Video-Text-to-Text
Transformers
English
qwen2
text-generation
DragonAura commited on
Commit
4e9ceb3
·
verified ·
1 Parent(s): 0741aff

Upload folder using huggingface_hub

Browse files
Files changed (3) hide show
  1. all_parameters.bin +3 -0
  2. checkpoint-12000.bin +3 -0
  3. config.json +167 -0
all_parameters.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:db59feaa6089c7cbc48ea10068af8547e12f01e921e010d9480e12046946f70c
3
+ size 18552045231
checkpoint-12000.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8f75c35528a16962309ab655d6697b5493d78fc76502e0576d5a6a1749a541eb
3
+ size 17756383152
config.json ADDED
@@ -0,0 +1,167 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "output/OV-Qwen2-7B-AM9",
3
+ "add_time_token": true,
4
+ "architectures": [
5
+ "LlavaQwenForCausalLM"
6
+ ],
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 151643,
9
+ "eos_token_id": 151645,
10
+ "freeze_mm_mlp_adapter": false,
11
+ "freeze_mm_vision_resampler": false,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 3584,
14
+ "image_aspect_ratio": "anyres",
15
+ "image_crop_resolution": 224,
16
+ "image_grid_pinpoints": [
17
+ [
18
+ 384,
19
+ 768
20
+ ],
21
+ [
22
+ 768,
23
+ 384
24
+ ],
25
+ [
26
+ 768,
27
+ 768
28
+ ],
29
+ [
30
+ 1152,
31
+ 384
32
+ ],
33
+ [
34
+ 384,
35
+ 1152
36
+ ]
37
+ ],
38
+ "image_split_resolution": 224,
39
+ "initializer_range": 0.02,
40
+ "intermediate_size": 18944,
41
+ "max_position_embeddings": 32768,
42
+ "max_window_layers": 28,
43
+ "mf_split_init": true,
44
+ "mm_hidden_size": 1152,
45
+ "mm_newline_position": "grid",
46
+ "mm_patch_merge_type": "spatial_unpad",
47
+ "mm_pooling_position": "after",
48
+ "mm_projector_lr": 2e-05,
49
+ "mm_projector_type": "mlp2x_gelu",
50
+ "mm_resampler_type": null,
51
+ "mm_spatial_pool_mode": "max",
52
+ "mm_spatial_pool_stride": 2,
53
+ "mm_tunable_parts": "mm_vision_tower,mm_mlp_adapter,mm_language_model",
54
+ "mm_use_im_patch_token": false,
55
+ "mm_use_im_start_end": false,
56
+ "mm_vision_select_feature": "patch",
57
+ "mm_vision_select_layer": -2,
58
+ "mm_vision_tower": "google/siglip-so400m-patch14-384",
59
+ "mm_vision_tower_lr": null,
60
+ "modality_max_length": "None",
61
+ "model_args": {
62
+ "add_time_token": true,
63
+ "audio_visual": false,
64
+ "beats_path": null,
65
+ "flash_attn": false,
66
+ "fps": 16,
67
+ "freeze_backbone": false,
68
+ "freeze_beats": true,
69
+ "freeze_final_linear": false,
70
+ "freeze_speech_QFormer": false,
71
+ "freeze_whisper": true,
72
+ "from_stage_1_5": false,
73
+ "image_processor": "google/siglip-so400m-patch14-384",
74
+ "lora_alpha": 16,
75
+ "lora_dropout": 0.05,
76
+ "lora_enable": false,
77
+ "lora_r": 64,
78
+ "mf_split_init": true,
79
+ "mm_mask_drop_mode": "fixed",
80
+ "mm_mask_drop_ratio": 0.25,
81
+ "mm_mask_drop_ratio_lower": null,
82
+ "mm_mask_drop_ratio_upper": null,
83
+ "mm_mask_drop_skip_percentage": 0.0,
84
