mazesmazes commited on
Commit
ee255e9
·
verified ·
1 Parent(s): 3768e1a

Training in progress, step 5000

Browse files
Files changed (4) hide show
  1. config.json +239 -322
  2. generation_config.json +4 -5
  3. model.safetensors +2 -2
  4. training_args.bin +1 -1
config.json CHANGED
@@ -4,157 +4,104 @@
4
  ],
5
  "attn_implementation": "sdpa",
6
  "audio_config": {
7
- "_name_or_path": "openai/whisper-base",
8
- "activation_dropout": 0.0,
9
- "activation_function": "gelu",
10
- "apply_spec_augment": false,
11
  "architectures": [
12
- "WhisperForConditionalGeneration"
13
  ],
14
- "attention_dropout": 0.0,
15
- "begin_suppress_tokens": [
16
- 220,
17
- 50257
18
- ],
19
- "bos_token_id": 50257,
20
- "classifier_proj_size": 256,
21
- "d_model": 512,
22
- "decoder_attention_heads": 8,
23
- "decoder_ffn_dim": 2048,
24
- "decoder_layerdrop": 0.0,
25
- "decoder_layers": 6,
26
- "decoder_start_token_id": 50258,
27
- "dropout": 0.0,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
28
  "dtype": "bfloat16",
29
- "encoder_attention_heads": 8,
30
- "encoder_ffn_dim": 2048,
31
- "encoder_layerdrop": 0.0,
32
- "encoder_layers": 6,
33
- "eos_token_id": 50257,
34
- "forced_decoder_ids": [
35
- [
36
- 1,
37
- 50259
 
 
 
 
 
 
 
38
  ],
39
- [
40
- 2,
41
- 50359
42
- ],
43
- [
44
- 3,
45
- 50363
46
- ]
47
- ],
48
- "init_std": 0.02,
49
- "mask_feature_length": 10,
50
- "mask_feature_min_masks": 0,
51
- "mask_feature_prob": 0.0,
52
- "mask_time_length": 10,
53
- "mask_time_min_masks": 2,
54
- "mask_time_prob": 0.05,
55
- "max_source_positions": 1500,
56
- "max_target_positions": 448,
57
- "median_filter_width": 7,
58
- "model_type": "whisper",
59
- "num_mel_bins": 80,
60
- "pad_token_id": 50257,
61
- "scale_embedding": false,
62
- "suppress_tokens": [
63
- 1,
64
- 2,
65
- 7,
66
- 8,
67
- 9,
68
- 10,
69
- 14,
70
- 25,
71
- 26,
72
- 27,
73
- 28,
74
- 29,
75
- 31,
76
- 58,
77
- 59,
78
- 60,
79
- 61,
80
- 62,
81
- 63,
82
- 90,
83
- 91,
84
- 92,
85
- 93,
86
- 359,
87
- 503,
88
- 522,
89
- 542,
90
- 873,
91
- 893,
92
- 902,
93
- 918,
94
- 922,
95
- 931,
96
- 1350,
97
- 1853,
98
- 1982,
99
- 2460,
100
- 2627,
101
- 3246,
102
- 3253,
103
- 3268,
104
- 3536,
105
- 3846,
106
- 3961,
107
- 4183,
108
- 4667,
109
- 6585,
110
- 6647,
111
- 7273,
112
- 9061,
113
- 9383,
114
- 10428,
115
- 10929,
116
- 11938,
117
- 12033,
118
- 12331,
119
- 12562,
120
- 13793,
121
- 14157,
122
- 14635,
123
- 15265,
124
- 15618,
125
- 16553,
126
- 16604,
127
- 18362,
128
- 18956,
129
- 20075,
130
- 21675,
131
- 22520,
132
- 26130,
133
- 26161,
134
- 26435,
135
- 28279,
136
- 29464,
137
- 31650,
138
- 32302,
139
- 32470,
140
- 36865,
141
- 42863,
142
- 47425,
143
- 49870,
144
- 50254,
145
- 50258,
146
- 50358,
147
- 50359,
148
- 50360,
149
- 50361,
150
- 50362
151
- ],
152
- "tie_word_embeddings": true,
153
- "use_cache": true,
154
- "use_weighted_layer_sum": false,
155
- "vocab_size": 51865
156
  },
157
- "audio_model_id": "openai/whisper-base",
158
  "audio_sample_rate": 16000,
159
  "auto_map": {
160
  "AutoConfig": "asr_config.ASRConfig",
@@ -176,155 +123,102 @@
176
  "downsample_rate": 5,
177
  "dtype": "bfloat16",
178
  "encoder": {
179
- "_name_or_path": "openai/whisper-base",
180
- "activation_dropout": 0.0,
181
- "activation_function": "gelu",
182
- "apply_spec_augment": false,
183
  "architectures": [
184
- "WhisperForConditionalGeneration"
185
- ],
186
- "attention_dropout": 0.0,
187
- "begin_suppress_tokens": [
188
- 220,
189
- 50257
190
  ],
191
- "bos_token_id": 50257,
192
- "classifier_proj_size": 256,
193
- "d_model": 512,
194
- "decoder_attention_heads": 8,
195
- "decoder_ffn_dim": 2048,
196
- "decoder_layerdrop": 0.0,
197
- "decoder_layers": 6,
198
- "decoder_start_token_id": 50258,
199
- "dropout": 0.0,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
200
  "dtype": "bfloat16",
201
- "encoder_attention_heads": 8,
202
- "encoder_ffn_dim": 2048,
203
- "encoder_layerdrop": 0.0,
204
- "encoder_layers": 6,
205
- "eos_token_id": 50257,
206
- "forced_decoder_ids": [
207
- [
208
- 1,
209
- 50259
210
- ],
211
- [
212
- 2,
213
- 50359
 
