5171_dilations_0, groups = var_5171_groups_0, pad = var_5171_pad_0, pad_type = var_5171_pad_type_0, strides = var_5171_strides_0, weight = layers_13_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_355_cast_fp16)[name = tensor("op_5171_cast_fp16")]; + tensor var_5177_pad_type_0 = const()[name = tensor("op_5177_pad_type_0"), val = tensor("valid")]; + tensor var_5177_strides_0 = const()[name = tensor("op_5177_strides_0"), val = tensor([1, 1])]; + tensor var_5177_pad_0 = const()[name = tensor("op_5177_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5177_dilations_0 = const()[name = tensor("op_5177_dilations_0"), val = tensor([1, 1])]; + tensor var_5177_groups_0 = const()[name = tensor("op_5177_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77722240))), name = tensor("layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77684928))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5177_cast_fp16 = conv(dilations = var_5177_dilations_0, groups = var_5177_groups_0, pad = var_5177_pad_0, pad_type = var_5177_pad_type_0, strides = var_5177_strides_0, weight = layers_13_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_355_cast_fp16)[name = tensor("op_5177_cast_fp16")]; + tensor input_357_cast_fp16 = add(x = var_5171_cast_fp16, y = var_5177_cast_fp16)[name = tensor("input_357_cast_fp16")]; + tensor input_359_cast_fp16 = silu(x = input_357_cast_fp16)[name = tensor("input_359_cast_fp16")]; + tensor var_5188_pad_type_0 = const()[name = tensor("op_5188_pad_type_0"), val = tensor("valid")]; + tensor var_5188_strides_0 = const()[name = tensor("op_5188_strides_0"), val = tensor([1, 1])]; + tensor var_5188_pad_0 = const()[name = tensor("op_5188_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5188_dilations_0 = const()[name = tensor("op_5188_dilations_0"), val = tensor([1, 1])]; + tensor var_5188_groups_0 = const()[name = tensor("op_5188_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(77853376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78639872))), name = tensor("layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_13_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78640064)))]; + tensor var_5188_cast_fp16 = conv(bias = layers_13_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_5188_dilations_0, groups = var_5188_groups_0, pad = var_5188_pad_0, pad_type = var_5188_pad_type_0, strides = var_5188_strides_0, weight = layers_13_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_359_cast_fp16)[name = tensor("op_5188_cast_fp16")]; + tensor var_5194_pad_type_0 = const()[name = tensor("op_5194_pad_type_0"), val = tensor("valid")]; + tensor var_5194_strides_0 = const()[name = tensor("op_5194_strides_0"), val = tensor([1, 1])]; + tensor var_5194_pad_0 = const()[name = tensor("op_5194_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5194_dilations_0 = const()[name = tensor("op_5194_dilations_0"), val = tensor([1, 1])]; + tensor var_5194_groups_0 = const()[name = tensor("op_5194_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78683136))), name = tensor("layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78641152))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5194_cast_fp16 = conv(dilations = var_5194_dilations_0, groups = var_5194_groups_0, pad = var_5194_pad_0, pad_type = var_5194_pad_type_0, strides = var_5194_strides_0, weight = layers_13_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_359_cast_fp16)[name = tensor("op_5194_cast_fp16")]; + tensor x_81_cast_fp16 = add(x = var_5188_cast_fp16, y = var_5194_cast_fp16)[name = tensor("x_81_cast_fp16")]; + tensor var_5196_to_fp16 = const()[name = tensor("op_5196_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5197_cast_fp16 = mul(x = x_81_cast_fp16, y = var_5196_to_fp16)[name = tensor("op_5197_cast_fp16")]; + tensor inputs_133_cast_fp16 = add(x = inputs_131_cast_fp16, y = var_5197_cast_fp16)[name = tensor("inputs_133_cast_fp16")]; + tensor out_133_axes_0 = const()[name = tensor("out_133_axes_0"), val = tensor([1])]; + tensor var_5207_to_fp16 = const()[name = tensor("op_5207_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_133_cast_fp16 = layer_norm(axes = out_133_axes_0, epsilon = var_5207_to_fp16, x = inputs_133_cast_fp16)[name = tensor("out_133_cast_fp16")]; + tensor obj_55_gamma_0_to_fp16 = const()[name = tensor("obj_55_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78814272)))]; + tensor obj_55_beta_0_to_fp16 = const()[name = tensor("obj_55_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78815360)))]; + tensor obj_55_epsilon_0_to_fp16 = const()[name = tensor("obj_55_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_55_cast_fp16 = batch_norm(beta = obj_55_beta_0_to_fp16, epsilon = obj_55_epsilon_0_to_fp16, gamma = obj_55_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_133_cast_fp16)[name = tensor("obj_55_cast_fp16")]; + tensor var_5232_pad_type_0 = const()[name = tensor("op_5232_pad_type_0"), val = tensor("valid")]; + tensor var_5232_strides_0 = const()[name = tensor("op_5232_strides_0"), val = tensor([1, 1])]; + tensor var_5232_pad_0 = const()[name = tensor("op_5232_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5232_dilations_0 = const()[name = tensor("op_5232_dilations_0"), val = tensor([1, 1])]; + tensor var_5232_groups_0 = const()[name = tensor("op_5232_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(78816448))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79013120))), name = tensor("layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_13_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79013312)))]; + tensor var_5232_cast_fp16 = conv(bias = layers_13_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_5232_dilations_0, groups = var_5232_groups_0, pad = var_5232_pad_0, pad_type = var_5232_pad_type_0, strides = var_5232_strides_0, weight = layers_13_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5232_cast_fp16")]; + tensor var_5238_pad_type_0 = const()[name = tensor("op_5238_pad_type_0"), val = tensor("valid")]; + tensor var_5238_strides_0 = const()[name = tensor("op_5238_strides_0"), val = tensor([1, 1])]; + tensor var_5238_pad_0 = const()[name = tensor("op_5238_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5238_dilations_0 = const()[name = tensor("op_5238_dilations_0"), val = tensor([1, 1])]; + tensor var_5238_groups_0 = const()[name = tensor("op_5238_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79024320))), name = tensor("layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79014400))), shape = tensor([512, 512, 1, 1])]; + tensor var_5238_cast_fp16 = conv(dilations = var_5238_dilations_0, groups = var_5238_groups_0, pad = var_5238_pad_0, pad_type = var_5238_pad_type_0, strides = var_5238_strides_0, weight = layers_13_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5238_cast_fp16")]; + tensor query_53_cast_fp16 = add(x = var_5232_cast_fp16, y = var_5238_cast_fp16)[name = tensor("query_53_cast_fp16")]; + tensor var_5247_pad_type_0 = const()[name = tensor("op_5247_pad_type_0"), val = tensor("valid")]; + tensor var_5247_strides_0 = const()[name = tensor("op_5247_strides_0"), val = tensor([1, 1])]; + tensor var_5247_pad_0 = const()[name = tensor("op_5247_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5247_dilations_0 = const()[name = tensor("op_5247_dilations_0"), val = tensor([1, 1])]; + tensor var_5247_groups_0 = const()[name = tensor("op_5247_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79057152))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79253824))), name = tensor("layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_5247_cast_fp16 = conv(dilations = var_5247_dilations_0, groups = var_5247_groups_0, pad = var_5247_pad_0, pad_type = var_5247_pad_type_0, strides = var_5247_strides_0, weight = layers_13_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5247_cast_fp16")]; + tensor var_5253_pad_type_0 = const()[name = tensor("op_5253_pad_type_0"), val = tensor("valid")]; + tensor var_5253_strides_0 = const()[name = tensor("op_5253_strides_0"), val = tensor([1, 1])]; + tensor var_5253_pad_0 = const()[name = tensor("op_5253_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5253_dilations_0 = const()[name = tensor("op_5253_dilations_0"), val = tensor([1, 1])]; + tensor var_5253_groups_0 = const()[name = tensor("op_5253_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79263680))), name = tensor("layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79254016))), shape = tensor([512, 512, 1, 1])]; + tensor var_5253_cast_fp16 = conv(dilations = var_5253_dilations_0, groups = var_5253_groups_0, pad = var_5253_pad_0, pad_type = var_5253_pad_type_0, strides = var_5253_strides_0, weight = layers_13_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5253_cast_fp16")]; + tensor key_27_cast_fp16 = add(x = var_5247_cast_fp16, y = var_5253_cast_fp16)[name = tensor("key_27_cast_fp16")]; + tensor var_5263_pad_type_0 = const()[name = tensor("op_5263_pad_type_0"), val = tensor("valid")]; + tensor var_5263_strides_0 = const()[name = tensor("op_5263_strides_0"), val = tensor([1, 1])]; + tensor var_5263_pad_0 = const()[name = tensor("op_5263_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5263_dilations_0 = const()[name = tensor("op_5263_dilations_0"), val = tensor([1, 1])]; + tensor var_5263_groups_0 = const()[name = tensor("op_5263_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79296512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79493184))), name = tensor("layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_13_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79493376)))]; + tensor var_5263_cast_fp16 = conv(bias = layers_13_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_5263_dilations_0, groups = var_5263_groups_0, pad = var_5263_pad_0, pad_type = var_5263_pad_type_0, strides = var_5263_strides_0, weight = layers_13_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_55_cast_fp16)[name = tensor("op_5263_cast_fp16")]; + tensor var_5269_pad_type_0 = const()[name = tensor("op_5269_pad_type_0"), val = tensor("valid")]; + tensor var_5269_strides_0 = const()[name = tensor("op_5269_strides_0"), val = tensor([1, 1])]; + tensor var_5269_pad_0 = const()[name = tensor("op_5269_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5269_dilations_0 = const()[name = tensor("op_5269_dilations_0"), val = tensor([1, 1])]; + tensor var_5269_groups_0 = const()[name = tensor("op_5269_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79502784))), name = tensor("layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79494464))), shape = tensor([512, 512, 1, 1])]; + tensor var_5269_cast_fp16 = conv(dilations = var_5269_dilations_0, groups = var_5269_groups_0, pad = var_5269_pad_0, pad_type = var_5269_pad_type_0, strides = var_5269_strides_0, weight = layers_13_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_55_cast_fp16)[name = tensor("op_5269_cast_fp16")]; + tensor value_27_cast_fp16 = add(x = var_5263_cast_fp16, y = var_5269_cast_fp16)[name = tensor("value_27_cast_fp16")]; + tensor var_5272_to_fp16 = const()[name = tensor("op_5272_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79535616)))]; + tensor query_55_cast_fp16 = add(x = query_53_cast_fp16, y = var_5272_to_fp16)[name = tensor("query_55_cast_fp16")]; + tensor var_5275_to_fp16 = const()[name = tensor("op_5275_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79536704)))]; + tensor q_with_bias_v_27_cast_fp16 = add(x = query_53_cast_fp16, y = var_5275_to_fp16)[name = tensor("q_with_bias_v_27_cast_fp16")]; + tensor var_5285_pad_type_0 = const()[name = tensor("op_5285_pad_type_0"), val = tensor("valid")]; + tensor var_5285_strides_0 = const()[name = tensor("op_5285_strides_0"), val = tensor([1, 1])]; + tensor var_5285_pad_0 = const()[name = tensor("op_5285_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5285_dilations_0 = const()[name = tensor("op_5285_dilations_0"), val = tensor([1, 1])]; + tensor var_5285_groups_0 = const()[name = tensor("op_5285_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79537792))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79734464))), name = tensor("layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_5285_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5285_dilations_0, groups = var_5285_groups_0, pad = var_5285_pad_0, pad_type = var_5285_pad_type_0, strides = var_5285_strides_0, weight = layers_13_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_5285_cast_fp16")]; + tensor var_5291_pad_type_0 = const()[name = tensor("op_5291_pad_type_0"), val = tensor("valid")]; + tensor var_5291_strides_0 = const()[name = tensor("op_5291_strides_0"), val = tensor([1, 1])]; + tensor var_5291_pad_0 = const()[name = tensor("op_5291_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5291_dilations_0 = const()[name = tensor("op_5291_dilations_0"), val = tensor([1, 1])]; + tensor var_5291_groups_0 = const()[name = tensor("op_5291_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79756544))), name = tensor("layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79734656))), shape = tensor([512, 512, 1, 1])]; + tensor var_5291_cast_fp16 = conv(dilations = var_5291_dilations_0, groups = var_5291_groups_0, pad = var_5291_pad_0, pad_type = var_5291_pad_type_0, strides = var_5291_strides_0, weight = layers_13_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_5291_cast_fp16")]; + tensor p_27_cast_fp16 = add(x = var_5285_cast_fp16, y = var_5291_cast_fp16)[name = tensor("p_27_cast_fp16")]; + tensor var_5295 = const()[name = tensor("op_5295"), val = tensor([1, 8, 64, 188])]; + tensor var_5296_cast_fp16 = reshape(shape = var_5295, x = q_with_bias_v_27_cast_fp16)[name = tensor("op_5296_cast_fp16")]; + tensor var_5297 = const()[name = tensor("op_5297"), val = tensor([1, 8, 64, -1])]; + tensor var_5298_cast_fp16 = reshape(shape = var_5297, x = p_27_cast_fp16)[name = tensor("op_5298_cast_fp16")]; + tensor matrix_bd_105_transpose_x_0 = const()[name = tensor("matrix_bd_105_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_105_transpose_y_0 = const()[name = tensor("matrix_bd_105_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_105_cast_fp16 = matmul(transpose_x = matrix_bd_105_transpose_x_0, transpose_y = matrix_bd_105_transpose_y_0, x = var_5296_cast_fp16, y = var_5298_cast_fp16)[name = tensor("matrix_bd_105_cast_fp16")]; + tensor matrix_bd_107_pad_0 = const()[name = tensor("matrix_bd_107_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_107_mode_0 = const()[name = tensor("matrix_bd_107_mode_0"), val = tensor("constant")]; + tensor const_153_to_fp16 = const()[name = tensor("const_153_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_107_cast_fp16 = pad(constant_val = const_153_to_fp16, mode = matrix_bd_107_mode_0, pad = matrix_bd_107_pad_0, x = matrix_bd_105_cast_fp16)[name = tensor("matrix_bd_107_cast_fp16")]; + tensor var_5307 = const()[name = tensor("op_5307"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_109_cast_fp16 = reshape(shape = var_5307, x = matrix_bd_107_cast_fp16)[name = tensor("matrix_bd_109_cast_fp16")]; + tensor var_5311_begin_0 = const()[name = tensor("op_5311_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_5311_end_0 = const()[name = tensor("op_5311_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_5311_end_mask_0 = const()[name = tensor("op_5311_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5311_cast_fp16 = slice_by_index(begin = var_5311_begin_0, end = var_5311_end_0, end_mask = var_5311_end_mask_0, x = matrix_bd_109_cast_fp16)[name = tensor("op_5311_cast_fp16")]; + tensor var_5312 = const()[name = tensor("op_5312"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_111_cast_fp16 = reshape(shape = var_5312, x = var_5311_cast_fp16)[name = tensor("matrix_bd_111_cast_fp16")]; + tensor var_5317_begin_0 = const()[name = tensor("op_5317_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5317_end_0 = const()[name = tensor("op_5317_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_5317_end_mask_0 = const()[name = tensor("op_5317_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5317_cast_fp16 = slice_by_index(begin = var_5317_begin_0, end = var_5317_end_0, end_mask = var_5317_end_mask_0, x = matrix_bd_111_cast_fp16)[name = tensor("op_5317_cast_fp16")]; + tensor var_5318_to_fp16 = const()[name = tensor("op_5318_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_27_cast_fp16 = mul(x = var_5317_cast_fp16, y = var_5318_to_fp16)[name = tensor("qk_mask_27_cast_fp16")]; + tensor var_5322 = const()[name = tensor("op_5322"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_27_cast_fp16 = reshape(shape = var_5322, x = query_55_cast_fp16)[name = tensor("mh_q_27_cast_fp16")]; + tensor var_5324_to_fp16 = const()[name = tensor("op_5324_to_fp16"), val = tensor(0x1p-3)]; + tensor var_5325_cast_fp16 = mul(x = mh_q_27_cast_fp16, y = var_5324_to_fp16)[name = tensor("op_5325_cast_fp16")]; + tensor var_5328 = const()[name = tensor("op_5328"), val = tensor([1, 8, 64, 188])]; + tensor var_5329_cast_fp16 = reshape(shape = var_5328, x = key_27_cast_fp16)[name = tensor("op_5329_cast_fp16")]; + tensor mh_w_53_transpose_x_0 = const()[name = tensor("mh_w_53_transpose_x_0"), val = tensor(true)]; + tensor mh_w_53_transpose_y_0 = const()[name = tensor("mh_w_53_transpose_y_0"), val = tensor(false)]; + tensor mh_w_53_cast_fp16 = matmul(transpose_x = mh_w_53_transpose_x_0, transpose_y = mh_w_53_transpose_y_0, x = var_5325_cast_fp16, y = var_5329_cast_fp16)[name = tensor("mh_w_53_cast_fp16")]; + tensor mh_w_55_cast_fp16 = add(x = mh_w_53_cast_fp16, y = qk_mask_27_cast_fp16)[name = tensor("mh_w_55_cast_fp16")]; + tensor var_5333_cast_fp16 = softmax(axis = var_5120, x = mh_w_55_cast_fp16)[name = tensor("op_5333_cast_fp16")]; + tensor var_5334 = const()[name = tensor("op_5334"), val = tensor([1, 8, 64, 188])]; + tensor var_5335_cast_fp16 = reshape(shape = var_5334, x = value_27_cast_fp16)[name = tensor("op_5335_cast_fp16")]; + tensor attn_27_transpose_x_0 = const()[name = tensor("attn_27_transpose_x_0"), val = tensor(false)]; + tensor attn_27_transpose_y_0 = const()[name = tensor("attn_27_transpose_y_0"), val = tensor(true)]; + tensor attn_27_cast_fp16 = matmul(transpose_x = attn_27_transpose_x_0, transpose_y = attn_27_transpose_y_0, x = var_5335_cast_fp16, y = var_5333_cast_fp16)[name = tensor("attn_27_cast_fp16")]; + tensor var_5338 = const()[name = tensor("op_5338"), val = tensor([1, 512, 1, 188])]; + tensor input_361_cast_fp16 = reshape(shape = var_5338, x = attn_27_cast_fp16)[name = tensor("input_361_cast_fp16")]; + tensor var_5348_pad_type_0 = const()[name = tensor("op_5348_pad_type_0"), val = tensor("valid")]; + tensor var_5348_strides_0 = const()[name = tensor("op_5348_strides_0"), val = tensor([1, 1])]; + tensor var_5348_pad_0 = const()[name = tensor("op_5348_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5348_dilations_0 = const()[name = tensor("op_5348_dilations_0"), val = tensor([1, 1])]; + tensor var_5348_groups_0 = const()[name = tensor("op_5348_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79789376))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79986048))), name = tensor("layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_13_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79986240)))]; + tensor var_5348_cast_fp16 = conv(bias = layers_13_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_5348_dilations_0, groups = var_5348_groups_0, pad = var_5348_pad_0, pad_type = var_5348_pad_type_0, strides = var_5348_strides_0, weight = layers_13_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_361_cast_fp16)[name = tensor("op_5348_cast_fp16")]; + tensor var_5354_pad_type_0 = const()[name = tensor("op_5354_pad_type_0"), val = tensor("valid")]; + tensor var_5354_strides_0 = const()[name = tensor("op_5354_strides_0"), val = tensor([1, 1])]; + tensor var_5354_pad_0 = const()[name = tensor("op_5354_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5354_dilations_0 = const()[name = tensor("op_5354_dilations_0"), val = tensor([1, 1])]; + tensor var_5354_groups_0 = const()[name = tensor("op_5354_groups_0"), val = tensor(1)]; + tensor layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79996608))), name = tensor("layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(79987328))), shape = tensor([512, 512, 1, 1])]; + tensor var_5354_cast_fp16 = conv(dilations = var_5354_dilations_0, groups = var_5354_groups_0, pad = var_5354_pad_0, pad_type = var_5354_pad_type_0, strides = var_5354_strides_0, weight = layers_13_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_361_cast_fp16)[name = tensor("op_5354_cast_fp16")]; + tensor obj_57_cast_fp16 = add(x = var_5348_cast_fp16, y = var_5354_cast_fp16)[name = tensor("obj_57_cast_fp16")]; + tensor inputs_135_cast_fp16 = add(x = inputs_133_cast_fp16, y = obj_57_cast_fp16)[name = tensor("inputs_135_cast_fp16")]; + tensor out_135_axes_0 = const()[name = tensor("out_135_axes_0"), val = tensor([1])]; + tensor var_5365_to_fp16 = const()[name = tensor("op_5365_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_135_cast_fp16 = layer_norm(axes = out_135_axes_0, epsilon = var_5365_to_fp16, x = inputs_135_cast_fp16)[name = tensor("out_135_cast_fp16")]; + tensor input_363_gamma_0_to_fp16 = const()[name = tensor("input_363_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80029440)))]; + tensor input_363_beta_0_to_fp16 = const()[name = tensor("input_363_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80030528)))]; + tensor input_363_epsilon_0_to_fp16 = const()[name = tensor("input_363_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_363_cast_fp16 = batch_norm(beta = input_363_beta_0_to_fp16, epsilon = input_363_epsilon_0_to_fp16, gamma = input_363_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_135_cast_fp16)[name = tensor("input_363_cast_fp16")]; + tensor var_5387_pad_type_0 = const()[name = tensor("op_5387_pad_type_0"), val = tensor("valid")]; + tensor var_5387_strides_0 = const()[name = tensor("op_5387_strides_0"), val = tensor([1, 1])]; + tensor var_5387_pad_0 = const()[name = tensor("op_5387_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5387_dilations_0 = const()[name = tensor("op_5387_dilations_0"), val = tensor([1, 1])]; + tensor var_5387_groups_0 = const()[name = tensor("op_5387_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80031616))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80424896))), name = tensor("layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_13_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80425088)))]; + tensor var_5387_cast_fp16 = conv(bias = layers_13_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_5387_dilations_0, groups = var_5387_groups_0, pad = var_5387_pad_0, pad_type = var_5387_pad_type_0, strides = var_5387_strides_0, weight = layers_13_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_363_cast_fp16)[name = tensor("op_5387_cast_fp16")]; + tensor var_5393_pad_type_0 = const()[name = tensor("op_5393_pad_type_0"), val = tensor("valid")]; + tensor var_5393_strides_0 = const()[name = tensor("op_5393_strides_0"), val = tensor([1, 1])]; + tensor var_5393_pad_0 = const()[name = tensor("op_5393_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5393_dilations_0 = const()[name = tensor("op_5393_dilations_0"), val = tensor([1, 1])]; + tensor var_5393_groups_0 = const()[name = tensor("op_5393_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80446784))), name = tensor("layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80427200))), shape = tensor([1024, 512, 1, 1])]; + tensor var_5393_cast_fp16 = conv(dilations = var_5393_dilations_0, groups = var_5393_groups_0, pad = var_5393_pad_0, pad_type = var_5393_pad_type_0, strides = var_5393_strides_0, weight = layers_13_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_363_cast_fp16)[name = tensor("op_5393_cast_fp16")]; + tensor input_365_cast_fp16 = add(x = var_5387_cast_fp16, y = var_5393_cast_fp16)[name = tensor("input_365_cast_fp16")]; + tensor input_367_split_num_splits_0 = const()[name = tensor("input_367_split_num_splits_0"), val = tensor(2)]; + tensor input_367_split_axis_0 = const()[name = tensor("input_367_split_axis_0"), val = tensor(1)]; + tensor input_367_split_cast_fp16_0, tensor input_367_split_cast_fp16_1 = split(axis = input_367_split_axis_0, num_splits = input_367_split_num_splits_0, x = input_365_cast_fp16)[name = tensor("input_367_split_cast_fp16")]; + tensor input_367_split_1_sigmoid_cast_fp16 = sigmoid(x = input_367_split_cast_fp16_1)[name = tensor("input_367_split_1_sigmoid_cast_fp16")]; + tensor input_367_cast_fp16 = mul(x = input_367_split_cast_fp16_0, y = input_367_split_1_sigmoid_cast_fp16)[name = tensor("input_367_cast_fp16")]; + tensor input_369_pad_type_0 = const()[name = tensor("input_369_pad_type_0"), val = tensor("custom")]; + tensor input_369_pad_0 = const()[name = tensor("input_369_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_369_groups_0 = const()[name = tensor("input_369_groups_0"), val = tensor(512)]; + tensor input_369_strides_0 = const()[name = tensor("input_369_strides_0"), val = tensor([1, 1])]; + tensor input_369_dilations_0 = const()[name = tensor("input_369_dilations_0"), val = tensor([1, 1])]; + tensor const_217_to_fp16 = const()[name = tensor("const_217_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80512384)))]; + tensor const_218_to_fp16 = const()[name = tensor("const_218_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80521664)))]; + tensor input_371_cast_fp16 = conv(bias = const_218_to_fp16, dilations = input_369_dilations_0, groups = input_369_groups_0, pad = input_369_pad_0, pad_type = input_369_pad_type_0, strides = input_369_strides_0, weight = const_217_to_fp16, x = input_367_cast_fp16)[name = tensor("input_371_cast_fp16")]; + tensor input_373_cast_fp16 = silu(x = input_371_cast_fp16)[name = tensor("input_373_cast_fp16")]; + tensor var_5417_pad_type_0 = const()[name = tensor("op_5417_pad_type_0"), val = tensor("valid")]; + tensor var_5417_strides_0 = const()[name = tensor("op_5417_strides_0"), val = tensor([1, 1])]; + tensor var_5417_pad_0 = const()[name = tensor("op_5417_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5417_dilations_0 = const()[name = tensor("op_5417_dilations_0"), val = tensor([1, 1])]; + tensor var_5417_groups_0 = const()[name = tensor("op_5417_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80522752))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80719424))), name = tensor("layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_13_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80719616)))]; + tensor var_5417_cast_fp16 = conv(bias = layers_13_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_5417_dilations_0, groups = var_5417_groups_0, pad = var_5417_pad_0, pad_type = var_5417_pad_type_0, strides = var_5417_strides_0, weight = layers_13_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_373_cast_fp16)[name = tensor("op_5417_cast_fp16")]; + tensor var_5423_pad_type_0 = const()[name = tensor("op_5423_pad_type_0"), val = tensor("valid")]; + tensor var_5423_strides_0 = const()[name = tensor("op_5423_strides_0"), val = tensor([1, 1])]; + tensor var_5423_pad_0 = const()[name = tensor("op_5423_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5423_dilations_0 = const()[name = tensor("op_5423_dilations_0"), val = tensor([1, 1])]; + tensor var_5423_groups_0 = const()[name = tensor("op_5423_groups_0"), val = tensor(1)]; + tensor layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80730112))), name = tensor("layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80720704))), shape = tensor([512, 512, 1, 1])]; + tensor var_5423_cast_fp16 = conv(dilations = var_5423_dilations_0, groups = var_5423_groups_0, pad = var_5423_pad_0, pad_type = var_5423_pad_type_0, strides = var_5423_strides_0, weight = layers_13_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_373_cast_fp16)[name = tensor("op_5423_cast_fp16")]; + tensor x_83_cast_fp16 = add(x = var_5417_cast_fp16, y = var_5423_cast_fp16)[name = tensor("x_83_cast_fp16")]; + tensor inputs_137_cast_fp16 = add(x = inputs_135_cast_fp16, y = x_83_cast_fp16)[name = tensor("inputs_137_cast_fp16")]; + tensor out_137_axes_0 = const()[name = tensor("out_137_axes_0"), val = tensor([1])]; + tensor var_5434_to_fp16 = const()[name = tensor("op_5434_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_137_cast_fp16 = layer_norm(axes = out_137_axes_0, epsilon = var_5434_to_fp16, x = inputs_137_cast_fp16)[name = tensor("out_137_cast_fp16")]; + tensor input_375_gamma_0_to_fp16 = const()[name = tensor("input_375_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80762944)))]; + tensor input_375_beta_0_to_fp16 = const()[name = tensor("input_375_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80764032)))]; + tensor input_375_epsilon_0_to_fp16 = const()[name = tensor("input_375_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_375_cast_fp16 = batch_norm(beta = input_375_beta_0_to_fp16, epsilon = input_375_epsilon_0_to_fp16, gamma = input_375_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_137_cast_fp16)[name = tensor("input_375_cast_fp16")]; + tensor var_5454_pad_type_0 = const()[name = tensor("op_5454_pad_type_0"), val = tensor("valid")]; + tensor var_5454_strides_0 = const()[name = tensor("op_5454_strides_0"), val = tensor([1, 1])]; + tensor var_5454_pad_0 = const()[name = tensor("op_5454_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5454_dilations_0 = const()[name = tensor("op_5454_dilations_0"), val = tensor([1, 1])]; + tensor var_5454_groups_0 = const()[name = tensor("op_5454_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(80765120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81551616))), name = tensor("layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_13_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81551808)))]; + tensor var_5454_cast_fp16 = conv(bias = layers_13_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_5454_dilations_0, groups = var_5454_groups_0, pad = var_5454_pad_0, pad_type = var_5454_pad_type_0, strides = var_5454_strides_0, weight = layers_13_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_375_cast_fp16)[name = tensor("op_5454_cast_fp16")]; + tensor var_5460_pad_type_0 = const()[name = tensor("op_5460_pad_type_0"), val = tensor("valid")]; + tensor var_5460_strides_0 = const()[name = tensor("op_5460_strides_0"), val = tensor([1, 1])]; + tensor var_5460_pad_0 = const()[name = tensor("op_5460_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5460_dilations_0 = const()[name = tensor("op_5460_dilations_0"), val = tensor([1, 1])]; + tensor var_5460_groups_0 = const()[name = tensor("op_5460_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81591104))), name = tensor("layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81555968))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5460_cast_fp16 = conv(dilations = var_5460_dilations_0, groups = var_5460_groups_0, pad = var_5460_pad_0, pad_type = var_5460_pad_type_0, strides = var_5460_strides_0, weight = layers_13_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_375_cast_fp16)[name = tensor("op_5460_cast_fp16")]; + tensor input_377_cast_fp16 = add(x = var_5454_cast_fp16, y = var_5460_cast_fp16)[name = tensor("input_377_cast_fp16")]; + tensor input_379_cast_fp16 = silu(x = input_377_cast_fp16)[name = tensor("input_379_cast_fp16")]; + tensor var_5471_pad_type_0 = const()[name = tensor("op_5471_pad_type_0"), val = tensor("valid")]; + tensor var_5471_strides_0 = const()[name = tensor("op_5471_strides_0"), val = tensor([1, 1])]; + tensor var_5471_pad_0 = const()[name = tensor("op_5471_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5471_dilations_0 = const()[name = tensor("op_5471_dilations_0"), val = tensor([1, 1])]; + tensor var_5471_groups_0 = const()[name = tensor("op_5471_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(81722240))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82508736))), name = tensor("layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_13_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_13_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82508928)))]; + tensor var_5471_cast_fp16 = conv(bias = layers_13_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_5471_dilations_0, groups = var_5471_groups_0, pad = var_5471_pad_0, pad_type = var_5471_pad_type_0, strides = var_5471_strides_0, weight = layers_13_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_379_cast_fp16)[name = tensor("op_5471_cast_fp16")]; + tensor var_5477_pad_type_0 = const()[name = tensor("op_5477_pad_type_0"), val = tensor("valid")]; + tensor var_5477_strides_0 = const()[name = tensor("op_5477_strides_0"), val = tensor([1, 1])]; + tensor var_5477_pad_0 = const()[name = tensor("op_5477_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5477_dilations_0 = const()[name = tensor("op_5477_dilations_0"), val = tensor([1, 1])]; + tensor var_5477_groups_0 = const()[name = tensor("op_5477_groups_0"), val = tensor(1)]; + tensor layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82552320))), name = tensor("layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82510016))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5477_cast_fp16 = conv(dilations = var_5477_dilations_0, groups = var_5477_groups_0, pad = var_5477_pad_0, pad_type = var_5477_pad_type_0, strides = var_5477_strides_0, weight = layers_13_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_379_cast_fp16)[name = tensor("op_5477_cast_fp16")]; + tensor x_85_cast_fp16 = add(x = var_5471_cast_fp16, y = var_5477_cast_fp16)[name = tensor("x_85_cast_fp16")]; + tensor var_5479_to_fp16 = const()[name = tensor("op_5479_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5480_cast_fp16 = mul(x = x_85_cast_fp16, y = var_5479_to_fp16)[name = tensor("op_5480_cast_fp16")]; + tensor inputs_139_cast_fp16 = add(x = inputs_137_cast_fp16, y = var_5480_cast_fp16)[name = tensor("inputs_139_cast_fp16")]; + tensor out_139_axes_0 = const()[name = tensor("out_139_axes_0"), val = tensor([1])]; + tensor var_5490_to_fp16 = const()[name = tensor("op_5490_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_139_cast_fp16 = layer_norm(axes = out_139_axes_0, epsilon = var_5490_to_fp16, x = inputs_139_cast_fp16)[name = tensor("out_139_cast_fp16")]; + tensor inputs_141_gamma_0_to_fp16 = const()[name = tensor("inputs_141_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82683456)))]; + tensor inputs_141_beta_0_to_fp16 = const()[name = tensor("inputs_141_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82684544)))]; + tensor inputs_141_epsilon_0_to_fp16 = const()[name = tensor("inputs_141_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_141_cast_fp16 = batch_norm(beta = inputs_141_beta_0_to_fp16, epsilon = inputs_141_epsilon_0_to_fp16, gamma = inputs_141_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_139_cast_fp16)[name = tensor("inputs_141_cast_fp16")]; + tensor var_5504 = const()[name = tensor("op_5504"), val = tensor(3)]; + tensor out_141_axes_0 = const()[name = tensor("out_141_axes_0"), val = tensor([1])]; + tensor var_5535_to_fp16 = const()[name = tensor("op_5535_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_141_cast_fp16 = layer_norm(axes = out_141_axes_0, epsilon = var_5535_to_fp16, x = inputs_141_cast_fp16)[name = tensor("out_141_cast_fp16")]; + tensor input_381_gamma_0_to_fp16 = const()[name = tensor("input_381_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82685632)))]; + tensor input_381_beta_0_to_fp16 = const()[name = tensor("input_381_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82686720)))]; + tensor input_381_epsilon_0_to_fp16 = const()[name = tensor("input_381_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_381_cast_fp16 = batch_norm(beta = input_381_beta_0_to_fp16, epsilon = input_381_epsilon_0_to_fp16, gamma = input_381_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_141_cast_fp16)[name = tensor("input_381_cast_fp16")]; + tensor var_5555_pad_type_0 = const()[name = tensor("op_5555_pad_type_0"), val = tensor("valid")]; + tensor var_5555_strides_0 = const()[name = tensor("op_5555_strides_0"), val = tensor([1, 1])]; + tensor var_5555_pad_0 = const()[name = tensor("op_5555_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5555_dilations_0 = const()[name = tensor("op_5555_dilations_0"), val = tensor([1, 1])]; + tensor var_5555_groups_0 = const()[name = tensor("op_5555_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(82687808))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83474304))), name = tensor("layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_14_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83474496)))]; + tensor var_5555_cast_fp16 = conv(bias = layers_14_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5555_dilations_0, groups = var_5555_groups_0, pad = var_5555_pad_0, pad_type = var_5555_pad_type_0, strides = var_5555_strides_0, weight = layers_14_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_381_cast_fp16)[name = tensor("op_5555_cast_fp16")]; + tensor var_5561_pad_type_0 = const()[name = tensor("op_5561_pad_type_0"), val = tensor("valid")]; + tensor var_5561_strides_0 = const()[name = tensor("op_5561_strides_0"), val = tensor([1, 1])]; + tensor var_5561_pad_0 = const()[name = tensor("op_5561_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5561_dilations_0 = const()[name = tensor("op_5561_dilations_0"), val = tensor([1, 1])]; + tensor var_5561_groups_0 = const()[name = tensor("op_5561_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83513984))), name = tensor("layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83478656))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5561_cast_fp16 = conv(dilations = var_5561_dilations_0, groups = var_5561_groups_0, pad = var_5561_pad_0, pad_type = var_5561_pad_type_0, strides = var_5561_strides_0, weight = layers_14_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_381_cast_fp16)[name = tensor("op_5561_cast_fp16")]; + tensor input_383_cast_fp16 = add(x = var_5555_cast_fp16, y = var_5561_cast_fp16)[name = tensor("input_383_cast_fp16")]; + tensor