Ghaser commited on
Commit
cffa6e9
·
verified ·
1 Parent(s): f09132e

Upload config.json

Browse files
Files changed (1) hide show
  1. config.json +56 -18
config.json CHANGED
@@ -165,37 +165,75 @@
165
  "use_sliding_window": false,
166
  "video_token_id": 151656,
167
  "vision_config": {
168
- "fullatt_block_indexes": [
169
- 7,
170
- 15,
171
- 23,
172
- 31
173
- ],
174
- "hidden_act": "silu",
175
- "hidden_size": 1280,
176
- "in_chans": 3,
177
- "intermediate_size": 3420,
178
- "model_type": "qwen2_vl",
179
- "out_hidden_size": 3584,
180
- "spatial_patch_size": 14,
181
- "tokens_per_second": 2,
182
- "window_size": 112
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
183
  },
184
  "vision_end_token_id": 151653,
185
  "vision_hidden_size": 1152,
186
  "vision_spatial_pool_stride": 12,
187
  "vision_start_token_id": 151652,
188
  "vision_token_id": 151654,
189
- "vision_tower": "google/siglip-so400m-patch14-384",
190
  "vision_tower_pretrained": null,
191
  "vision_vision_select_layer": -2,
192
- "vision_vision_tower": "google/siglip-so400m-patch14-384",
193
  "visual_aligner_type": "linear",
194
  "vocab_size": 152064,
195
  "whisper_hidden_size": 1280,
196
  "whisper_projector_type": "linear",
197
  "whisper_query_tokens_size": 200,
198
- "whisper_tower": "openai/whisper-large-v3",
199
  "idim": 2048,
200
  "odim": 4096,
201
  "encoder_pre_norm_type": "ln",
 
165
  "use_sliding_window": false,
166
  "video_token_id": 151656,
167
  "vision_config": {
168
+ "hidden_size": 1152,
169
+ "image_size": 384,
170
+ "intermediate_size": 4304,
171
+ "model_type": "siglip_vision_model",
172
+ "num_attention_heads": 16,
173
+ "num_hidden_layers": 27,
174
+ "patch_size": 14
175
+ },
176
+ "audio_config": {
177
+ "activation_dropout": 0.0,
178
+ "activation_function": "gelu",
179
+ "apply_spec_augment": false,
180
+ "architectures": [
181
+ "WhisperForConditionalGeneration"
182
+ ],
183
+ "attention_dropout": 0.0,
184
+ "begin_suppress_tokens": [
185
+ 220,
186
+ 50257
187
+ ],
188
+ "bos_token_id": 50257,
189
+ "classifier_proj_size": 256,
190
+ "d_model": 1280,
191
+ "decoder_attention_heads": 20,
192
+ "decoder_ffn_dim": 5120,
193
+ "decoder_layerdrop": 0.0,
194
+ "decoder_layers": 32,
195
+ "decoder_start_token_id": 50258,
196
+ "dropout": 0.0,
197
+ "encoder_attention_heads": 20,
198
+ "encoder_ffn_dim": 5120,
199
+ "encoder_layerdrop": 0.0,
200
+ "encoder_layers": 32,
201
+ "eos_token_id": 50257,
202
+ "init_std": 0.02,
203
+ "is_encoder_decoder": true,
204
+ "mask_feature_length": 10,
205
+ "mask_feature_min_masks": 0,
206
+ "mask_feature_prob": 0.0,
207
+ "mask_time_length": 10,
208
+ "mask_time_min_masks": 2,
209
+ "mask_time_prob": 0.05,
210
+ "max_length": 448,
211
+ "max_source_positions": 1500,
212
+ "max_target_positions": 448,
213
+ "median_filter_width": 7,
214
+ "model_type": "whisper",
215
+ "num_hidden_layers": 32,
216
+ "num_mel_bins": 128,
217
+ "pad_token_id": 50256,
218
+ "scale_embedding": false,
219
+ "torch_dtype": "float16",
220
+ "transformers_version": "4.36.0.dev0",
221
+ "use_cache": true,
222
+ "use_weighted_layer_sum": false,
223
+ "vocab_size": 51866
224
  },
225
  "vision_end_token_id": 151653,
226
  "vision_hidden_size": 1152,
227
  "vision_spatial_pool_stride": 12,
228
  "vision_start_token_id": 151652,
229
  "vision_token_id": 151654,
 
230
  "vision_tower_pretrained": null,
231
  "vision_vision_select_layer": -2,
 
232
  "visual_aligner_type": "linear",
233
  "vocab_size": 152064,
234
  "whisper_hidden_size": 1280,
235
  "whisper_projector_type": "linear",
236
  "whisper_query_tokens_size": 200,
 
237
  "idim": 2048,
238
  "odim": 4096,
239
  "encoder_pre_norm_type": "ln",