 
 
214
  ],
215
- [
216
- 3,
217
- 50363
218
- ]
219
- ],
220
- "init_std": 0.02,
221
- "mask_feature_length": 10,
222
- "mask_feature_min_masks": 0,
223
- "mask_feature_prob": 0.0,
224
- "mask_time_length": 10,
225
- "mask_time_min_masks": 2,
226
- "mask_time_prob": 0.05,
227
- "max_source_positions": 1500,
228
- "max_target_positions": 448,
229
- "median_filter_width": 7,
230
- "model_type": "whisper",
231
- "num_mel_bins": 80,
232
- "pad_token_id": 50257,
233
- "scale_embedding": false,
234
- "suppress_tokens": [
235
- 1,
236
- 2,
237
- 7,
238
- 8,
239
- 9,
240
- 10,
241
- 14,
242
- 25,
243
- 26,
244
- 27,
245
- 28,
246
- 29,
247
- 31,
248
- 58,
249
- 59,
250
- 60,
251
- 61,
252
- 62,
253
- 63,
254
- 90,
255
- 91,
256
- 92,
257
- 93,
258
- 359,
259
- 503,
260
- 522,
261
- 542,
262
- 873,
263
- 893,
264
- 902,
265
- 918,
266
- 922,
267
- 931,
268
- 1350,
269
- 1853,
270
- 1982,
271
- 2460,
272
- 2627,
273
- 3246,
274
- 3253,
275
- 3268,
276
- 3536,
277
- 3846,
278
- 3961,
279
- 4183,
280
- 4667,
281
- 6585,
282
- 6647,
283
- 7273,
284
- 9061,
285
- 9383,
286
- 10428,
287
- 10929,
288
- 11938,
289
- 12033,
290
- 12331,
291
- 12562,
292
- 13793,
293
- 14157,
294
- 14635,
295
- 15265,
296
- 15618,
297
- 16553,
298
- 16604,
299
- 18362,
300
- 18956,
301
- 20075,
302
- 21675,
303
- 22520,
304
- 26130,
305
- 26161,
306
- 26435,
307
- 28279,
308
- 29464,
309
- 31650,
310
- 32302,
311
- 32470,
312
- 36865,
313
- 42863,
314
- 47425,
315
- 49870,
316
- 50254,
317
- 50258,
318
- 50358,
319
- 50359,
320
- 50360,
321
- 50361,
322
- 50362
323
- ],
324
- "tie_word_embeddings": true,
325
- "use_cache": true,
326
- "use_weighted_layer_sum": false,
327
- "vocab_size": 51865
328
  },
329
  "encoder_conv_layers": [
330
  [
@@ -338,13 +232,13 @@
338
  2
339
  ]
340
  ],
341
- "encoder_dim": 512,
342
  "freeze_projector": false,
343
  "freq_mask_length": 27,
344
  "inference_warmup_tokens": 10,
345
  "label_smoothing": 0.0,
346
  "length_penalty": 1.0,
347
- "llm_dim": 960,
348
  "lora_alpha": 32,
349
  "lora_dropout": 0.0,
350
  "lora_rank": 8,
@@ -370,7 +264,7 @@
370
  "pipeline_tag": "automatic-speech-recognition",
371
  "pretrained_model_path": "mazesmazes/tiny-audio-embedded",
372
  "projector_dropout": 0.0,
373
- "projector_hidden_dim": 512,
374
  "projector_init_std": 0.02,
375
  "projector_num_layers": 2,
376
  "projector_pool_stride": 4,
@@ -385,46 +279,69 @@
385
  "system_prompt": "",
386
  "temperature": null,
387
  "text_config": {
388
- "_name_or_path": "HuggingFaceTB/SmolLM2-360M-Instruct",
389
  "architectures": [
390
- "LlamaForCausalLM"
391
  ],
392
  "attention_bias": false,
393
  "attention_dropout": 0.0,
394
- "bos_token_id": 1,
395
  "dtype": "bfloat16",
396
- "eos_token_id": 2,
397
- "head_dim": 64,
398
  "hidden_act": "silu",
399
- "hidden_size": 960,
400
  "initializer_range": 0.02,
401
- "intermediate_size": 2560,
402
- "is_llama_config": true,
403
- "max_position_embeddings": 8192,
404
- "mlp_bias": false,
405
- "model_type": "llama",
406
- "num_attention_heads": 15,
407
- "num_hidden_layers": 32,
408
- "num_key_value_heads": 5,
409
- "pad_token_id": 2,
410
- "pretraining_tp": 1,
411
- "rms_norm_eps": 1e-05,
412
- "rope_interleaved": false,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
413
  "rope_parameters": {
414
- "rope_theta": 100000,
415
  "rope_type": "default"
416
  },
 
