Upload config.json with huggingface_hub
Browse files- config.json +139 -0
config.json
ADDED
@@ -0,0 +1,139 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"_name_or_path": "./checkpoints/mlcd-embodied-pretrain",
|
3 |
+
"add_faster_video": false,
|
4 |
+
"add_time_instruction": false,
|
5 |
+
"architectures": [
|
6 |
+
"LlavaQwenForCausalLM"
|
7 |
+
],
|
8 |
+
"attention_dropout": 0.0,
|
9 |
+
"bos_token_id": 151643,
|
10 |
+
"eos_token_id": 151645,
|
11 |
+
"faster_token_stride": 10,
|
12 |
+
"force_sample": false,
|
13 |
+
"hidden_act": "silu",
|
14 |
+
"hidden_size": 3584,
|
15 |
+
"image_aspect_ratio": "anyres",
|
16 |
+
"image_crop_resolution": null,
|
17 |
+
"image_grid_pinpoints": [
|
18 |
+
[
|
19 |
+
336,
|
20 |
+
336
|
21 |
+
],
|
22 |
+
[
|
23 |
+
336,
|
24 |
+
672
|
25 |
+
],
|
26 |
+
[
|
27 |
+
336,
|
28 |
+
1008
|
29 |
+
],
|
30 |
+
[
|
31 |
+
336,
|
32 |
+
1344
|
33 |
+
],
|
34 |
+
[
|
35 |
+
336,
|
36 |
+
1680
|
37 |
+
],
|
38 |
+
[
|
39 |
+
336,
|
40 |
+
2016
|
41 |
+
],
|
42 |
+
[
|
43 |
+
672,
|
44 |
+
336
|
45 |
+
],
|
46 |
+
[
|
47 |
+
672,
|
48 |
+
672
|
49 |
+
],
|
50 |
+
[
|
51 |
+
672,
|
52 |
+
1008
|
53 |
+
],
|
54 |
+
[
|
55 |
+
672,
|
56 |
+
1344
|
57 |
+
],
|
58 |
+
[
|
59 |
+
672,
|
60 |
+
1680
|
61 |
+
],
|
62 |
+
[
|
63 |
+
1008,
|
64 |
+
336
|
65 |
+
],
|
66 |
+
[
|
67 |
+
1008,
|
68 |
+
672
|
69 |
+
],
|
70 |
+
[
|
71 |
+
1008,
|
72 |
+
1008
|
73 |
+
],
|
74 |
+
[
|
75 |
+
1008,
|
76 |
+
1344
|
77 |
+
],
|
78 |
+
[
|
79 |
+
1008,
|
80 |
+
1680
|
81 |
+
],
|
82 |
+
[
|
83 |
+
1344,
|
84 |
+
336
|
85 |
+
],
|
86 |
+
[
|
87 |
+
1344,
|
88 |
+
672
|
89 |
+
],
|
90 |
+
[
|
91 |
+
1344,
|
92 |
+
1008
|
93 |
+
],
|
94 |
+
[
|
95 |
+
1344,
|
96 |
+
1344
|
97 |
+
]
|
98 |
+
],
|
99 |
+
"image_split_resolution": null,
|
100 |
+
"initializer_range": 0.02,
|
101 |
+
"intermediate_size": 18944,
|
102 |
+
"max_position_embeddings": 32768,
|
103 |
+
"max_window_layers": 28,
|
104 |
+
"mm_hidden_size": 1024,
|
105 |
+
"mm_newline_position": "grid",
|
106 |
+
"mm_patch_merge_type": "spatial_unpad",
|
107 |
+
"mm_projector_lr": null,
|
108 |
+
"mm_projector_type": "mlp2x_gelu",
|
109 |
+
"mm_resampler_type": null,
|
110 |
+
"mm_spatial_pool_mode": "bilinear",
|
111 |
+
"mm_spatial_pool_stride": null,
|
112 |
+
"mm_tunable_parts": "mm_vision_tower,mm_mlp_adapter,mm_language_model",
|
113 |
+
"mm_use_im_patch_token": false,
|
114 |
+
"mm_use_im_start_end": false,
|
115 |
+
"mm_vision_select_feature": "patch",
|
116 |
+
"mm_vision_select_layer": -2,
|
117 |
+
"mm_vision_tower": "DeepGlint-AI/mlcd-vit-large-patch14-336",
|
118 |
+
"mm_vision_tower_lr": 2e-06,
|
119 |
+
"model_type": "qwen2",
|
120 |
+
"num_attention_heads": 28,
|
121 |
+
"num_hidden_layers": 28,
|
122 |
+
"num_key_value_heads": 4,
|
123 |
+
"pos_skipping_range": 4096,
|
124 |
+
"rms_norm_eps": 1e-06,
|
125 |
+
"rope_scaling": null,
|
126 |
+
"rope_theta": 1000000.0,
|
127 |
+
"sliding_window": null,
|
128 |
+
"tie_word_embeddings": false,
|
129 |
+
"tokenizer_model_max_length": 13768,
|
130 |
+
"tokenizer_padding_side": "right",
|
131 |
+
"torch_dtype": "bfloat16",
|
132 |
+
"transformers_version": "4.44.2",
|
133 |
+
"use_cache": true,
|
134 |
+
"use_mm_proj": true,
|
135 |
+
"use_pos_skipping": false,
|
136 |
+
"use_sliding_window": false,
|
137 |
+
"vision_tower_pretrained": null,
|
138 |
+
"vocab_size": 152064
|
139 |
+
}
|