input_385_cast_fp16 = silu(x = input_383_cast_fp16)[name = tensor("input_385_cast_fp16")]; + tensor var_5572_pad_type_0 = const()[name = tensor("op_5572_pad_type_0"), val = tensor("valid")]; + tensor var_5572_strides_0 = const()[name = tensor("op_5572_strides_0"), val = tensor([1, 1])]; + tensor var_5572_pad_0 = const()[name = tensor("op_5572_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5572_dilations_0 = const()[name = tensor("op_5572_dilations_0"), val = tensor([1, 1])]; + tensor var_5572_groups_0 = const()[name = tensor("op_5572_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(83645120))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84431616))), name = tensor("layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_14_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84431808)))]; + tensor var_5572_cast_fp16 = conv(bias = layers_14_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_5572_dilations_0, groups = var_5572_groups_0, pad = var_5572_pad_0, pad_type = var_5572_pad_type_0, strides = var_5572_strides_0, weight = layers_14_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_385_cast_fp16)[name = tensor("op_5572_cast_fp16")]; + tensor var_5578_pad_type_0 = const()[name = tensor("op_5578_pad_type_0"), val = tensor("valid")]; + tensor var_5578_strides_0 = const()[name = tensor("op_5578_strides_0"), val = tensor([1, 1])]; + tensor var_5578_pad_0 = const()[name = tensor("op_5578_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5578_dilations_0 = const()[name = tensor("op_5578_dilations_0"), val = tensor([1, 1])]; + tensor var_5578_groups_0 = const()[name = tensor("op_5578_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84471232))), name = tensor("layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84432896))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5578_cast_fp16 = conv(dilations = var_5578_dilations_0, groups = var_5578_groups_0, pad = var_5578_pad_0, pad_type = var_5578_pad_type_0, strides = var_5578_strides_0, weight = layers_14_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_385_cast_fp16)[name = tensor("op_5578_cast_fp16")]; + tensor x_87_cast_fp16 = add(x = var_5572_cast_fp16, y = var_5578_cast_fp16)[name = tensor("x_87_cast_fp16")]; + tensor var_5580_to_fp16 = const()[name = tensor("op_5580_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5581_cast_fp16 = mul(x = x_87_cast_fp16, y = var_5580_to_fp16)[name = tensor("op_5581_cast_fp16")]; + tensor inputs_143_cast_fp16 = add(x = inputs_141_cast_fp16, y = var_5581_cast_fp16)[name = tensor("inputs_143_cast_fp16")]; + tensor out_143_axes_0 = const()[name = tensor("out_143_axes_0"), val = tensor([1])]; + tensor var_5591_to_fp16 = const()[name = tensor("op_5591_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_143_cast_fp16 = layer_norm(axes = out_143_axes_0, epsilon = var_5591_to_fp16, x = inputs_143_cast_fp16)[name = tensor("out_143_cast_fp16")]; + tensor obj_59_gamma_0_to_fp16 = const()[name = tensor("obj_59_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84602368)))]; + tensor obj_59_beta_0_to_fp16 = const()[name = tensor("obj_59_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84603456)))]; + tensor obj_59_epsilon_0_to_fp16 = const()[name = tensor("obj_59_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_59_cast_fp16 = batch_norm(beta = obj_59_beta_0_to_fp16, epsilon = obj_59_epsilon_0_to_fp16, gamma = obj_59_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_143_cast_fp16)[name = tensor("obj_59_cast_fp16")]; + tensor var_5616_pad_type_0 = const()[name = tensor("op_5616_pad_type_0"), val = tensor("valid")]; + tensor var_5616_strides_0 = const()[name = tensor("op_5616_strides_0"), val = tensor([1, 1])]; + tensor var_5616_pad_0 = const()[name = tensor("op_5616_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5616_dilations_0 = const()[name = tensor("op_5616_dilations_0"), val = tensor([1, 1])]; + tensor var_5616_groups_0 = const()[name = tensor("op_5616_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84604544))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84801216))), name = tensor("layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_14_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84801408)))]; + tensor var_5616_cast_fp16 = conv(bias = layers_14_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_5616_dilations_0, groups = var_5616_groups_0, pad = var_5616_pad_0, pad_type = var_5616_pad_type_0, strides = var_5616_strides_0, weight = layers_14_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5616_cast_fp16")]; + tensor var_5622_pad_type_0 = const()[name = tensor("op_5622_pad_type_0"), val = tensor("valid")]; + tensor var_5622_strides_0 = const()[name = tensor("op_5622_strides_0"), val = tensor([1, 1])]; + tensor var_5622_pad_0 = const()[name = tensor("op_5622_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5622_dilations_0 = const()[name = tensor("op_5622_dilations_0"), val = tensor([1, 1])]; + tensor var_5622_groups_0 = const()[name = tensor("op_5622_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84809984))), name = tensor("layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84802496))), shape = tensor([512, 512, 1, 1])]; + tensor var_5622_cast_fp16 = conv(dilations = var_5622_dilations_0, groups = var_5622_groups_0, pad = var_5622_pad_0, pad_type = var_5622_pad_type_0, strides = var_5622_strides_0, weight = layers_14_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5622_cast_fp16")]; + tensor query_57_cast_fp16 = add(x = var_5616_cast_fp16, y = var_5622_cast_fp16)[name = tensor("query_57_cast_fp16")]; + tensor var_5631_pad_type_0 = const()[name = tensor("op_5631_pad_type_0"), val = tensor("valid")]; + tensor var_5631_strides_0 = const()[name = tensor("op_5631_strides_0"), val = tensor([1, 1])]; + tensor var_5631_pad_0 = const()[name = tensor("op_5631_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5631_dilations_0 = const()[name = tensor("op_5631_dilations_0"), val = tensor([1, 1])]; + tensor var_5631_groups_0 = const()[name = tensor("op_5631_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(84842816))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85039488))), name = tensor("layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_5631_cast_fp16 = conv(dilations = var_5631_dilations_0, groups = var_5631_groups_0, pad = var_5631_pad_0, pad_type = var_5631_pad_type_0, strides = var_5631_strides_0, weight = layers_14_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5631_cast_fp16")]; + tensor var_5637_pad_type_0 = const()[name = tensor("op_5637_pad_type_0"), val = tensor("valid")]; + tensor var_5637_strides_0 = const()[name = tensor("op_5637_strides_0"), val = tensor([1, 1])]; + tensor var_5637_pad_0 = const()[name = tensor("op_5637_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5637_dilations_0 = const()[name = tensor("op_5637_dilations_0"), val = tensor([1, 1])]; + tensor var_5637_groups_0 = const()[name = tensor("op_5637_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85047232))), name = tensor("layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85039680))), shape = tensor([512, 512, 1, 1])]; + tensor var_5637_cast_fp16 = conv(dilations = var_5637_dilations_0, groups = var_5637_groups_0, pad = var_5637_pad_0, pad_type = var_5637_pad_type_0, strides = var_5637_strides_0, weight = layers_14_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5637_cast_fp16")]; + tensor key_29_cast_fp16 = add(x = var_5631_cast_fp16, y = var_5637_cast_fp16)[name = tensor("key_29_cast_fp16")]; + tensor var_5647_pad_type_0 = const()[name = tensor("op_5647_pad_type_0"), val = tensor("valid")]; + tensor var_5647_strides_0 = const()[name = tensor("op_5647_strides_0"), val = tensor([1, 1])]; + tensor var_5647_pad_0 = const()[name = tensor("op_5647_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5647_dilations_0 = const()[name = tensor("op_5647_dilations_0"), val = tensor([1, 1])]; + tensor var_5647_groups_0 = const()[name = tensor("op_5647_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85080064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85276736))), name = tensor("layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_14_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85276928)))]; + tensor var_5647_cast_fp16 = conv(bias = layers_14_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_5647_dilations_0, groups = var_5647_groups_0, pad = var_5647_pad_0, pad_type = var_5647_pad_type_0, strides = var_5647_strides_0, weight = layers_14_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_59_cast_fp16)[name = tensor("op_5647_cast_fp16")]; + tensor var_5653_pad_type_0 = const()[name = tensor("op_5653_pad_type_0"), val = tensor("valid")]; + tensor var_5653_strides_0 = const()[name = tensor("op_5653_strides_0"), val = tensor([1, 1])]; + tensor var_5653_pad_0 = const()[name = tensor("op_5653_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5653_dilations_0 = const()[name = tensor("op_5653_dilations_0"), val = tensor([1, 1])]; + tensor var_5653_groups_0 = const()[name = tensor("op_5653_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85285760))), name = tensor("layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85278016))), shape = tensor([512, 512, 1, 1])]; + tensor var_5653_cast_fp16 = conv(dilations = var_5653_dilations_0, groups = var_5653_groups_0, pad = var_5653_pad_0, pad_type = var_5653_pad_type_0, strides = var_5653_strides_0, weight = layers_14_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_59_cast_fp16)[name = tensor("op_5653_cast_fp16")]; + tensor value_29_cast_fp16 = add(x = var_5647_cast_fp16, y = var_5653_cast_fp16)[name = tensor("value_29_cast_fp16")]; + tensor var_5656_to_fp16 = const()[name = tensor("op_5656_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85318592)))]; + tensor query_59_cast_fp16 = add(x = query_57_cast_fp16, y = var_5656_to_fp16)[name = tensor("query_59_cast_fp16")]; + tensor var_5659_to_fp16 = const()[name = tensor("op_5659_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85319680)))]; + tensor q_with_bias_v_29_cast_fp16 = add(x = query_57_cast_fp16, y = var_5659_to_fp16)[name = tensor("q_with_bias_v_29_cast_fp16")]; + tensor var_5669_pad_type_0 = const()[name = tensor("op_5669_pad_type_0"), val = tensor("valid")]; + tensor var_5669_strides_0 = const()[name = tensor("op_5669_strides_0"), val = tensor([1, 1])]; + tensor var_5669_pad_0 = const()[name = tensor("op_5669_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5669_dilations_0 = const()[name = tensor("op_5669_dilations_0"), val = tensor([1, 1])]; + tensor var_5669_groups_0 = const()[name = tensor("op_5669_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85320768))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85517440))), name = tensor("layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_5669_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_5669_dilations_0, groups = var_5669_groups_0, pad = var_5669_pad_0, pad_type = var_5669_pad_type_0, strides = var_5669_strides_0, weight = layers_14_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_5669_cast_fp16")]; + tensor var_5675_pad_type_0 = const()[name = tensor("op_5675_pad_type_0"), val = tensor("valid")]; + tensor var_5675_strides_0 = const()[name = tensor("op_5675_strides_0"), val = tensor([1, 1])]; + tensor var_5675_pad_0 = const()[name = tensor("op_5675_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5675_dilations_0 = const()[name = tensor("op_5675_dilations_0"), val = tensor([1, 1])]; + tensor var_5675_groups_0 = const()[name = tensor("op_5675_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85539840))), name = tensor("layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85517632))), shape = tensor([512, 512, 1, 1])]; + tensor var_5675_cast_fp16 = conv(dilations = var_5675_dilations_0, groups = var_5675_groups_0, pad = var_5675_pad_0, pad_type = var_5675_pad_type_0, strides = var_5675_strides_0, weight = layers_14_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_5675_cast_fp16")]; + tensor p_29_cast_fp16 = add(x = var_5669_cast_fp16, y = var_5675_cast_fp16)[name = tensor("p_29_cast_fp16")]; + tensor var_5679 = const()[name = tensor("op_5679"), val = tensor([1, 8, 64, 188])]; + tensor var_5680_cast_fp16 = reshape(shape = var_5679, x = q_with_bias_v_29_cast_fp16)[name = tensor("op_5680_cast_fp16")]; + tensor var_5681 = const()[name = tensor("op_5681"), val = tensor([1, 8, 64, -1])]; + tensor var_5682_cast_fp16 = reshape(shape = var_5681, x = p_29_cast_fp16)[name = tensor("op_5682_cast_fp16")]; + tensor matrix_bd_113_transpose_x_0 = const()[name = tensor("matrix_bd_113_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_113_transpose_y_0 = const()[name = tensor("matrix_bd_113_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_113_cast_fp16 = matmul(transpose_x = matrix_bd_113_transpose_x_0, transpose_y = matrix_bd_113_transpose_y_0, x = var_5680_cast_fp16, y = var_5682_cast_fp16)[name = tensor("matrix_bd_113_cast_fp16")]; + tensor matrix_bd_115_pad_0 = const()[name = tensor("matrix_bd_115_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_115_mode_0 = const()[name = tensor("matrix_bd_115_mode_0"), val = tensor("constant")]; + tensor const_164_to_fp16 = const()[name = tensor("const_164_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_115_cast_fp16 = pad(constant_val = const_164_to_fp16, mode = matrix_bd_115_mode_0, pad = matrix_bd_115_pad_0, x = matrix_bd_113_cast_fp16)[name = tensor("matrix_bd_115_cast_fp16")]; + tensor var_5691 = const()[name = tensor("op_5691"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_117_cast_fp16 = reshape(shape = var_5691, x = matrix_bd_115_cast_fp16)[name = tensor("matrix_bd_117_cast_fp16")]; + tensor var_5695_begin_0 = const()[name = tensor("op_5695_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_5695_end_0 = const()[name = tensor("op_5695_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_5695_end_mask_0 = const()[name = tensor("op_5695_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_5695_cast_fp16 = slice_by_index(begin = var_5695_begin_0, end = var_5695_end_0, end_mask = var_5695_end_mask_0, x = matrix_bd_117_cast_fp16)[name = tensor("op_5695_cast_fp16")]; + tensor var_5696 = const()[name = tensor("op_5696"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_119_cast_fp16 = reshape(shape = var_5696, x = var_5695_cast_fp16)[name = tensor("matrix_bd_119_cast_fp16")]; + tensor var_5701_begin_0 = const()[name = tensor("op_5701_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5701_end_0 = const()[name = tensor("op_5701_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_5701_end_mask_0 = const()[name = tensor("op_5701_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_5701_cast_fp16 = slice_by_index(begin = var_5701_begin_0, end = var_5701_end_0, end_mask = var_5701_end_mask_0, x = matrix_bd_119_cast_fp16)[name = tensor("op_5701_cast_fp16")]; + tensor var_5702_to_fp16 = const()[name = tensor("op_5702_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_29_cast_fp16 = mul(x = var_5701_cast_fp16, y = var_5702_to_fp16)[name = tensor("qk_mask_29_cast_fp16")]; + tensor var_5706 = const()[name = tensor("op_5706"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_29_cast_fp16 = reshape(shape = var_5706, x = query_59_cast_fp16)[name = tensor("mh_q_29_cast_fp16")]; + tensor var_5708_to_fp16 = const()[name = tensor("op_5708_to_fp16"), val = tensor(0x1p-3)]; + tensor var_5709_cast_fp16 = mul(x = mh_q_29_cast_fp16, y = var_5708_to_fp16)[name = tensor("op_5709_cast_fp16")]; + tensor var_5712 = const()[name = tensor("op_5712"), val = tensor([1, 8, 64, 188])]; + tensor var_5713_cast_fp16 = reshape(shape = var_5712, x = key_29_cast_fp16)[name = tensor("op_5713_cast_fp16")]; + tensor mh_w_57_transpose_x_0 = const()[name = tensor("mh_w_57_transpose_x_0"), val = tensor(true)]; + tensor mh_w_57_transpose_y_0 = const()[name = tensor("mh_w_57_transpose_y_0"), val = tensor(false)]; + tensor mh_w_57_cast_fp16 = matmul(transpose_x = mh_w_57_transpose_x_0, transpose_y = mh_w_57_transpose_y_0, x = var_5709_cast_fp16, y = var_5713_cast_fp16)[name = tensor("mh_w_57_cast_fp16")]; + tensor mh_w_59_cast_fp16 = add(x = mh_w_57_cast_fp16, y = qk_mask_29_cast_fp16)[name = tensor("mh_w_59_cast_fp16")]; + tensor var_5717_cast_fp16 = softmax(axis = var_5504, x = mh_w_59_cast_fp16)[name = tensor("op_5717_cast_fp16")]; + tensor var_5718 = const()[name = tensor("op_5718"), val = tensor([1, 8, 64, 188])]; + tensor var_5719_cast_fp16 = reshape(shape = var_5718, x = value_29_cast_fp16)[name = tensor("op_5719_cast_fp16")]; + tensor attn_29_transpose_x_0 = const()[name = tensor("attn_29_transpose_x_0"), val = tensor(false)]; + tensor attn_29_transpose_y_0 = const()[name = tensor("attn_29_transpose_y_0"), val = tensor(true)]; + tensor attn_29_cast_fp16 = matmul(transpose_x = attn_29_transpose_x_0, transpose_y = attn_29_transpose_y_0, x = var_5719_cast_fp16, y = var_5717_cast_fp16)[name = tensor("attn_29_cast_fp16")]; + tensor var_5722 = const()[name = tensor("op_5722"), val = tensor([1, 512, 1, 188])]; + tensor input_387_cast_fp16 = reshape(shape = var_5722, x = attn_29_cast_fp16)[name = tensor("input_387_cast_fp16")]; + tensor var_5732_pad_type_0 = const()[name = tensor("op_5732_pad_type_0"), val = tensor("valid")]; + tensor var_5732_strides_0 = const()[name = tensor("op_5732_strides_0"), val = tensor([1, 1])]; + tensor var_5732_pad_0 = const()[name = tensor("op_5732_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5732_dilations_0 = const()[name = tensor("op_5732_dilations_0"), val = tensor([1, 1])]; + tensor var_5732_groups_0 = const()[name = tensor("op_5732_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85572672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85769344))), name = tensor("layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_14_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85769536)))]; + tensor var_5732_cast_fp16 = conv(bias = layers_14_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_5732_dilations_0, groups = var_5732_groups_0, pad = var_5732_pad_0, pad_type = var_5732_pad_type_0, strides = var_5732_strides_0, weight = layers_14_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_387_cast_fp16)[name = tensor("op_5732_cast_fp16")]; + tensor var_5738_pad_type_0 = const()[name = tensor("op_5738_pad_type_0"), val = tensor("valid")]; + tensor var_5738_strides_0 = const()[name = tensor("op_5738_strides_0"), val = tensor([1, 1])]; + tensor var_5738_pad_0 = const()[name = tensor("op_5738_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5738_dilations_0 = const()[name = tensor("op_5738_dilations_0"), val = tensor([1, 1])]; + tensor var_5738_groups_0 = const()[name = tensor("op_5738_groups_0"), val = tensor(1)]; + tensor layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85779392))), name = tensor("layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85770624))), shape = tensor([512, 512, 1, 1])]; + tensor var_5738_cast_fp16 = conv(dilations = var_5738_dilations_0, groups = var_5738_groups_0, pad = var_5738_pad_0, pad_type = var_5738_pad_type_0, strides = var_5738_strides_0, weight = layers_14_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_387_cast_fp16)[name = tensor("op_5738_cast_fp16")]; + tensor obj_61_cast_fp16 = add(x = var_5732_cast_fp16, y = var_5738_cast_fp16)[name = tensor("obj_61_cast_fp16")]; + tensor inputs_145_cast_fp16 = add(x = inputs_143_cast_fp16, y = obj_61_cast_fp16)[name = tensor("inputs_145_cast_fp16")]; + tensor out_145_axes_0 = const()[name = tensor("out_145_axes_0"), val = tensor([1])]; + tensor var_5749_to_fp16 = const()[name = tensor("op_5749_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_145_cast_fp16 = layer_norm(axes = out_145_axes_0, epsilon = var_5749_to_fp16, x = inputs_145_cast_fp16)[name = tensor("out_145_cast_fp16")]; + tensor input_389_gamma_0_to_fp16 = const()[name = tensor("input_389_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85812224)))]; + tensor input_389_beta_0_to_fp16 = const()[name = tensor("input_389_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85813312)))]; + tensor input_389_epsilon_0_to_fp16 = const()[name = tensor("input_389_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_389_cast_fp16 = batch_norm(beta = input_389_beta_0_to_fp16, epsilon = input_389_epsilon_0_to_fp16, gamma = input_389_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_145_cast_fp16)[name = tensor("input_389_cast_fp16")]; + tensor var_5771_pad_type_0 = const()[name = tensor("op_5771_pad_type_0"), val = tensor("valid")]; + tensor var_5771_strides_0 = const()[name = tensor("op_5771_strides_0"), val = tensor([1, 1])]; + tensor var_5771_pad_0 = const()[name = tensor("op_5771_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5771_dilations_0 = const()[name = tensor("op_5771_dilations_0"), val = tensor([1, 1])]; + tensor var_5771_groups_0 = const()[name = tensor("op_5771_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(85814400))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86207680))), name = tensor("layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_14_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86207872)))]; + tensor var_5771_cast_fp16 = conv(bias = layers_14_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_5771_dilations_0, groups = var_5771_groups_0, pad = var_5771_pad_0, pad_type = var_5771_pad_type_0, strides = var_5771_strides_0, weight = layers_14_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_389_cast_fp16)[name = tensor("op_5771_cast_fp16")]; + tensor var_5777_pad_type_0 = const()[name = tensor("op_5777_pad_type_0"), val = tensor("valid")]; + tensor var_5777_strides_0 = const()[name = tensor("op_5777_strides_0"), val = tensor([1, 1])]; + tensor var_5777_pad_0 = const()[name = tensor("op_5777_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5777_dilations_0 = const()[name = tensor("op_5777_dilations_0"), val = tensor([1, 1])]; + tensor var_5777_groups_0 = const()[name = tensor("op_5777_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86229248))), name = tensor("layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86209984))), shape = tensor([1024, 512, 1, 1])]; + tensor var_5777_cast_fp16 = conv(dilations = var_5777_dilations_0, groups = var_5777_groups_0, pad = var_5777_pad_0, pad_type = var_5777_pad_type_0, strides = var_5777_strides_0, weight = layers_14_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_389_cast_fp16)[name = tensor("op_5777_cast_fp16")]; + tensor input_391_cast_fp16 = add(x = var_5771_cast_fp16, y = var_5777_cast_fp16)[name = tensor("input_391_cast_fp16")]; + tensor input_393_split_num_splits_0 = const()[name = tensor("input_393_split_num_splits_0"), val = tensor(2)]; + tensor input_393_split_axis_0 = const()[name = tensor("input_393_split_axis_0"), val = tensor(1)]; + tensor input_393_split_cast_fp16_0, tensor input_393_split_cast_fp16_1 = split(axis = input_393_split_axis_0, num_splits = input_393_split_num_splits_0, x = input_391_cast_fp16)[name = tensor("input_393_split_cast_fp16")]; + tensor input_393_split_1_sigmoid_cast_fp16 = sigmoid(x = input_393_split_cast_fp16_1)[name = tensor("input_393_split_1_sigmoid_cast_fp16")]; + tensor input_393_cast_fp16 = mul(x = input_393_split_cast_fp16_0, y = input_393_split_1_sigmoid_cast_fp16)[name = tensor("input_393_cast_fp16")]; + tensor input_395_pad_type_0 = const()[name = tensor("input_395_pad_type_0"), val = tensor("custom")]; + tensor input_395_pad_0 = const()[name = tensor("input_395_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_395_groups_0 = const()[name = tensor("input_395_groups_0"), val = tensor(512)]; + tensor input_395_strides_0 = const()[name = tensor("input_395_strides_0"), val = tensor([1, 1])]; + tensor input_395_dilations_0 = const()[name = tensor("input_395_dilations_0"), val = tensor([1, 1])]; + tensor const_219_to_fp16 = const()[name = tensor("const_219_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86294848)))]; + tensor const_220_to_fp16 = const()[name = tensor("const_220_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86304128)))]; + tensor input_397_cast_fp16 = conv(bias = const_220_to_fp16, dilations = input_395_dilations_0, groups = input_395_groups_0, pad = input_395_pad_0, pad_type = input_395_pad_type_0, strides = input_395_strides_0, weight = const_219_to_fp16, x = input_393_cast_fp16)[name = tensor("input_397_cast_fp16")]; + tensor input_399_cast_fp16 = silu(x = input_397_cast_fp16)[name = tensor("input_399_cast_fp16")]; + tensor var_5801_pad_type_0 = const()[name = tensor("op_5801_pad_type_0"), val = tensor("valid")]; + tensor var_5801_strides_0 = const()[name = tensor("op_5801_strides_0"), val = tensor([1, 1])]; + tensor var_5801_pad_0 = const()[name = tensor("op_5801_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5801_dilations_0 = const()[name = tensor("op_5801_dilations_0"), val = tensor([1, 1])]; + tensor var_5801_groups_0 = const()[name = tensor("op_5801_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86305216))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86501888))), name = tensor("layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_14_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86502080)))]; + tensor var_5801_cast_fp16 = conv(bias = layers_14_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_5801_dilations_0, groups = var_5801_groups_0, pad = var_5801_pad_0, pad_type = var_5801_pad_type_0, strides = var_5801_strides_0, weight = layers_14_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_399_cast_fp16)[name = tensor("op_5801_cast_fp16")]; + tensor var_5807_pad_type_0 = const()[name = tensor("op_5807_pad_type_0"), val = tensor("valid")]; + tensor var_5807_strides_0 = const()[name = tensor("op_5807_strides_0"), val = tensor([1, 1])]; + tensor var_5807_pad_0 = const()[name = tensor("op_5807_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5807_dilations_0 = const()[name = tensor("op_5807_dilations_0"), val = tensor([1, 1])]; + tensor var_5807_groups_0 = const()[name = tensor("op_5807_groups_0"), val = tensor(1)]; + tensor layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86511872))), name = tensor("layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86503168))), shape = tensor([512, 512, 1, 1])]; + tensor var_5807_cast_fp16 = conv(dilations = var_5807_dilations_0, groups = var_5807_groups_0, pad = var_5807_pad_0, pad_type = var_5807_pad_type_0, strides = var_5807_strides_0, weight = layers_14_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_399_cast_fp16)[name = tensor("op_5807_cast_fp16")]; + tensor x_89_cast_fp16 = add(x = var_5801_cast_fp16, y = var_5807_cast_fp16)[name = tensor("x_89_cast_fp16")]; + tensor inputs_147_cast_fp16 = add(x = inputs_145_cast_fp16, y = x_89_cast_fp16)[name = tensor("inputs_147_cast_fp16")]; + tensor out_147_axes_0 = const()[name = tensor("out_147_axes_0"), val = tensor([1])]; + tensor var_5818_to_fp16 = const()[name = tensor("op_5818_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_147_cast_fp16 = layer_norm(axes = out_147_axes_0, epsilon = var_5818_to_fp16, x = inputs_147_cast_fp16)[name = tensor("out_147_cast_fp16")]; + tensor input_401_gamma_0_to_fp16 = const()[name = tensor("input_401_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86544704)))]; + tensor input_401_beta_0_to_fp16 = const()[name = tensor("input_401_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86545792)))]; + tensor input_401_epsilon_0_to_fp16 = const()[name = tensor("input_401_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_401_cast_fp16 = batch_norm(beta = input_401_beta_0_to_fp16, epsilon = input_401_epsilon_0_to_fp16, gamma = input_401_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_147_cast_fp16)[name = tensor("input_401_cast_fp16")]; + tensor var_5838_pad_type_0 = const()[name = tensor("op_5838_pad_type_0"), val = tensor("valid")]; + tensor var_5838_strides_0 = const()[name = tensor("op_5838_strides_0"), val = tensor([1, 1])]; + tensor var_5838_pad_0 = const()[name = tensor("op_5838_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5838_dilations_0 = const()[name = tensor("op_5838_dilations_0"), val = tensor([1, 1])]; + tensor var_5838_groups_0 = const()[name = tensor("op_5838_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(86546880))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87333376))), name = tensor("layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_14_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87333568)))]; + tensor var_5838_cast_fp16 = conv(bias = layers_14_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_5838_dilations_0, groups = var_5838_groups_0, pad = var_5838_pad_0, pad_type = var_5838_pad_type_0, strides = var_5838_strides_0, weight = layers_14_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_401_cast_fp16)[name = tensor("op_5838_cast_fp16")]; + tensor var_5844_pad_type_0 = const()[name = tensor("op_5844_pad_type_0"), val = tensor("valid")]; + tensor var_5844_strides_0 = const()[name = tensor("op_5844_strides_0"), val = tensor([1, 1])]; + tensor var_5844_pad_0 = const()[name = tensor("op_5844_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5844_dilations_0 = const()[name = tensor("op_5844_dilations_0"), val = tensor([1, 1])]; + tensor var_5844_groups_0 = const()[name = tensor("op_5844_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87370176))), name = tensor("layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87337728))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5844_cast_fp16 = conv(dilations = var_5844_dilations_0, groups = var_5844_groups_0, pad = var_5844_pad_0, pad_type = var_5844_pad_type_0, strides = var_5844_strides_0, weight = layers_14_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_401_cast_fp16)[name = tensor("op_5844_cast_fp16")]; + tensor input_403_cast_fp16 = add(x = var_5838_cast_fp16, y = var_5844_cast_fp16)[name = tensor("input_403_cast_fp16")]; + tensor input_405_cast_fp16 = silu(x = input_403_cast_fp16)[name = tensor("input_405_cast_fp16")]; + tensor var_5855_pad_type_0 = const()[name = tensor("op_5855_pad_type_0"), val = tensor("valid")]; + tensor var_5855_strides_0 = const()[name = tensor("op_5855_strides_0"), val = tensor([1, 1])]; + tensor var_5855_pad_0 = const()[name = tensor("op_5855_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5855_dilations_0 = const()[name = tensor("op_5855_dilations_0"), val = tensor([1, 1])]; + tensor var_5855_groups_0 = const()[name = tensor("op_5855_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(87501312))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88287808))), name = tensor("layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_14_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_14_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88288000)))]; + tensor var_5855_cast_fp16 = conv(bias = layers_14_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_5855_dilations_0, groups = var_5855_groups_0, pad = var_5855_pad_0, pad_type = var_5855_pad_type_0, strides = var_5855_strides_0, weight = layers_14_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_405_cast_fp16)[name = tensor("op_5855_cast_fp16")]; + tensor var_5861_pad_type_0 = const()[name = tensor("op_5861_pad_type_0"), val = tensor("valid")]; + tensor var_5861_strides_0 = const()[name = tensor("op_5861_strides_0"), val = tensor([1, 1])]; + tensor var_5861_pad_0 = const()[name = tensor("op_5861_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5861_dilations_0 = const()[name = tensor("op_5861_dilations_0"), val = tensor([1, 1])]; + tensor var_5861_groups_0 = const()[name = tensor("op_5861_groups_0"), val = tensor(1)]; + tensor layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88327936))), name = tensor("layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88289088))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5861_cast_fp16 = conv(dilations = var_5861_dilations_0, groups = var_5861_groups_0, pad = var_5861_pad_0, pad_type = var_5861_pad_type_0, strides = var_5861_strides_0, weight = layers_14_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_405_cast_fp16)[name = tensor("op_5861_cast_fp16")]; + tensor x_91_cast_fp16 = add(x = var_5855_cast_fp16, y = var_5861_cast_fp16)[name = tensor("x_91_cast_fp16")]; + tensor var_5863_to_fp16 = const()[name = tensor("op_5863_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5864_cast_fp16 = mul(x = x_91_cast_fp16, y = var_5863_to_fp16)[name = tensor("op_5864_cast_fp16")]; + tensor inputs_149_cast_fp16 = add(x = inputs_147_cast_fp16, y = var_5864_cast_fp16)[name = tensor("inputs_149_cast_fp16")]; + tensor out_149_axes_0 = const()[name = tensor("out_149_axes_0"), val = tensor([1])]; + tensor var_5874_to_fp16 = const()[name = tensor("op_5874_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_149_cast_fp16 = layer_norm(axes = out_149_axes_0, epsilon = var_5874_to_fp16, x = inputs_149_cast_fp16)[name = tensor("out_149_cast_fp16")]; + tensor inputs_151_gamma_0_to_fp16 = const()[name = tensor("inputs_151_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88459072)))]; + tensor inputs_151_beta_0_to_fp16 = const()[name = tensor("inputs_151_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88460160)))]; + tensor inputs_151_epsilon_0_to_fp16 = const()[name = tensor("inputs_151_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_151_cast_fp16 = batch_norm(beta = inputs_151_beta_0_to_fp16, epsilon = inputs_151_epsilon_0_to_fp16, gamma = inputs_151_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_149_cast_fp16)[name = tensor("inputs_151_cast_fp16")]; + tensor var_5888 = const()[name = tensor("op_5888"), val = tensor(3)]; + tensor out_151_axes_0 = const()[name = tensor("out_151_axes_0"), val = tensor([1])]; + tensor var_5919_to_fp16 = const()[name = tensor("op_5919_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_151_cast_fp16 = layer_norm(axes = out_151_axes_0, epsilon = var_5919_to_fp16, x = inputs_151_cast_fp16)[name = tensor("out_151_cast_fp16")]; + tensor input_407_gamma_0_to_fp16 = const()[name = tensor("input_407_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88461248)))]; + tensor input_407_beta_0_to_fp16 = const()[name = tensor("input_407_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88462336)))]; + tensor input_407_epsilon_0_to_fp16 = const()[name = tensor("input_407_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_407_cast_fp16 = batch_norm(beta = input_407_beta_0_to_fp16, epsilon = input_407_epsilon_0_to_fp16, gamma = input_407_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_151_cast_fp16)[name = tensor("input_407_cast_fp16")]; + tensor var_5939_pad_type_0 = const()[name = tensor("op_5939_pad_type_0"), val = tensor("valid")]; + tensor var_5939_strides_0 = const()[name = tensor("op_5939_strides_0"), val = tensor([1, 1])]; + tensor var_5939_pad_0 = const()[name = tensor("op_5939_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5939_dilations_0 = const()[name = tensor("op_5939_dilations_0"), val = tensor([1, 1])]; + tensor var_5939_groups_0 = const()[name = tensor("op_5939_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(88463424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89249920))), name = tensor("layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_15_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89250112)))]; + tensor var_5939_cast_fp16 = conv(bias = layers_15_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_5939_dilations_0, groups = var_5939_groups_0, pad = var_5939_pad_0, pad_type = var_5939_pad_type_0, strides = var_5939_strides_0, weight = layers_15_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_407_cast_fp16)[name = tensor("op_5939_cast_fp16")]; + tensor var_5945_pad_type_0 = const()[name = tensor("op_5945_pad_type_0"), val = tensor("valid")]; + tensor var_5945_strides_0 = const()[name = tensor("op_5945_strides_0"), val = tensor([1, 1])]; + tensor var_5945_pad_0 = const()[name = tensor("op_5945_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5945_dilations_0 = const()[name = tensor("op_5945_dilations_0"), val = tensor([1, 1])]; + tensor var_5945_groups_0 = const()[name = tensor("op_5945_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89285760))), name = tensor("layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89254272))), shape = tensor([2048, 512, 1, 1])]; + tensor var_5945_cast_fp16 = conv(dilations = var_5945_dilations_0, groups = var_5945_groups_0, pad = var_5945_pad_0, pad_type = var_5945_pad_type_0, strides = var_5945_strides_0, weight = layers_15_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_407_cast_fp16)[name = tensor("op_5945_cast_fp16")]; + tensor input_409_cast_fp16 = add(x = var_5939_cast_fp16, y = var_5945_cast_fp16)[name = tensor("input_409_cast_fp16")]; + tensor input_411_cast_fp16 = silu(x = input_409_cast_fp16)[name = tensor("input_411_cast_fp16")]; + tensor var_5956_pad_type_0 = const()[name = tensor("op_5956_pad_type_0"), val = tensor("valid")]; + tensor var_5956_strides_0 = const()[name = tensor("op_5956_strides_0"), val = tensor([1, 1])]; + tensor var_5956_pad_0 = const()[name = tensor("op_5956_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5956_dilations_0 = const()[name = tensor("op_5956_dilations_0"), val = tensor([1, 1])]; + tensor var_5956_groups_0 = const()[name = tensor("op_5956_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(89416896))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90203392))), name = tensor("layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_15_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90203584)))]; + tensor var_5956_cast_fp16 = conv(bias = layers_15_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_5956_dilations_0, groups = var_5956_groups_0, pad = var_5956_pad_0, pad_type = var_5956_pad_type_0, strides = var_5956_strides_0, weight = layers_15_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_411_cast_fp16)[name = tensor("op_5956_cast_fp16")]; + tensor var_5962_pad_type_0 = const()[name = tensor("op_5962_pad_type_0"), val = tensor("valid")]; + tensor var_5962_strides_0 = const()[name = tensor("op_5962_strides_0"), val = tensor([1, 1])]; + tensor var_5962_pad_0 = const()[name = tensor("op_5962_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_5962_dilations_0 = const()[name = tensor("op_5962_dilations_0"), val = tensor([1, 1])]; + tensor var_5962_groups_0 = const()[name = tensor("op_5962_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90244800))), name = tensor("layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90204672))), shape = tensor([512, 2048, 1, 1])]; + tensor var_5962_cast_fp16 = conv(dilations = var_5962_dilations_0, groups = var_5962_groups_0, pad = var_5962_pad_0, pad_type = var_5962_pad_type_0, strides = var_5962_strides_0, weight = layers_15_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_411_cast_fp16)[name = tensor("op_5962_cast_fp16")]; + tensor x_93_cast_fp16 = add(x = var_5956_cast_fp16, y = var_5962_cast_fp16)[name = tensor("x_93_cast_fp16")]; + tensor var_5964_to_fp16 = const()[name = tensor("op_5964_to_fp16"), val = tensor(0x1p-1)]; + tensor var_5965_cast_fp16 = mul(x = x_93_cast_fp16, y = var_5964_to_fp16)[name = tensor("op_5965_cast_fp16")]; + tensor inputs_153_cast_fp16 = add(x = inputs_151_cast_fp16, y = var_5965_cast_fp16)[name = tensor("inputs_153_cast_fp16")]; + tensor out_153_axes_0 = const()[name = tensor("out_153_axes_0"), val = tensor([1])]; + tensor var_5975_to_fp16 = const()[name = tensor("op_5975_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_153_cast_fp16 = layer_norm(axes = out_153_axes_0, epsilon = var_5975_to_fp16, x = inputs_153_cast_fp16)[name = tensor("out_153_cast_fp16")]; + tensor obj_63_gamma_0_to_fp16 = const()[name = tensor("obj_63_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90375936)))]; + tensor obj_63_beta_0_to_fp16 = const()[name = tensor("obj_63_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90377024)))]; + tensor obj_63_epsilon_0_to_fp16 = const()[name = tensor("obj_63_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_63_cast_fp16 = batch_norm(beta = obj_63_beta_0_to_fp16, epsilon = obj_63_epsilon_0_to_fp16, gamma = obj_63_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_153_cast_fp16)[name = tensor("obj_63_cast_fp16")]; + tensor var_6000_pad_type_0 = const()[name = tensor("op_6000_pad_type_0"), val = tensor("valid")]; + tensor var_6000_strides_0 = const()[name = tensor("op_6000_strides_0"), val = tensor([1, 1])]; + tensor var_6000_pad_0 = const()[name = tensor("op_6000_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6000_dilations_0 = const()[name = tensor("op_6000_dilations_0"), val = tensor([1, 1])]; + tensor var_6000_groups_0 = const()[name = tensor("op_6000_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90378112))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90574784))), name = tensor("layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_15_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90574976)))]; + tensor var_6000_cast_fp16 = conv(bias = layers_15_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_6000_dilations_0, groups = var_6000_groups_0, pad = var_6000_pad_0, pad_type = var_6000_pad_type_0, strides = var_6000_strides_0, weight = layers_15_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_6000_cast_fp16")]; + tensor var_6006_pad_type_0 = const()[name = tensor("op_6006_pad_type_0"), val = tensor("valid")]; + tensor var_6006_strides_0 = const()[name = tensor("op_6006_strides_0"), val = tensor([1, 1])]; + tensor var_6006_pad_0 = const()[name = tensor("op_6006_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6006_dilations_0 = const()[name = tensor("op_6006_dilations_0"), val = tensor([1, 1])]; + tensor var_6006_groups_0 = const()[name = tensor("op_6006_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90583680))), name = tensor("layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90576064))), shape = tensor([512, 512, 1, 1])]; + tensor var_6006_cast_fp16 = conv(dilations = var_6006_dilations_0, groups = var_6006_groups_0, pad = var_6006_pad_0, pad_type = var_6006_pad_type_0, strides = var_6006_strides_0, weight = layers_15_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_6006_cast_fp16")]; + tensor query_61_cast_fp16 = add(x = var_6000_cast_fp16, y = var_6006_cast_fp16)[name = tensor("query_61_cast_fp16")]; + tensor var_6015_pad_type_0 = const()[name = tensor("op_6015_pad_type_0"), val = tensor("valid")]; + tensor var_6015_strides_0 = const()[name = tensor("op_6015_strides_0"), val = tensor([1, 1])]; + tensor var_6015_pad_0 = const()[name = tensor("op_6015_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6015_dilations_0 = const()[name = tensor("op_6015_dilations_0"), val = tensor([1, 1])]; + tensor var_6015_groups_0 = const()[name = tensor("op_6015_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90616512))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90813184))), name = tensor("layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_6015_cast_fp16 = conv(dilations = var_6015_dilations_0, groups = var_6015_groups_0, pad = var_6015_pad_0, pad_type = var_6015_pad_type_0, strides = var_6015_strides_0, weight = layers_15_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_6015_cast_fp16")]; + tensor var_6021_pad_type_0 = const()[name = tensor("op_6021_pad_type_0"), val = tensor("valid")]; + tensor var_6021_strides_0 = const()[name = tensor("op_6021_strides_0"), val = tensor([1, 1])]; + tensor var_6021_pad_0 = const()[name = tensor("op_6021_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6021_dilations_0 = const()[name = tensor("op_6021_dilations_0"), val = tensor([1, 1])]; + tensor var_6021_groups_0 = const()[name = tensor("op_6021_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90821184))), name = tensor("layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90813376))), shape = tensor([512, 512, 1, 1])]; + tensor var_6021_cast_fp16 = conv(dilations = var_6021_dilations_0, groups = var_6021_groups_0, pad = var_6021_pad_0, pad_type = var_6021_pad_type_0, strides = var_6021_strides_0, weight = layers_15_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_6021_cast_fp16")]; + tensor key_31_cast_fp16 = add(x = var_6015_cast_fp16, y = var_6021_cast_fp16)[name = tensor("key_31_cast_fp16")]; + tensor var_6031_pad_type_0 = const()[name = tensor("op_6031_pad_type_0"), val = tensor("valid")]; + tensor var_6031_strides_0 = const()[name = tensor("op_6031_strides_0"), val = tensor([1, 1])]; + tensor var_6031_pad_0 = const()[name = tensor("op_6031_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6031_dilations_0 = const()[name = tensor("op_6031_dilations_0"), val = tensor([1, 1])]; + tensor var_6031_groups_0 = const()[name = tensor("op_6031_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(90854016))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91050688))), name = tensor("layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_15_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91050880)))]; + tensor var_6031_cast_fp16 = conv(bias = layers_15_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_6031_dilations_0, groups = var_6031_groups_0, pad = var_6031_pad_0, pad_type = var_6031_pad_type_0, strides = var_6031_strides_0, weight = layers_15_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_63_cast_fp16)[name = tensor("op_6031_cast_fp16")]; + tensor var_6037_pad_type_0 = const()[name = tensor("op_6037_pad_type_0"), val = tensor("valid")]; + tensor var_6037_strides_0 = const()[name = tensor("op_6037_strides_0"), val = tensor([1, 1])]; + tensor var_6037_pad_0 = const()[name = tensor("op_6037_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6037_dilations_0 = const()[name = tensor("op_6037_dilations_0"), val = tensor([1, 1])]; + tensor var_6037_groups_0 = const()[name = tensor("op_6037_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91060672))), name = tensor("layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91051968))), shape = tensor([512, 512, 1, 1])]; + tensor var_6037_cast_fp16 = conv(dilations = var_6037_dilations_0, groups = var_6037_groups_0, pad = var_6037_pad_0, pad_type = var_6037_pad_type_0, strides = var_6037_strides_0, weight = layers_15_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_63_cast_fp16)[name = tensor("op_6037_cast_fp16")]; + tensor value_31_cast_fp16 = add(x = var_6031_cast_fp16, y = var_6037_cast_fp16)[name = tensor("value_31_cast_fp16")]; + tensor var_6040_to_fp16 = const()[name = tensor("op_6040_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91093504)))]; + tensor query_63_cast_fp16 = add(x = query_61_cast_fp16, y = var_6040_to_fp16)[name = tensor("query_63_cast_fp16")]; + tensor var_6043_to_fp16 = const()[name = tensor("op_6043_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91094592)))]; + tensor q_with_bias_v_31_cast_fp16 = add(x = query_61_cast_fp16, y = var_6043_to_fp16)[name = tensor("q_with_bias_v_31_cast_fp16")]; + tensor var_6053_pad_type_0 = const()[name = tensor("op_6053_pad_type_0"), val = tensor("valid")]; + tensor var_6053_strides_0 = const()[name = tensor("op_6053_strides_0"), val = tensor([1, 1])]; + tensor var_6053_pad_0 = const()[name = tensor("op_6053_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6053_dilations_0 = const()[name = tensor("op_6053_dilations_0"), val = tensor([1, 1])]; + tensor var_6053_groups_0 = const()[name = tensor("op_6053_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91095680))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91292352))), name = tensor("layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_6053_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6053_dilations_0, groups = var_6053_groups_0, pad = var_6053_pad_0, pad_type = var_6053_pad_type_0, strides = var_6053_strides_0, weight = layers_15_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_6053_cast_fp16")]; + tensor var_6059_pad_type_0 = const()[name = tensor("op_6059_pad_type_0"), val = tensor("valid")]; + tensor var_6059_strides_0 = const()[name = tensor("op_6059_strides_0"), val = tensor([1, 1])]; + tensor var_6059_pad_0 = const()[name = tensor("op_6059_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6059_dilations_0 = const()[name = tensor("op_6059_dilations_0"), val = tensor([1, 1])]; + tensor var_6059_groups_0 = const()[name = tensor("op_6059_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91315264))), name = tensor("layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91292544))), shape = tensor([512, 512, 1, 1])]; + tensor var_6059_cast_fp16 = conv(dilations = var_6059_dilations_0, groups = var_6059_groups_0, pad = var_6059_pad_0, pad_type = var_6059_pad_type_0, strides = var_6059_strides_0, weight = layers_15_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_6059_cast_fp16")]; + tensor p_31_cast_fp16 = add(x = var_6053_cast_fp16, y = var_6059_cast_fp16)[name = tensor("p_31_cast_fp16")]; + tensor var_6063 = const()[name = tensor("op_6063"), val = tensor([1, 8, 64, 188])]; + tensor var_6064_cast_fp16 = reshape(shape = var_6063, x = q_with_bias_v_31_cast_fp16)[name = tensor("op_6064_cast_fp16")]; + tensor var_6065 = const()[name = tensor("op_6065"), val = tensor([1, 8, 64, -1])]; + tensor var_6066_cast_fp16 = reshape(shape = var_6065, x = p_31_cast_fp16)[name = tensor("op_6066_cast_fp16")]; + tensor matrix_bd_121_transpose_x_0 = const()[name = tensor("matrix_bd_121_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_121_transpose_y_0 = const()[name = tensor("matrix_bd_121_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_121_cast_fp16 = matmul(transpose_x = matrix_bd_121_transpose_x_0, transpose_y = matrix_bd_121_transpose_y_0, x = var_6064_cast_fp16, y = var_6066_cast_fp16)[name = tensor("matrix_bd_121_cast_fp16")]; + tensor matrix_bd_123_pad_0 = const()[name = tensor("matrix_bd_123_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_123_mode_0 = const()[name = tensor("matrix_bd_123_mode_0"), val = tensor("constant")]; + tensor const_175_to_fp16 = const()[name = tensor("const_175_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_123_cast_fp16 = pad(constant_val = const_175_to_fp16, mode = matrix_bd_123_mode_0, pad = matrix_bd_123_pad_0, x = matrix_bd_121_cast_fp16)[name = tensor("matrix_bd_123_cast_fp16")]; + tensor var_6075 = const()[name = tensor("op_6075"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_125_cast_fp16 = reshape(shape = var_6075, x = matrix_bd_123_cast_fp16)[name = tensor("matrix_bd_125_cast_fp16")]; + tensor var_6079_begin_0 = const()[name = tensor("op_6079_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_6079_end_0 = const()[name = tensor("op_6079_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_6079_end_mask_0 = const()[name = tensor("op_6079_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6079_cast_fp16 = slice_by_index(begin = var_6079_begin_0, end = var_6079_end_0, end_mask = var_6079_end_mask_0, x = matrix_bd_125_cast_fp16)[name = tensor("op_6079_cast_fp16")]; + tensor var_6080 = const()[name = tensor("op_6080"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_127_cast_fp16 = reshape(shape = var_6080, x = var_6079_cast_fp16)[name = tensor("matrix_bd_127_cast_fp16")]; + tensor var_6085_begin_0 = const()[name = tensor("op_6085_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6085_end_0 = const()[name = tensor("op_6085_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_6085_end_mask_0 = const()[name = tensor("op_6085_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6085_cast_fp16 = slice_by_index(begin = var_6085_begin_0, end = var_6085_end_0, end_mask = var_6085_end_mask_0, x = matrix_bd_127_cast_fp16)[name = tensor("op_6085_cast_fp16")]; + tensor var_6086_to_fp16 = const()[name = tensor("op_6086_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_31_cast_fp16 = mul(x = var_6085_cast_fp16, y = var_6086_to_fp16)[name = tensor("qk_mask_31_cast_fp16")]; + tensor var_6090 = const()[name = tensor("op_6090"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_31_cast_fp16 = reshape(shape = var_6090, x = query_63_cast_fp16)[name = tensor("mh_q_31_cast_fp16")]; + tensor var_6092_to_fp16 = const()[name = tensor("op_6092_to_fp16"), val = tensor(0x1p-3)]; + tensor var_6093_cast_fp16 = mul(x = mh_q_31_cast_fp16, y = var_6092_to_fp16)[name = tensor("op_6093_cast_fp16")]; + tensor var_6096 = const()[name = tensor("op_6096"), val = tensor([1, 8, 64, 188])]; + tensor var_6097_cast_fp16 = reshape(shape = var_6096, x = key_31_cast_fp16)[name = tensor("op_6097_cast_fp16")]; + tensor mh_w_61_transpose_x_0 = const()[name = tensor("mh_w_61_transpose_x_0"), val = tensor(true)]; + tensor mh_w_61_transpose_y_0 = const()[name = tensor("mh_w_61_transpose_y_0"), val = tensor(false)]; + tensor mh_w_61_cast_fp16 = matmul(transpose_x = mh_w_61_transpose_x_0, transpose_y = mh_w_61_transpose_y_0, x = var_6093_cast_fp16, y = var_6097_cast_fp16)[name = tensor("mh_w_61_cast_fp16")]; + tensor mh_w_63_cast_fp16 = add(x = mh_w_61_cast_fp16, y = qk_mask_31_cast_fp16)[name = tensor("mh_w_63_cast_fp16")]; + tensor var_6101_cast_fp16 = softmax(axis = var_5888, x = mh_w_63_cast_fp16)[name = tensor("op_6101_cast_fp16")]; + tensor var_6102 = const()[name = tensor("op_6102"), val = tensor([1, 8, 64, 188])]; + tensor var_6103_cast_fp16 = reshape(shape = var_6102, x = value_31_cast_fp16)[name = tensor("op_6103_cast_fp16")]; + tensor attn_31_transpose_x_0 = const()[name = tensor("attn_31_transpose_x_0"), val = tensor(false)]; + tensor attn_31_transpose_y_0 = const()[name = tensor("attn_31_transpose_y_0"), val = tensor(true)]; + tensor attn_31_cast_fp16 = matmul(transpose_x = attn_31_transpose_x_0, transpose_y = attn_31_transpose_y_0, x = var_6103_cast_fp16, y = var_6101_cast_fp16)[name = tensor("attn_31_cast_fp16")]; + tensor var_6106 = const()[name = tensor("op_6106"), val = tensor([1, 512, 1, 188])]; + tensor input_413_cast_fp16 = reshape(shape = var_6106, x = attn_31_cast_fp16)[name = tensor("input_413_cast_fp16")]; + tensor var_6116_pad_type_0 = const()[name = tensor("op_6116_pad_type_0"), val = tensor("valid")]; + tensor var_6116_strides_0 = const()[name = tensor("op_6116_strides_0"), val = tensor([1, 1])]; + tensor var_6116_pad_0 = const()[name = tensor("op_6116_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6116_dilations_0 = const()[name = tensor("op_6116_dilations_0"), val = tensor([1, 1])]; + tensor var_6116_groups_0 = const()[name = tensor("op_6116_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91348096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91544768))), name = tensor("layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_15_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91544960)))]; + tensor var_6116_cast_fp16 = conv(bias = layers_15_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_6116_dilations_0, groups = var_6116_groups_0, pad = var_6116_pad_0, pad_type = var_6116_pad_type_0, strides = var_6116_strides_0, weight = layers_15_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_413_cast_fp16)[name = tensor("op_6116_cast_fp16")]; + tensor var_6122_pad_type_0 = const()[name = tensor("op_6122_pad_type_0"), val = tensor("valid")]; + tensor var_6122_strides_0 = const()[name = tensor("op_6122_strides_0"), val = tensor([1, 1])]; + tensor var_6122_pad_0 = const()[name = tensor("op_6122_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6122_dilations_0 = const()[name = tensor("op_6122_dilations_0"), val = tensor([1, 1])]; + tensor var_6122_groups_0 = const()[name = tensor("op_6122_groups_0"), val = tensor(1)]; + tensor layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91554496))), name = tensor("layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91546048))), shape = tensor([512, 512, 1, 1])]; + tensor var_6122_cast_fp16 = conv(dilations = var_6122_dilations_0, groups = var_6122_groups_0, pad = var_6122_pad_0, pad_type = var_6122_pad_type_0, strides = var_6122_strides_0, weight = layers_15_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_413_cast_fp16)[name = tensor("op_6122_cast_fp16")]; + tensor obj_65_cast_fp16 = add(x = var_6116_cast_fp16, y = var_6122_cast_fp16)[name = tensor("obj_65_cast_fp16")]; + tensor inputs_155_cast_fp16 = add(x = inputs_153_cast_fp16, y = obj_65_cast_fp16)[name = tensor("inputs_155_cast_fp16")]; + tensor out_155_axes_0 = const()[name = tensor("out_155_axes_0"), val = tensor([1])]; + tensor var_6133_to_fp16 = const()[name = tensor("op_6133_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_155_cast_fp16 = layer_norm(axes = out_155_axes_0, epsilon = var_6133_to_fp16, x = inputs_155_cast_fp16)[name = tensor("out_155_cast_fp16")]; + tensor input_415_gamma_0_to_fp16 = const()[name = tensor("input_415_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91587328)))]; + tensor input_415_beta_0_to_fp16 = const()[name = tensor("input_415_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91588416)))]; + tensor input_415_epsilon_0_to_fp16 = const()[name = tensor("input_415_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_415_cast_fp16 = batch_norm(beta = input_415_beta_0_to_fp16, epsilon = input_415_epsilon_0_to_fp16, gamma = input_415_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_155_cast_fp16)[name = tensor("input_415_cast_fp16")]; + tensor var_6155_pad_type_0 = const()[name = tensor("op_6155_pad_type_0"), val = tensor("valid")]; + tensor var_6155_strides_0 = const()[name = tensor("op_6155_strides_0"), val = tensor([1, 1])]; + tensor var_6155_pad_0 = const()[name = tensor("op_6155_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6155_dilations_0 = const()[name = tensor("op_6155_dilations_0"), val = tensor([1, 1])]; + tensor var_6155_groups_0 = const()[name = tensor("op_6155_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91589504))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91982784))), name = tensor("layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_15_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91982976)))]; + tensor var_6155_cast_fp16 = conv(bias = layers_15_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_6155_dilations_0, groups = var_6155_groups_0, pad = var_6155_pad_0, pad_type = var_6155_pad_type_0, strides = var_6155_strides_0, weight = layers_15_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_415_cast_fp16)[name = tensor("op_6155_cast_fp16")]; + tensor var_6161_pad_type_0 = const()[name = tensor("op_6161_pad_type_0"), val = tensor("valid")]; + tensor var_6161_strides_0 = const()[name = tensor("op_6161_strides_0"), val = tensor([1, 1])]; + tensor var_6161_pad_0 = const()[name = tensor("op_6161_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6161_dilations_0 = const()[name = tensor("op_6161_dilations_0"), val = tensor([1, 1])]; + tensor var_6161_groups_0 = const()[name = tensor("op_6161_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92003456))), name = tensor("layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(91985088))), shape = tensor([1024, 512, 1, 1])]; + tensor var_6161_cast_fp16 = conv(dilations = var_6161_dilations_0, groups = var_6161_groups_0, pad = var_6161_pad_0, pad_type = var_6161_pad_type_0, strides = var_6161_strides_0, weight = layers_15_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_415_cast_fp16)[name = tensor("op_6161_cast_fp16")]; + tensor input_417_cast_fp16 = add(x = var_6155_cast_fp16, y = var_6161_cast_fp16)[name = tensor("input_417_cast_fp16")]; + tensor input_419_split_num_splits_0 = const()[name = tensor("input_419_split_num_splits_0"), val = tensor(2)]; + tensor input_419_split_axis_0 = const()[name = tensor("input_419_split_axis_0"), val = tensor(1)]; + tensor input_419_split_cast_fp16_0, tensor input_419_split_cast_fp16_1 = split(axis = input_419_split_axis_0, num_splits = input_419_split_num_splits_0, x = input_417_cast_fp16)[name = tensor("input_419_split_cast_fp16")]; + tensor input_419_split_1_sigmoid_cast_fp16 = sigmoid(x = input_419_split_cast_fp16_1)[name = tensor("input_419_split_1_sigmoid_cast_fp16")]; + tensor input_419_cast_fp16 = mul(x = input_419_split_cast_fp16_0, y = input_419_split_1_sigmoid_cast_fp16)[name = tensor("input_419_cast_fp16")]; + tensor input_421_pad_type_0 = const()[name = tensor("input_421_pad_type_0"), val = tensor("custom")]; + tensor input_421_pad_0 = const()[name = tensor("input_421_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_421_groups_0 = const()[name = tensor("input_421_groups_0"), val = tensor(512)]; + tensor input_421_strides_0 = const()[name = tensor("input_421_strides_0"), val = tensor([1, 1])]; + tensor input_421_dilations_0 = const()[name = tensor("input_421_dilations_0"), val = tensor([1, 1])]; + tensor const_221_to_fp16 = const()[name = tensor("const_221_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92069056)))]; + tensor const_222_to_fp16 = const()[name = tensor("const_222_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92078336)))]; + tensor input_423_cast_fp16 = conv(bias = const_222_to_fp16, dilations = input_421_dilations_0, groups = input_421_groups_0, pad = input_421_pad_0, pad_type = input_421_pad_type_0, strides = input_421_strides_0, weight = const_221_to_fp16, x = input_419_cast_fp16)[name = tensor("input_423_cast_fp16")]; + tensor input_425_cast_fp16 = silu(x = input_423_cast_fp16)[name = tensor("input_425_cast_fp16")]; + tensor var_6185_pad_type_0 = const()[name = tensor("op_6185_pad_type_0"), val = tensor("valid")]; + tensor var_6185_strides_0 = const()[name = tensor("op_6185_strides_0"), val = tensor([1, 1])]; + tensor var_6185_pad_0 = const()[name = tensor("op_6185_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6185_dilations_0 = const()[name = tensor("op_6185_dilations_0"), val = tensor([1, 1])]; + tensor var_6185_groups_0 = const()[name = tensor("op_6185_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92079424))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92276096))), name = tensor("layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_15_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92276288)))]; + tensor var_6185_cast_fp16 = conv(bias = layers_15_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_6185_dilations_0, groups = var_6185_groups_0, pad = var_6185_pad_0, pad_type = var_6185_pad_type_0, strides = var_6185_strides_0, weight = layers_15_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_425_cast_fp16)[name = tensor("op_6185_cast_fp16")]; + tensor var_6191_pad_type_0 = const()[name = tensor("op_6191_pad_type_0"), val = tensor("valid")]; + tensor var_6191_strides_0 = const()[name = tensor("op_6191_strides_0"), val = tensor([1, 1])]; + tensor var_6191_pad_0 = const()[name = tensor("op_6191_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6191_dilations_0 = const()[name = tensor("op_6191_dilations_0"), val = tensor([1, 1])]; + tensor var_6191_groups_0 = const()[name = tensor("op_6191_groups_0"), val = tensor(1)]; + tensor layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92285568))), name = tensor("layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92277376))), shape = tensor([512, 512, 1, 1])]; + tensor var_6191_cast_fp16 = conv(dilations = var_6191_dilations_0, groups = var_6191_groups_0, pad = var_6191_pad_0, pad_type = var_6191_pad_type_0, strides = var_6191_strides_0, weight = layers_15_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_425_cast_fp16)[name = tensor("op_6191_cast_fp16")]; + tensor x_95_cast_fp16 = add(x = var_6185_cast_fp16, y = var_6191_cast_fp16)[name = tensor("x_95_cast_fp16")]; + tensor inputs_157_cast_fp16 = add(x = inputs_155_cast_fp16, y = x_95_cast_fp16)[name = tensor("inputs_157_cast_fp16")]; + tensor out_157_axes_0 = const()[name = tensor("out_157_axes_0"), val = tensor([1])]; + tensor var_6202_to_fp16 = const()[name = tensor("op_6202_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_157_cast_fp16 = layer_norm(axes = out_157_axes_0, epsilon = var_6202_to_fp16, x = inputs_157_cast_fp16)[name = tensor("out_157_cast_fp16")]; + tensor input_427_gamma_0_to_fp16 = const()[name = tensor("input_427_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92318400)))]; + tensor input_427_beta_0_to_fp16 = const()[name = tensor("input_427_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92319488)))]; + tensor input_427_epsilon_0_to_fp16 = const()[name = tensor("input_427_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_427_cast_fp16 = batch_norm(beta = input_427_beta_0_to_fp16, epsilon = input_427_epsilon_0_to_fp16, gamma = input_427_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_157_cast_fp16)[name = tensor("input_427_cast_fp16")]; + tensor var_6222_pad_type_0 = const()[name = tensor("op_6222_pad_type_0"), val = tensor("valid")]; + tensor var_6222_strides_0 = const()[name = tensor("op_6222_strides_0"), val = tensor([1, 1])]; + tensor var_6222_pad_0 = const()[name = tensor("op_6222_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6222_dilations_0 = const()[name = tensor("op_6222_dilations_0"), val = tensor([1, 1])]; + tensor var_6222_groups_0 = const()[name = tensor("op_6222_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(92320576))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93107072))), name = tensor("layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_15_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93107264)))]; + tensor var_6222_cast_fp16 = conv(bias = layers_15_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_6222_dilations_0, groups = var_6222_groups_0, pad = var_6222_pad_0, pad_type = var_6222_pad_type_0, strides = var_6222_strides_0, weight = layers_15_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_427_cast_fp16)[name = tensor("op_6222_cast_fp16")]; + tensor var_6228_pad_type_0 = const()[name = tensor("op_6228_pad_type_0"), val = tensor("valid")]; + tensor var_6228_strides_0 = const()[name = tensor("op_6228_strides_0"), val = tensor([1, 1])]; + tensor var_6228_pad_0 = const()[name = tensor("op_6228_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6228_dilations_0 = const()[name = tensor("op_6228_dilations_0"), val = tensor([1, 1])]; + tensor var_6228_groups_0 = const()[name = tensor("op_6228_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93144704))), name = tensor("layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93111424))), shape = tensor([2048, 512, 1, 1])]; + tensor var_6228_cast_fp16 = conv(dilations = var_6228_dilations_0, groups = var_6228_groups_0, pad = var_6228_pad_0, pad_type = var_6228_pad_type_0, strides = var_6228_strides_0, weight = layers_15_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_427_cast_fp16)[name = tensor("op_6228_cast_fp16")]; + tensor input_429_cast_fp16 = add(x = var_6222_cast_fp16, y = var_6228_cast_fp16)[name = tensor("input_429_cast_fp16")]; + tensor input_431_cast_fp16 = silu(x = input_429_cast_fp16)[name = tensor("input_431_cast_fp16")]; + tensor var_6239_pad_type_0 = const()[name = tensor("op_6239_pad_type_0"), val = tensor("valid")]; + tensor var_6239_strides_0 = const()[name = tensor("op_6239_strides_0"), val = tensor([1, 1])]; + tensor var_6239_pad_0 = const()[name = tensor("op_6239_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6239_dilations_0 = const()[name = tensor("op_6239_dilations_0"), val = tensor([1, 1])]; + tensor var_6239_groups_0 = const()[name = tensor("op_6239_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(93275840))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94062336))), name = tensor("layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_15_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_15_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94062528)))]; + tensor var_6239_cast_fp16 = conv(bias = layers_15_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_6239_dilations_0, groups = var_6239_groups_0, pad = var_6239_pad_0, pad_type = var_6239_pad_type_0, strides = var_6239_strides_0, weight = layers_15_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_431_cast_fp16)[name = tensor("op_6239_cast_fp16")]; + tensor var_6245_pad_type_0 = const()[name = tensor("op_6245_pad_type_0"), val = tensor("valid")]; + tensor var_6245_strides_0 = const()[name = tensor("op_6245_strides_0"), val = tensor([1, 1])]; + tensor var_6245_pad_0 = const()[name = tensor("op_6245_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6245_dilations_0 = const()[name = tensor("op_6245_dilations_0"), val = tensor([1, 1])]; + tensor var_6245_groups_0 = const()[name = tensor("op_6245_groups_0"), val = tensor(1)]; + tensor layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94104576))), name = tensor("layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94063616))), shape = tensor([512, 2048, 1, 1])]; + tensor var_6245_cast_fp16 = conv(dilations = var_6245_dilations_0, groups = var_6245_groups_0, pad = var_6245_pad_0, pad_type = var_6245_pad_type_0, strides = var_6245_strides_0, weight = layers_15_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_431_cast_fp16)[name = tensor("op_6245_cast_fp16")]; + tensor x_97_cast_fp16 = add(x = var_6239_cast_fp16, y = var_6245_cast_fp16)[name = tensor("x_97_cast_fp16")]; + tensor var_6247_to_fp16 = const()[name = tensor("op_6247_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6248_cast_fp16 = mul(x = x_97_cast_fp16, y = var_6247_to_fp16)[name = tensor("op_6248_cast_fp16")]; + tensor inputs_159_cast_fp16 = add(x = inputs_157_cast_fp16, y = var_6248_cast_fp16)[name = tensor("inputs_159_cast_fp16")]; + tensor out_159_axes_0 = const()[name = tensor("out_159_axes_0"), val = tensor([1])]; + tensor var_6258_to_fp16 = const()[name = tensor("op_6258_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_159_cast_fp16 = layer_norm(axes = out_159_axes_0, epsilon = var_6258_to_fp16, x = inputs_159_cast_fp16)[name = tensor("out_159_cast_fp16")]; + tensor inputs_161_gamma_0_to_fp16 = const()[name = tensor("inputs_161_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94235712)))]; + tensor inputs_161_beta_0_to_fp16 = const()[name = tensor("inputs_161_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94236800)))]; + tensor inputs_161_epsilon_0_to_fp16 = const()[name = tensor("inputs_161_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor inputs_161_cast_fp16 = batch_norm(beta = inputs_161_beta_0_to_fp16, epsilon = inputs_161_epsilon_0_to_fp16, gamma = inputs_161_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_159_cast_fp16)[name = tensor("inputs_161_cast_fp16")]; + tensor var_6272 = const()[name = tensor("op_6272"), val = tensor(3)]; + tensor out_161_axes_0 = const()[name = tensor("out_161_axes_0"), val = tensor([1])]; + tensor var_6303_to_fp16 = const()[name = tensor("op_6303_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_161_cast_fp16 = layer_norm(axes = out_161_axes_0, epsilon = var_6303_to_fp16, x = inputs_161_cast_fp16)[name = tensor("out_161_cast_fp16")]; + tensor input_433_gamma_0_to_fp16 = const()[name = tensor("input_433_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94237888)))]; + tensor input_433_beta_0_to_fp16 = const()[name = tensor("input_433_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94238976)))]; + tensor input_433_epsilon_0_to_fp16 = const()[name = tensor("input_433_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_433_cast_fp16 = batch_norm(beta = input_433_beta_0_to_fp16, epsilon = input_433_epsilon_0_to_fp16, gamma = input_433_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_161_cast_fp16)[name = tensor("input_433_cast_fp16")]; + tensor var_6323_pad_type_0 = const()[name = tensor("op_6323_pad_type_0"), val = tensor("valid")]; + tensor var_6323_strides_0 = const()[name = tensor("op_6323_strides_0"), val = tensor([1, 1])]; + tensor var_6323_pad_0 = const()[name = tensor("op_6323_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6323_dilations_0 = const()[name = tensor("op_6323_dilations_0"), val = tensor([1, 1])]; + tensor var_6323_groups_0 = const()[name = tensor("op_6323_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(94240064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95026560))), name = tensor("layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_16_feed_forward1_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_feed_forward1_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95026752)))]; + tensor var_6323_cast_fp16 = conv(bias = layers_16_feed_forward1_fc1_inlier_module_bias_to_fp16, dilations = var_6323_dilations_0, groups = var_6323_groups_0, pad = var_6323_pad_0, pad_type = var_6323_pad_type_0, strides = var_6323_strides_0, weight = layers_16_feed_forward1_fc1_inlier_module_weight_to_fp16_palettized, x = input_433_cast_fp16)[name = tensor("op_6323_cast_fp16")]; + tensor var_6329_pad_type_0 = const()[name = tensor("op_6329_pad_type_0"), val = tensor("valid")]; + tensor var_6329_strides_0 = const()[name = tensor("op_6329_strides_0"), val = tensor([1, 1])]; + tensor var_6329_pad_0 = const()[name = tensor("op_6329_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6329_dilations_0 = const()[name = tensor("op_6329_dilations_0"), val = tensor([1, 1])]; + tensor var_6329_groups_0 = const()[name = tensor("op_6329_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95064576))), name = tensor("layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95030912))), shape = tensor([2048, 512, 1, 1])]; + tensor var_6329_cast_fp16 = conv(dilations = var_6329_dilations_0, groups = var_6329_groups_0, pad = var_6329_pad_0, pad_type = var_6329_pad_type_0, strides = var_6329_strides_0, weight = layers_16_feed_forward1_fc1_outlier_module_weight_to_fp16_sparsified, x = input_433_cast_fp16)[name = tensor("op_6329_cast_fp16")]; + tensor input_435_cast_fp16 = add(x = var_6323_cast_fp16, y = var_6329_cast_fp16)[name = tensor("input_435_cast_fp16")]; + tensor input_437_cast_fp16 = silu(x = input_435_cast_fp16)[name = tensor("input_437_cast_fp16")]; + tensor var_6340_pad_type_0 = const()[name = tensor("op_6340_pad_type_0"), val = tensor("valid")]; + tensor var_6340_strides_0 = const()[name = tensor("op_6340_strides_0"), val = tensor([1, 1])]; + tensor var_6340_pad_0 = const()[name = tensor("op_6340_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6340_dilations_0 = const()[name = tensor("op_6340_dilations_0"), val = tensor([1, 1])]; + tensor var_6340_groups_0 = const()[name = tensor("op_6340_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95195712))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95982208))), name = tensor("layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_16_feed_forward1_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_feed_forward1_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95982400)))]; + tensor var_6340_cast_fp16 = conv(bias = layers_16_feed_forward1_fc2_inlier_module_bias_to_fp16, dilations = var_6340_dilations_0, groups = var_6340_groups_0, pad = var_6340_pad_0, pad_type = var_6340_pad_type_0, strides = var_6340_strides_0, weight = layers_16_feed_forward1_fc2_inlier_module_weight_to_fp16_palettized, x = input_437_cast_fp16)[name = tensor("op_6340_cast_fp16")]; + tensor var_6346_pad_type_0 = const()[name = tensor("op_6346_pad_type_0"), val = tensor("valid")]; + tensor var_6346_strides_0 = const()[name = tensor("op_6346_strides_0"), val = tensor([1, 1])]; + tensor var_6346_pad_0 = const()[name = tensor("op_6346_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6346_dilations_0 = const()[name = tensor("op_6346_dilations_0"), val = tensor([1, 1])]; + tensor var_6346_groups_0 = const()[name = tensor("op_6346_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96026752))), name = tensor("layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(95983488))), shape = tensor([512, 2048, 1, 1])]; + tensor var_6346_cast_fp16 = conv(dilations = var_6346_dilations_0, groups = var_6346_groups_0, pad = var_6346_pad_0, pad_type = var_6346_pad_type_0, strides = var_6346_strides_0, weight = layers_16_feed_forward1_fc2_outlier_module_weight_to_fp16_sparsified, x = input_437_cast_fp16)[name = tensor("op_6346_cast_fp16")]; + tensor x_99_cast_fp16 = add(x = var_6340_cast_fp16, y = var_6346_cast_fp16)[name = tensor("x_99_cast_fp16")]; + tensor var_6348_to_fp16 = const()[name = tensor("op_6348_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6349_cast_fp16 = mul(x = x_99_cast_fp16, y = var_6348_to_fp16)[name = tensor("op_6349_cast_fp16")]; + tensor inputs_163_cast_fp16 = add(x = inputs_161_cast_fp16, y = var_6349_cast_fp16)[name = tensor("inputs_163_cast_fp16")]; + tensor out_163_axes_0 = const()[name = tensor("out_163_axes_0"), val = tensor([1])]; + tensor var_6359_to_fp16 = const()[name = tensor("op_6359_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_163_cast_fp16 = layer_norm(axes = out_163_axes_0, epsilon = var_6359_to_fp16, x = inputs_163_cast_fp16)[name = tensor("out_163_cast_fp16")]; + tensor obj_67_gamma_0_to_fp16 = const()[name = tensor("obj_67_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96157888)))]; + tensor obj_67_beta_0_to_fp16 = const()[name = tensor("obj_67_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96158976)))]; + tensor obj_67_epsilon_0_to_fp16 = const()[name = tensor("obj_67_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor obj_67_cast_fp16 = batch_norm(beta = obj_67_beta_0_to_fp16, epsilon = obj_67_epsilon_0_to_fp16, gamma = obj_67_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_163_cast_fp16)[name = tensor("obj_67_cast_fp16")]; + tensor var_6384_pad_type_0 = const()[name = tensor("op_6384_pad_type_0"), val = tensor("valid")]; + tensor var_6384_strides_0 = const()[name = tensor("op_6384_strides_0"), val = tensor([1, 1])]; + tensor var_6384_pad_0 = const()[name = tensor("op_6384_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6384_dilations_0 = const()[name = tensor("op_6384_dilations_0"), val = tensor([1, 1])]; + tensor var_6384_groups_0 = const()[name = tensor("op_6384_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96160064))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96356736))), name = tensor("layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_16_self_attn_q_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_q_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96356928)))]; + tensor var_6384_cast_fp16 = conv(bias = layers_16_self_attn_q_proj_inlier_module_bias_to_fp16, dilations = var_6384_dilations_0, groups = var_6384_groups_0, pad = var_6384_pad_0, pad_type = var_6384_pad_type_0, strides = var_6384_strides_0, weight = layers_16_self_attn_q_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6384_cast_fp16")]; + tensor var_6390_pad_type_0 = const()[name = tensor("op_6390_pad_type_0"), val = tensor("valid")]; + tensor var_6390_strides_0 = const()[name = tensor("op_6390_strides_0"), val = tensor([1, 1])]; + tensor var_6390_pad_0 = const()[name = tensor("op_6390_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6390_dilations_0 = const()[name = tensor("op_6390_dilations_0"), val = tensor([1, 1])]; + tensor var_6390_groups_0 = const()[name = tensor("op_6390_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96365312))), name = tensor("layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96358016))), shape = tensor([512, 512, 1, 1])]; + tensor var_6390_cast_fp16 = conv(dilations = var_6390_dilations_0, groups = var_6390_groups_0, pad = var_6390_pad_0, pad_type = var_6390_pad_type_0, strides = var_6390_strides_0, weight = layers_16_self_attn_q_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6390_cast_fp16")]; + tensor query_65_cast_fp16 = add(x = var_6384_cast_fp16, y = var_6390_cast_fp16)[name = tensor("query_65_cast_fp16")]; + tensor var_6399_pad_type_0 = const()[name = tensor("op_6399_pad_type_0"), val = tensor("valid")]; + tensor var_6399_strides_0 = const()[name = tensor("op_6399_strides_0"), val = tensor([1, 1])]; + tensor var_6399_pad_0 = const()[name = tensor("op_6399_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6399_dilations_0 = const()[name = tensor("op_6399_dilations_0"), val = tensor([1, 1])]; + tensor var_6399_groups_0 = const()[name = tensor("op_6399_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96398144))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96594816))), name = tensor("layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_6399_cast_fp16 = conv(dilations = var_6399_dilations_0, groups = var_6399_groups_0, pad = var_6399_pad_0, pad_type = var_6399_pad_type_0, strides = var_6399_strides_0, weight = layers_16_self_attn_k_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6399_cast_fp16")]; + tensor var_6405_pad_type_0 = const()[name = tensor("op_6405_pad_type_0"), val = tensor("valid")]; + tensor var_6405_strides_0 = const()[name = tensor("op_6405_strides_0"), val = tensor([1, 1])]; + tensor var_6405_pad_0 = const()[name = tensor("op_6405_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6405_dilations_0 = const()[name = tensor("op_6405_dilations_0"), val = tensor([1, 1])]; + tensor var_6405_groups_0 = const()[name = tensor("op_6405_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96602560))), name = tensor("layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96595008))), shape = tensor([512, 512, 1, 1])]; + tensor var_6405_cast_fp16 = conv(dilations = var_6405_dilations_0, groups = var_6405_groups_0, pad = var_6405_pad_0, pad_type = var_6405_pad_type_0, strides = var_6405_strides_0, weight = layers_16_self_attn_k_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6405_cast_fp16")]; + tensor key_cast_fp16 = add(x = var_6399_cast_fp16, y = var_6405_cast_fp16)[name = tensor("key_cast_fp16")]; + tensor var_6415_pad_type_0 = const()[name = tensor("op_6415_pad_type_0"), val = tensor("valid")]; + tensor var_6415_strides_0 = const()[name = tensor("op_6415_strides_0"), val = tensor([1, 1])]; + tensor var_6415_pad_0 = const()[name = tensor("op_6415_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6415_dilations_0 = const()[name = tensor("op_6415_dilations_0"), val = tensor([1, 1])]; + tensor var_6415_groups_0 = const()[name = tensor("op_6415_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96635392))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96832064))), name = tensor("layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_16_self_attn_v_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_v_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96832256)))]; + tensor var_6415_cast_fp16 = conv(bias = layers_16_self_attn_v_proj_inlier_module_bias_to_fp16, dilations = var_6415_dilations_0, groups = var_6415_groups_0, pad = var_6415_pad_0, pad_type = var_6415_pad_type_0, strides = var_6415_strides_0, weight = layers_16_self_attn_v_proj_inlier_module_weight_to_fp16_palettized, x = obj_67_cast_fp16)[name = tensor("op_6415_cast_fp16")]; + tensor var_6421_pad_type_0 = const()[name = tensor("op_6421_pad_type_0"), val = tensor("valid")]; + tensor var_6421_strides_0 = const()[name = tensor("op_6421_strides_0"), val = tensor([1, 1])]; + tensor var_6421_pad_0 = const()[name = tensor("op_6421_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6421_dilations_0 = const()[name = tensor("op_6421_dilations_0"), val = tensor([1, 1])]; + tensor var_6421_groups_0 = const()[name = tensor("op_6421_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96841664))), name = tensor("layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96833344))), shape = tensor([512, 512, 1, 1])]; + tensor var_6421_cast_fp16 = conv(dilations = var_6421_dilations_0, groups = var_6421_groups_0, pad = var_6421_pad_0, pad_type = var_6421_pad_type_0, strides = var_6421_strides_0, weight = layers_16_self_attn_v_proj_outlier_module_weight_to_fp16_sparsified, x = obj_67_cast_fp16)[name = tensor("op_6421_cast_fp16")]; + tensor value_cast_fp16 = add(x = var_6415_cast_fp16, y = var_6421_cast_fp16)[name = tensor("value_cast_fp16")]; + tensor var_6424_to_fp16 = const()[name = tensor("op_6424_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96874496)))]; + tensor query_cast_fp16 = add(x = query_65_cast_fp16, y = var_6424_to_fp16)[name = tensor("query_cast_fp16")]; + tensor var_6427_to_fp16 = const()[name = tensor("op_6427_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96875584)))]; + tensor q_with_bias_v_cast_fp16 = add(x = query_65_cast_fp16, y = var_6427_to_fp16)[name = tensor("q_with_bias_v_cast_fp16")]; + tensor var_6437_pad_type_0 = const()[name = tensor("op_6437_pad_type_0"), val = tensor("valid")]; + tensor var_6437_strides_0 = const()[name = tensor("op_6437_strides_0"), val = tensor([1, 1])]; + tensor var_6437_pad_0 = const()[name = tensor("op_6437_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6437_dilations_0 = const()[name = tensor("op_6437_dilations_0"), val = tensor([1, 1])]; + tensor var_6437_groups_0 = const()[name = tensor("op_6437_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(96876672))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97073344))), name = tensor("layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor var_6437_cast_fp16 = conv(bias = input_17_mean_0_to_fp16, dilations = var_6437_dilations_0, groups = var_6437_groups_0, pad = var_6437_pad_0, pad_type = var_6437_pad_type_0, strides = var_6437_strides_0, weight = layers_16_self_attn_linear_pos_inlier_module_weight_to_fp16_palettized, x = obj_3_cast_fp16)[name = tensor("op_6437_cast_fp16")]; + tensor var_6443_pad_type_0 = const()[name = tensor("op_6443_pad_type_0"), val = tensor("valid")]; + tensor var_6443_strides_0 = const()[name = tensor("op_6443_strides_0"), val = tensor([1, 1])]; + tensor var_6443_pad_0 = const()[name = tensor("op_6443_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6443_dilations_0 = const()[name = tensor("op_6443_dilations_0"), val = tensor([1, 1])]; + tensor var_6443_groups_0 = const()[name = tensor("op_6443_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97094016))), name = tensor("layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97073536))), shape = tensor([512, 512, 1, 1])]; + tensor var_6443_cast_fp16 = conv(dilations = var_6443_dilations_0, groups = var_6443_groups_0, pad = var_6443_pad_0, pad_type = var_6443_pad_type_0, strides = var_6443_strides_0, weight = layers_16_self_attn_linear_pos_outlier_module_weight_to_fp16_sparsified, x = obj_3_cast_fp16)[name = tensor("op_6443_cast_fp16")]; + tensor p_cast_fp16 = add(x = var_6437_cast_fp16, y = var_6443_cast_fp16)[name = tensor("p_cast_fp16")]; + tensor var_6447 = const()[name = tensor("op_6447"), val = tensor([1, 8, 64, 188])]; + tensor var_6448_cast_fp16 = reshape(shape = var_6447, x = q_with_bias_v_cast_fp16)[name = tensor("op_6448_cast_fp16")]; + tensor var_6449 = const()[name = tensor("op_6449"), val = tensor([1, 8, 64, -1])]; + tensor var_6450_cast_fp16 = reshape(shape = var_6449, x = p_cast_fp16)[name = tensor("op_6450_cast_fp16")]; + tensor matrix_bd_129_transpose_x_0 = const()[name = tensor("matrix_bd_129_transpose_x_0"), val = tensor(true)]; + tensor matrix_bd_129_transpose_y_0 = const()[name = tensor("matrix_bd_129_transpose_y_0"), val = tensor(false)]; + tensor matrix_bd_129_cast_fp16 = matmul(transpose_x = matrix_bd_129_transpose_x_0, transpose_y = matrix_bd_129_transpose_y_0, x = var_6448_cast_fp16, y = var_6450_cast_fp16)[name = tensor("matrix_bd_129_cast_fp16")]; + tensor matrix_bd_131_pad_0 = const()[name = tensor("matrix_bd_131_pad_0"), val = tensor([0, 0, 0, 0, 0, 0, 1, 0])]; + tensor matrix_bd_131_mode_0 = const()[name = tensor("matrix_bd_131_mode_0"), val = tensor("constant")]; + tensor const_186_to_fp16 = const()[name = tensor("const_186_to_fp16"), val = tensor(0x0p+0)]; + tensor matrix_bd_131_cast_fp16 = pad(constant_val = const_186_to_fp16, mode = matrix_bd_131_mode_0, pad = matrix_bd_131_pad_0, x = matrix_bd_129_cast_fp16)[name = tensor("matrix_bd_131_cast_fp16")]; + tensor var_6459 = const()[name = tensor("op_6459"), val = tensor([1, 8, -1, 188])]; + tensor matrix_bd_133_cast_fp16 = reshape(shape = var_6459, x = matrix_bd_131_cast_fp16)[name = tensor("matrix_bd_133_cast_fp16")]; + tensor var_6463_begin_0 = const()[name = tensor("op_6463_begin_0"), val = tensor([0, 0, 1, 0])]; + tensor var_6463_end_0 = const()[name = tensor("op_6463_end_0"), val = tensor([1, 8, 376, 188])]; + tensor var_6463_end_mask_0 = const()[name = tensor("op_6463_end_mask_0"), val = tensor([true, true, true, true])]; + tensor var_6463_cast_fp16 = slice_by_index(begin = var_6463_begin_0, end = var_6463_end_0, end_mask = var_6463_end_mask_0, x = matrix_bd_133_cast_fp16)[name = tensor("op_6463_cast_fp16")]; + tensor var_6464 = const()[name = tensor("op_6464"), val = tensor([1, 8, 188, 375])]; + tensor matrix_bd_cast_fp16 = reshape(shape = var_6464, x = var_6463_cast_fp16)[name = tensor("matrix_bd_cast_fp16")]; + tensor var_6469_begin_0 = const()[name = tensor("op_6469_begin_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6469_end_0 = const()[name = tensor("op_6469_end_0"), val = tensor([1, 8, 188, 188])]; + tensor var_6469_end_mask_0 = const()[name = tensor("op_6469_end_mask_0"), val = tensor([true, true, true, false])]; + tensor var_6469_cast_fp16 = slice_by_index(begin = var_6469_begin_0, end = var_6469_end_0, end_mask = var_6469_end_mask_0, x = matrix_bd_cast_fp16)[name = tensor("op_6469_cast_fp16")]; + tensor var_6470_to_fp16 = const()[name = tensor("op_6470_to_fp16"), val = tensor(0x1p-3)]; + tensor qk_mask_cast_fp16 = mul(x = var_6469_cast_fp16, y = var_6470_to_fp16)[name = tensor("qk_mask_cast_fp16")]; + tensor var_6474 = const()[name = tensor("op_6474"), val = tensor([1, 8, 64, 188])]; + tensor mh_q_cast_fp16 = reshape(shape = var_6474, x = query_cast_fp16)[name = tensor("mh_q_cast_fp16")]; + tensor var_6476_to_fp16 = const()[name = tensor("op_6476_to_fp16"), val = tensor(0x1p-3)]; + tensor var_6477_cast_fp16 = mul(x = mh_q_cast_fp16, y = var_6476_to_fp16)[name = tensor("op_6477_cast_fp16")]; + tensor var_6480 = const()[name = tensor("op_6480"), val = tensor([1, 8, 64, 188])]; + tensor var_6481_cast_fp16 = reshape(shape = var_6480, x = key_cast_fp16)[name = tensor("op_6481_cast_fp16")]; + tensor mh_w_65_transpose_x_0 = const()[name = tensor("mh_w_65_transpose_x_0"), val = tensor(true)]; + tensor mh_w_65_transpose_y_0 = const()[name = tensor("mh_w_65_transpose_y_0"), val = tensor(false)]; + tensor mh_w_65_cast_fp16 = matmul(transpose_x = mh_w_65_transpose_x_0, transpose_y = mh_w_65_transpose_y_0, x = var_6477_cast_fp16, y = var_6481_cast_fp16)[name = tensor("mh_w_65_cast_fp16")]; + tensor mh_w_cast_fp16 = add(x = mh_w_65_cast_fp16, y = qk_mask_cast_fp16)[name = tensor("mh_w_cast_fp16")]; + tensor var_6485_cast_fp16 = softmax(axis = var_6272, x = mh_w_cast_fp16)[name = tensor("op_6485_cast_fp16")]; + tensor var_6486 = const()[name = tensor("op_6486"), val = tensor([1, 8, 64, 188])]; + tensor var_6487_cast_fp16 = reshape(shape = var_6486, x = value_cast_fp16)[name = tensor("op_6487_cast_fp16")]; + tensor attn_transpose_x_0 = const()[name = tensor("attn_transpose_x_0"), val = tensor(false)]; + tensor attn_transpose_y_0 = const()[name = tensor("attn_transpose_y_0"), val = tensor(true)]; + tensor attn_cast_fp16 = matmul(transpose_x = attn_transpose_x_0, transpose_y = attn_transpose_y_0, x = var_6487_cast_fp16, y = var_6485_cast_fp16)[name = tensor("attn_cast_fp16")]; + tensor var_6490 = const()[name = tensor("op_6490"), val = tensor([1, 512, 1, 188])]; + tensor input_439_cast_fp16 = reshape(shape = var_6490, x = attn_cast_fp16)[name = tensor("input_439_cast_fp16")]; + tensor var_6500_pad_type_0 = const()[name = tensor("op_6500_pad_type_0"), val = tensor("valid")]; + tensor var_6500_strides_0 = const()[name = tensor("op_6500_strides_0"), val = tensor([1, 1])]; + tensor var_6500_pad_0 = const()[name = tensor("op_6500_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6500_dilations_0 = const()[name = tensor("op_6500_dilations_0"), val = tensor([1, 1])]; + tensor var_6500_groups_0 = const()[name = tensor("op_6500_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97126848))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97323520))), name = tensor("layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_16_self_attn_o_proj_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_self_attn_o_proj_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97323712)))]; + tensor var_6500_cast_fp16 = conv(bias = layers_16_self_attn_o_proj_inlier_module_bias_to_fp16, dilations = var_6500_dilations_0, groups = var_6500_groups_0, pad = var_6500_pad_0, pad_type = var_6500_pad_type_0, strides = var_6500_strides_0, weight = layers_16_self_attn_o_proj_inlier_module_weight_to_fp16_palettized, x = input_439_cast_fp16)[name = tensor("op_6500_cast_fp16")]; + tensor var_6506_pad_type_0 = const()[name = tensor("op_6506_pad_type_0"), val = tensor("valid")]; + tensor var_6506_strides_0 = const()[name = tensor("op_6506_strides_0"), val = tensor([1, 1])]; + tensor var_6506_pad_0 = const()[name = tensor("op_6506_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6506_dilations_0 = const()[name = tensor("op_6506_dilations_0"), val = tensor([1, 1])]; + tensor var_6506_groups_0 = const()[name = tensor("op_6506_groups_0"), val = tensor(1)]; + tensor layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97333248))), name = tensor("layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97324800))), shape = tensor([512, 512, 1, 1])]; + tensor var_6506_cast_fp16 = conv(dilations = var_6506_dilations_0, groups = var_6506_groups_0, pad = var_6506_pad_0, pad_type = var_6506_pad_type_0, strides = var_6506_strides_0, weight = layers_16_self_attn_o_proj_outlier_module_weight_to_fp16_sparsified, x = input_439_cast_fp16)[name = tensor("op_6506_cast_fp16")]; + tensor obj_cast_fp16 = add(x = var_6500_cast_fp16, y = var_6506_cast_fp16)[name = tensor("obj_cast_fp16")]; + tensor inputs_165_cast_fp16 = add(x = inputs_163_cast_fp16, y = obj_cast_fp16)[name = tensor("inputs_165_cast_fp16")]; + tensor out_165_axes_0 = const()[name = tensor("out_165_axes_0"), val = tensor([1])]; + tensor var_6517_to_fp16 = const()[name = tensor("op_6517_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_165_cast_fp16 = layer_norm(axes = out_165_axes_0, epsilon = var_6517_to_fp16, x = inputs_165_cast_fp16)[name = tensor("out_165_cast_fp16")]; + tensor input_441_gamma_0_to_fp16 = const()[name = tensor("input_441_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97366080)))]; + tensor input_441_beta_0_to_fp16 = const()[name = tensor("input_441_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97367168)))]; + tensor input_441_epsilon_0_to_fp16 = const()[name = tensor("input_441_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_441_cast_fp16 = batch_norm(beta = input_441_beta_0_to_fp16, epsilon = input_441_epsilon_0_to_fp16, gamma = input_441_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_165_cast_fp16)[name = tensor("input_441_cast_fp16")]; + tensor var_6539_pad_type_0 = const()[name = tensor("op_6539_pad_type_0"), val = tensor("valid")]; + tensor var_6539_strides_0 = const()[name = tensor("op_6539_strides_0"), val = tensor([1, 1])]; + tensor var_6539_pad_0 = const()[name = tensor("op_6539_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6539_dilations_0 = const()[name = tensor("op_6539_dilations_0"), val = tensor([1, 1])]; + tensor var_6539_groups_0 = const()[name = tensor("op_6539_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97368256))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97761536))), name = tensor("layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized"), shape = tensor([1024, 512, 1, 1])]; + tensor layers_16_conv_pointwise_conv1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_conv_pointwise_conv1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97761728)))]; + tensor var_6539_cast_fp16 = conv(bias = layers_16_conv_pointwise_conv1_inlier_module_bias_to_fp16, dilations = var_6539_dilations_0, groups = var_6539_groups_0, pad = var_6539_pad_0, pad_type = var_6539_pad_type_0, strides = var_6539_strides_0, weight = layers_16_conv_pointwise_conv1_inlier_module_weight_to_fp16_palettized, x = input_441_cast_fp16)[name = tensor("op_6539_cast_fp16")]; + tensor var_6545_pad_type_0 = const()[name = tensor("op_6545_pad_type_0"), val = tensor("valid")]; + tensor var_6545_strides_0 = const()[name = tensor("op_6545_strides_0"), val = tensor([1, 1])]; + tensor var_6545_pad_0 = const()[name = tensor("op_6545_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6545_dilations_0 = const()[name = tensor("op_6545_dilations_0"), val = tensor([1, 1])]; + tensor var_6545_groups_0 = const()[name = tensor("op_6545_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97782720))), name = tensor("layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97763840))), shape = tensor([1024, 512, 1, 1])]; + tensor var_6545_cast_fp16 = conv(dilations = var_6545_dilations_0, groups = var_6545_groups_0, pad = var_6545_pad_0, pad_type = var_6545_pad_type_0, strides = var_6545_strides_0, weight = layers_16_conv_pointwise_conv1_outlier_module_weight_to_fp16_sparsified, x = input_441_cast_fp16)[name = tensor("op_6545_cast_fp16")]; + tensor input_443_cast_fp16 = add(x = var_6539_cast_fp16, y = var_6545_cast_fp16)[name = tensor("input_443_cast_fp16")]; + tensor input_445_split_num_splits_0 = const()[name = tensor("input_445_split_num_splits_0"), val = tensor(2)]; + tensor input_445_split_axis_0 = const()[name = tensor("input_445_split_axis_0"), val = tensor(1)]; + tensor input_445_split_cast_fp16_0, tensor input_445_split_cast_fp16_1 = split(axis = input_445_split_axis_0, num_splits = input_445_split_num_splits_0, x = input_443_cast_fp16)[name = tensor("input_445_split_cast_fp16")]; + tensor input_445_split_1_sigmoid_cast_fp16 = sigmoid(x = input_445_split_cast_fp16_1)[name = tensor("input_445_split_1_sigmoid_cast_fp16")]; + tensor input_445_cast_fp16 = mul(x = input_445_split_cast_fp16_0, y = input_445_split_1_sigmoid_cast_fp16)[name = tensor("input_445_cast_fp16")]; + tensor input_447_pad_type_0 = const()[name = tensor("input_447_pad_type_0"), val = tensor("custom")]; + tensor input_447_pad_0 = const()[name = tensor("input_447_pad_0"), val = tensor([0, 0, 4, 4])]; + tensor input_447_groups_0 = const()[name = tensor("input_447_groups_0"), val = tensor(512)]; + tensor input_447_strides_0 = const()[name = tensor("input_447_strides_0"), val = tensor([1, 1])]; + tensor input_447_dilations_0 = const()[name = tensor("input_447_dilations_0"), val = tensor([1, 1])]; + tensor const_223_to_fp16 = const()[name = tensor("const_223_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97848320)))]; + tensor const_224_to_fp16 = const()[name = tensor("const_224_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97857600)))]; + tensor input_449_cast_fp16 = conv(bias = const_224_to_fp16, dilations = input_447_dilations_0, groups = input_447_groups_0, pad = input_447_pad_0, pad_type = input_447_pad_type_0, strides = input_447_strides_0, weight = const_223_to_fp16, x = input_445_cast_fp16)[name = tensor("input_449_cast_fp16")]; + tensor input_451_cast_fp16 = silu(x = input_449_cast_fp16)[name = tensor("input_451_cast_fp16")]; + tensor var_6569_pad_type_0 = const()[name = tensor("op_6569_pad_type_0"), val = tensor("valid")]; + tensor var_6569_strides_0 = const()[name = tensor("op_6569_strides_0"), val = tensor([1, 1])]; + tensor var_6569_pad_0 = const()[name = tensor("op_6569_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6569_dilations_0 = const()[name = tensor("op_6569_dilations_0"), val = tensor([1, 1])]; + tensor var_6569_groups_0 = const()[name = tensor("op_6569_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(97858688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98055360))), name = tensor("layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 512, 1, 1])]; + tensor layers_16_conv_pointwise_conv2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_conv_pointwise_conv2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98055552)))]; + tensor var_6569_cast_fp16 = conv(bias = layers_16_conv_pointwise_conv2_inlier_module_bias_to_fp16, dilations = var_6569_dilations_0, groups = var_6569_groups_0, pad = var_6569_pad_0, pad_type = var_6569_pad_type_0, strides = var_6569_strides_0, weight = layers_16_conv_pointwise_conv2_inlier_module_weight_to_fp16_palettized, x = input_451_cast_fp16)[name = tensor("op_6569_cast_fp16")]; + tensor var_6575_pad_type_0 = const()[name = tensor("op_6575_pad_type_0"), val = tensor("valid")]; + tensor var_6575_strides_0 = const()[name = tensor("op_6575_strides_0"), val = tensor([1, 1])]; + tensor var_6575_pad_0 = const()[name = tensor("op_6575_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6575_dilations_0 = const()[name = tensor("op_6575_dilations_0"), val = tensor([1, 1])]; + tensor var_6575_groups_0 = const()[name = tensor("op_6575_groups_0"), val = tensor(1)]; + tensor layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98064256))), name = tensor("layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98056640))), shape = tensor([512, 512, 1, 1])]; + tensor var_6575_cast_fp16 = conv(dilations = var_6575_dilations_0, groups = var_6575_groups_0, pad = var_6575_pad_0, pad_type = var_6575_pad_type_0, strides = var_6575_strides_0, weight = layers_16_conv_pointwise_conv2_outlier_module_weight_to_fp16_sparsified, x = input_451_cast_fp16)[name = tensor("op_6575_cast_fp16")]; + tensor x_101_cast_fp16 = add(x = var_6569_cast_fp16, y = var_6575_cast_fp16)[name = tensor("x_101_cast_fp16")]; + tensor inputs_167_cast_fp16 = add(x = inputs_165_cast_fp16, y = x_101_cast_fp16)[name = tensor("inputs_167_cast_fp16")]; + tensor out_167_axes_0 = const()[name = tensor("out_167_axes_0"), val = tensor([1])]; + tensor var_6586_to_fp16 = const()[name = tensor("op_6586_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_167_cast_fp16 = layer_norm(axes = out_167_axes_0, epsilon = var_6586_to_fp16, x = inputs_167_cast_fp16)[name = tensor("out_167_cast_fp16")]; + tensor input_453_gamma_0_to_fp16 = const()[name = tensor("input_453_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98097088)))]; + tensor input_453_beta_0_to_fp16 = const()[name = tensor("input_453_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98098176)))]; + tensor input_453_epsilon_0_to_fp16 = const()[name = tensor("input_453_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor input_453_cast_fp16 = batch_norm(beta = input_453_beta_0_to_fp16, epsilon = input_453_epsilon_0_to_fp16, gamma = input_453_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_167_cast_fp16)[name = tensor("input_453_cast_fp16")]; + tensor var_6606_pad_type_0 = const()[name = tensor("op_6606_pad_type_0"), val = tensor("valid")]; + tensor var_6606_strides_0 = const()[name = tensor("op_6606_strides_0"), val = tensor([1, 1])]; + tensor var_6606_pad_0 = const()[name = tensor("op_6606_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6606_dilations_0 = const()[name = tensor("op_6606_dilations_0"), val = tensor([1, 1])]; + tensor var_6606_groups_0 = const()[name = tensor("op_6606_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98099264))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98885760))), name = tensor("layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized"), shape = tensor([2048, 512, 1, 1])]; + tensor layers_16_feed_forward2_fc1_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_feed_forward2_fc1_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98885952)))]; + tensor var_6606_cast_fp16 = conv(bias = layers_16_feed_forward2_fc1_inlier_module_bias_to_fp16, dilations = var_6606_dilations_0, groups = var_6606_groups_0, pad = var_6606_pad_0, pad_type = var_6606_pad_type_0, strides = var_6606_strides_0, weight = layers_16_feed_forward2_fc1_inlier_module_weight_to_fp16_palettized, x = input_453_cast_fp16)[name = tensor("op_6606_cast_fp16")]; + tensor var_6612_pad_type_0 = const()[name = tensor("op_6612_pad_type_0"), val = tensor("valid")]; + tensor var_6612_strides_0 = const()[name = tensor("op_6612_strides_0"), val = tensor([1, 1])]; + tensor var_6612_pad_0 = const()[name = tensor("op_6612_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6612_dilations_0 = const()[name = tensor("op_6612_dilations_0"), val = tensor([1, 1])]; + tensor var_6612_groups_0 = const()[name = tensor("op_6612_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98919552))), name = tensor("layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(98890112))), shape = tensor([2048, 512, 1, 1])]; + tensor var_6612_cast_fp16 = conv(dilations = var_6612_dilations_0, groups = var_6612_groups_0, pad = var_6612_pad_0, pad_type = var_6612_pad_type_0, strides = var_6612_strides_0, weight = layers_16_feed_forward2_fc1_outlier_module_weight_to_fp16_sparsified, x = input_453_cast_fp16)[name = tensor("op_6612_cast_fp16")]; + tensor input_455_cast_fp16 = add(x = var_6606_cast_fp16, y = var_6612_cast_fp16)[name = tensor("input_455_cast_fp16")]; + tensor input_457_cast_fp16 = silu(x = input_455_cast_fp16)[name = tensor("input_457_cast_fp16")]; + tensor var_6623_pad_type_0 = const()[name = tensor("op_6623_pad_type_0"), val = tensor("valid")]; + tensor var_6623_strides_0 = const()[name = tensor("op_6623_strides_0"), val = tensor([1, 1])]; + tensor var_6623_pad_0 = const()[name = tensor("op_6623_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6623_dilations_0 = const()[name = tensor("op_6623_dilations_0"), val = tensor([1, 1])]; + tensor var_6623_groups_0 = const()[name = tensor("op_6623_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99050688))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99837184))), name = tensor("layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized"), shape = tensor([512, 2048, 1, 1])]; + tensor layers_16_feed_forward2_fc2_inlier_module_bias_to_fp16 = const()[name = tensor("layers_16_feed_forward2_fc2_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99837376)))]; + tensor var_6623_cast_fp16 = conv(bias = layers_16_feed_forward2_fc2_inlier_module_bias_to_fp16, dilations = var_6623_dilations_0, groups = var_6623_groups_0, pad = var_6623_pad_0, pad_type = var_6623_pad_type_0, strides = var_6623_strides_0, weight = layers_16_feed_forward2_fc2_inlier_module_weight_to_fp16_palettized, x = input_457_cast_fp16)[name = tensor("op_6623_cast_fp16")]; + tensor var_6629_pad_type_0 = const()[name = tensor("op_6629_pad_type_0"), val = tensor("valid")]; + tensor var_6629_strides_0 = const()[name = tensor("op_6629_strides_0"), val = tensor([1, 1])]; + tensor var_6629_pad_0 = const()[name = tensor("op_6629_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6629_dilations_0 = const()[name = tensor("op_6629_dilations_0"), val = tensor([1, 1])]; + tensor var_6629_groups_0 = const()[name = tensor("op_6629_groups_0"), val = tensor(1)]; + tensor layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99870784))), name = tensor("layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(99838464))), shape = tensor([512, 2048, 1, 1])]; + tensor var_6629_cast_fp16 = conv(dilations = var_6629_dilations_0, groups = var_6629_groups_0, pad = var_6629_pad_0, pad_type = var_6629_pad_type_0, strides = var_6629_strides_0, weight = layers_16_feed_forward2_fc2_outlier_module_weight_to_fp16_sparsified, x = input_457_cast_fp16)[name = tensor("op_6629_cast_fp16")]; + tensor x_cast_fp16 = add(x = var_6623_cast_fp16, y = var_6629_cast_fp16)[name = tensor("x_cast_fp16")]; + tensor var_6631_to_fp16 = const()[name = tensor("op_6631_to_fp16"), val = tensor(0x1p-1)]; + tensor var_6632_cast_fp16 = mul(x = x_cast_fp16, y = var_6631_to_fp16)[name = tensor("op_6632_cast_fp16")]; + tensor inputs_cast_fp16 = add(x = inputs_167_cast_fp16, y = var_6632_cast_fp16)[name = tensor("inputs_cast_fp16")]; + tensor out_169_axes_0 = const()[name = tensor("out_169_axes_0"), val = tensor([1])]; + tensor var_6642_to_fp16 = const()[name = tensor("op_6642_to_fp16"), val = tensor(0x1.5p-17)]; + tensor out_169_cast_fp16 = layer_norm(axes = out_169_axes_0, epsilon = var_6642_to_fp16, x = inputs_cast_fp16)[name = tensor("out_169_cast_fp16")]; + tensor encoder_output_embeds_type_fp32_gamma_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_gamma_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100001920)))]; + tensor encoder_output_embeds_type_fp32_beta_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_beta_0_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100003008)))]; + tensor encoder_output_embeds_type_fp32_epsilon_0_to_fp16 = const()[name = tensor("encoder_output_embeds_type_fp32_epsilon_0_to_fp16"), val = tensor(0x1.5p-17)]; + tensor encoder_output_embeds = batch_norm(beta = encoder_output_embeds_type_fp32_beta_0_to_fp16, epsilon = encoder_output_embeds_type_fp32_epsilon_0_to_fp16, gamma = encoder_output_embeds_type_fp32_gamma_0_to_fp16, mean = input_17_mean_0_to_fp16, variance = input_17_variance_0_to_fp16, x = out_169_cast_fp16)[name = tensor("encoder_output_embeds_type_fp32_cast_fp16")]; + tensor var_6665_pad_type_0 = const()[name = tensor("op_6665_pad_type_0"), val = tensor("valid")]; + tensor var_6665_strides_0 = const()[name = tensor("op_6665_strides_0"), val = tensor([1, 1])]; + tensor var_6665_pad_0 = const()[name = tensor("op_6665_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6665_dilations_0 = const()[name = tensor("op_6665_dilations_0"), val = tensor([1, 1])]; + tensor var_6665_groups_0 = const()[name = tensor("op_6665_groups_0"), val = tensor(1)]; + tensor out_projection_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100004096))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100249920))), name = tensor("out_projection_inlier_module_weight_to_fp16_palettized"), shape = tensor([640, 512, 1, 1])]; + tensor out_projection_inlier_module_bias_to_fp16 = const()[name = tensor("out_projection_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100250112)))]; + tensor var_6665_cast_fp16 = conv(bias = out_projection_inlier_module_bias_to_fp16, dilations = var_6665_dilations_0, groups = var_6665_groups_0, pad = var_6665_pad_0, pad_type = var_6665_pad_type_0, strides = var_6665_strides_0, weight = out_projection_inlier_module_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("op_6665_cast_fp16")]; + tensor var_6671_pad_type_0 = const()[name = tensor("op_6671_pad_type_0"), val = tensor("valid")]; + tensor var_6671_strides_0 = const()[name = tensor("op_6671_strides_0"), val = tensor([1, 1])]; + tensor var_6671_pad_0 = const()[name = tensor("op_6671_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6671_dilations_0 = const()[name = tensor("op_6671_dilations_0"), val = tensor([1, 1])]; + tensor var_6671_groups_0 = const()[name = tensor("op_6671_groups_0"), val = tensor(1)]; + tensor out_projection_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100261568))), name = tensor("out_projection_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100251456))), shape = tensor([640, 512, 1, 1])]; + tensor var_6671_cast_fp16 = conv(dilations = var_6671_dilations_0, groups = var_6671_groups_0, pad = var_6671_pad_0, pad_type = var_6671_pad_type_0, strides = var_6671_strides_0, weight = out_projection_outlier_module_weight_to_fp16_sparsified, x = encoder_output_embeds)[name = tensor("op_6671_cast_fp16")]; + tensor joint_projected_encoder_output_embeds = add(x = var_6665_cast_fp16, y = var_6671_cast_fp16)[name = tensor("op_6672_cast_fp16")]; + tensor var_6686_pad_type_0 = const()[name = tensor("op_6686_pad_type_0"), val = tensor("valid")]; + tensor var_6686_strides_0 = const()[name = tensor("op_6686_strides_0"), val = tensor([1, 1])]; + tensor var_6686_pad_0 = const()[name = tensor("op_6686_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6686_dilations_0 = const()[name = tensor("op_6686_dilations_0"), val = tensor([1, 1])]; + tensor var_6686_groups_0 = const()[name = tensor("op_6686_groups_0"), val = tensor(1)]; + tensor ctc_head_inlier_module_weight_to_fp16_palettized = constexpr_lut_to_dense()[indices = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100302592))), lut = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100696256))), name = tensor("ctc_head_inlier_module_weight_to_fp16_palettized"), shape = tensor([1025, 512, 1, 1])]; + tensor ctc_head_inlier_module_bias_to_fp16 = const()[name = tensor("ctc_head_inlier_module_bias_to_fp16"), val = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100696448)))]; + tensor var_6686_cast_fp16 = conv(bias = ctc_head_inlier_module_bias_to_fp16, dilations = var_6686_dilations_0, groups = var_6686_groups_0, pad = var_6686_pad_0, pad_type = var_6686_pad_type_0, strides = var_6686_strides_0, weight = ctc_head_inlier_module_weight_to_fp16_palettized, x = encoder_output_embeds)[name = tensor("op_6686_cast_fp16")]; + tensor var_6692_pad_type_0 = const()[name = tensor("op_6692_pad_type_0"), val = tensor("valid")]; + tensor var_6692_strides_0 = const()[name = tensor("op_6692_strides_0"), val = tensor([1, 1])]; + tensor var_6692_pad_0 = const()[name = tensor("op_6692_pad_0"), val = tensor([0, 0, 0, 0])]; + tensor var_6692_dilations_0 = const()[name = tensor("op_6692_dilations_0"), val = tensor([1, 1])]; + tensor var_6692_groups_0 = const()[name = tensor("op_6692_groups_0"), val = tensor(1)]; + tensor ctc_head_outlier_module_weight_to_fp16_sparsified = constexpr_sparse_to_dense()[mask = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100712640))), name = tensor("ctc_head_outlier_module_weight_to_fp16_sparsified"), nonzero_data = tensor(BLOBFILE(path = tensor("@model_path/weights/weight.bin"), offset = tensor(100698624))), shape = tensor([1025, 512, 1, 1])]; + tensor var_6692_cast_fp16 = conv(dilations = var_6692_dilations_0, groups = var_6692_groups_0, pad = var_6692_pad_0, pad_type = var_6692_pad_type_0, strides = var_6692_strides_0, weight = ctc_head_outlier_module_weight_to_fp16_sparsified, x = encoder_output_embeds)[name = tensor("op_6692_cast_fp16")]; + tensor ctc_head_raw_output = add(x = var_6686_cast_fp16, y = var_6692_cast_fp16)[name = tensor("op_6693_cast_fp16")]; + tensor var_6715 = const()[name = tensor("op_6715"), val = tensor(1)]; + tensor var_6717_softmax_cast_fp16 = softmax(axis = var_6715, x = ctc_head_raw_output)[name = tensor("op_6717_softmax_cast_fp16")]; + tensor var_6717_epsilon_0 = const()[name = tensor("op_6717_epsilon_0"), val = tensor(0x1p-149)]; + tensor ctc_head_output = log(epsilon = var_6717_epsilon_0, x = var_6717_softmax_cast_fp16)[name = tensor("op_6717_cast_fp16")]; + } -> (encoder_output_embeds, joint_projected_encoder_output_embeds, ctc_head_raw_output, ctc_head_output); +} \ No newline at end of file