+ "mm_newline_position": "grid",
85
+ "mm_patch_merge_type": "spatial_unpad",
86
+ "mm_perceiver_depth": 3,
87
+ "mm_perceiver_ff_mult": 4,
88
+ "mm_perceiver_latents": 32,
89
+ "mm_perceiver_pretrained": null,
90
+ "mm_pooling_position": "after",
91
+ "mm_projector_type": "mlp2x_gelu",
92
+ "mm_qformer_depth": 3,
93
+ "mm_qformer_latents": 32,
94
+ "mm_qformer_pretrained": null,
95
+ "mm_resampler_type": "spatial_pool",
96
+ "mm_spatial_pool_mode": "max",
97
+ "mm_spatial_pool_out_channels": 1152,
98
+ "mm_spatial_pool_stride": 2,
99
+ "mm_use_im_patch_token": false,
100
+ "mm_use_im_start_end": false,
101
+ "mm_vision_select_feature": "patch",
102
+ "mm_vision_select_layer": -2,
103
+ "mm_vlmattention_bert_type": "qformer_pretrain",
104
+ "mm_vlmattention_compress_type": null,
105
+ "mm_vlmattention_num_query": 32,
106
+ "mm_vlmattention_pretrained": null,
107
+ "modality_max_length": "None",
108
+ "model_name_or_path": "output/OV-Qwen2-7B-AM9",
109
+ "multi_frame_num": 16,
110
+ "multi_frame_projector": true,
111
+ "niv_cnn_params": "[(10, 5), (3, 2), (3, 2), (3, 2), (3, 2), (2, 2), (2, 2)]",
112
+ "niv_in_channels": 4,
113
+ "niv_out_channels": 401,
114
+ "num_speech_query_token": 1,
115
+ "patchify_video_feature": false,
116
+ "pretrain_mm_mlp_adapter": null,
117
+ "salmonn_path": null,
118
+ "second_per_window": 0.2,
119
+ "second_stride": 0.2,
120
+ "spt_projector": false,
121
+ "tune_mm_mlp_adapter": false,
122
+ "tune_mm_vision_resampler": false,
123
+ "unfreeze_mm_vision_tower": false,
124
+ "use_final_linear": true,
125
+ "use_flash_tower": true,
126
+ "use_mf_attn": false,
127
+ "use_mfcnn": false,
128
+ "use_mftrans": false,
129
+ "use_niv": false,
130
+ "use_speech_Qformer": true,
131
+ "version": "qwen_1_5",
132
+ "vision_tower": "google/siglip-so400m-patch14-384",
133
+ "whisper_lora": false,
134
+ "whisper_path": "openai/whisper-large-v3",
135
+ "window_level_Qformer": true
136
+ },
137
+ "model_type": "qwen2",
138
+ "multi_frame_num": 16,
139
+ "multi_frame_projector": true,
140
+ "num_attention_heads": 28,
141
+ "num_hidden_layers": 28,
142
+ "num_key_value_heads": 4,
143
+ "pos_skipping_range": 4096,
144
+ "rms_norm_eps": 1e-06,
145
+ "rope_scaling": null,
146
+ "rope_theta": 1000000.0,
147
+ "sliding_window": 131072,
148
+ "spt_projector": false,
149
+ "tie_word_embeddings": false,
150
+ "tokenizer_model_max_length": 32768,
151
+ "tokenizer_padding_side": "right",
152
+ "torch_dtype": "bfloat16",
153
+ "transformers_version": "4.39.2",
154
+ "tune_mm_mlp_adapter": false,
155
+ "tune_mm_vision_resampler": false,
156
+ "unfreeze_mm_vision_tower": false,
157
+ "use_cache": false,
158
+ "use_flash_tower": true,
159
+ "use_mf_attn": false,
160
+ "use_mfcnn": false,
161
+ "use_mftrans": false,
162
+ "use_mm_proj": true,
163
+ "use_pos_skipping": false,
164
+ "use_sliding_window": false,
165
+ "vision_tower_pretrained": null,
166
+ "vocab_size": 152064
167
+ }