417
  "tie_word_embeddings": true,
418
- "transformers.js_config": {
419
- "kv_cache_dtype": {
420
- "fp16": "float16",
421
- "q4f16": "float16"
422
- }
423
- },
424
  "use_cache": true,
425
- "vocab_size": 49153
 
426
  },
427
- "text_model_id": "HuggingFaceTB/SmolLM2-360M-Instruct",
428
  "time_mask_length": 100,
429
  "top_k": null,
430
  "top_p": null,
@@ -432,5 +349,5 @@
432
  "use_cache": false,
433
  "use_lora": false,
434
  "use_specaugment": true,
435
- "vocab_size": 49153
436
  }
 
4
  ],
5
  "attn_implementation": "sdpa",
6
  "audio_config": {
7
+ "_name_or_path": "zai-org/GLM-ASR-Nano-2512",
 
 
 
8
  "architectures": [
9
+ "GlmAsrForConditionalGeneration"
10
  ],
11
+ "audio_config": {
12
+ "_name_or_path": "",
13
+ "architectures": null,
14
+ "attention_dropout": 0.0,
15
+ "chunk_size_feed_forward": 0,
16
+ "dtype": null,
17
+ "head_dim": 64,
18
+ "hidden_act": "gelu",
19
+ "hidden_size": 1280,
20
+ "id2label": {
21
+ "0": "LABEL_0",
22
+ "1": "LABEL_1"
23
+ },
24
+ "initializer_range": 0.02,
25
+ "intermediate_size": 5120,
26
+ "is_encoder_decoder": false,
27
+ "label2id": {
28
+ "LABEL_0": 0,
29
+ "LABEL_1": 1
30
+ },
31
+ "max_position_embeddings": 1500,
32
+ "model_type": "glmasr_encoder",
33
+ "num_attention_heads": 20,
34
+ "num_hidden_layers": 32,
35
+ "num_key_value_heads": 20,
36
+ "num_mel_bins": 128,
37
+ "output_attentions": false,
38
+ "output_hidden_states": false,
39
+ "partial_rotary_factor": 0.5,
40
+ "problem_type": null,
41
+ "return_dict": true,
42
+ "rope_parameters": {
43
+ "partial_rotary_factor": 0.5,
44
+ "rope_theta": 10000.0,
45
+ "rope_type": "default"
46
+ }
47
+ },
48
+ "audio_token_id": 59260,
49
  "dtype": "bfloat16",
50
+ "hidden_size": 2048,
51
+ "model_type": "glmasr",
52
+ "num_mel_bins": 128,
53
+ "projector_hidden_act": "gelu",
54
+ "text_config": {
55
+ "_name_or_path": "",
56
+ "architectures": null,
57
+ "attention_bias": false,
58
+ "attention_dropout": 0.0,
59
+ "bos_token_id": 1,
60
+ "chunk_size_feed_forward": 0,
61
+ "dtype": null,
62
+ "eos_token_id": [
63
+ 59246,
64
+ 59253,
65
+ 59255
66
  ],
67
+ "head_dim": 128,
68
+ "hidden_act": "silu",
69
+ "hidden_size": 2048,
70
+ "id2label": {
71
+ "0": "LABEL_0",
72
+ "1": "LABEL_1"
73
+ },
74
+ "initializer_range": 0.02,
75
+ "intermediate_size": 6144,
76
+ "is_encoder_decoder": false,
77
+ "label2id": {
78
+ "LABEL_0": 0,
79
+ "LABEL_1": 1
80
+ },
81
+ "max_position_embeddings": 8192,
82
+ "mlp_bias": false,
83
+ "model_type": "llama",
84
+ "num_attention_heads": 16,
85
+ "num_hidden_layers": 28,
86
+ "num_key_value_heads": 4,
87
+ "output_attentions": false,
88
+ "output_hidden_states": false,
89
+ "pad_token_id": null,
90
+ "pretraining_tp": 1,
91
+ "problem_type": null,
92
+ "return_dict": true,
93
+ "rms_norm_eps": 1e-05,
94
+ "rope_parameters": {
95
+ "rope_theta": 10000.0,
96
+ "rope_type": "default"
97
+ },
98
+ "tie_word_embeddings": false,
99
+ "use_cache": true,
100
+ "vocab_size": 59264
101
+ },
102
+ "vocab_size": 59264
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
103
  },
104
+ "audio_model_id": "zai-org/GLM-ASR-Nano-2512",
105
  "audio_sample_rate": 16000,
106
  "auto_map": {
107
  "AutoConfig": "asr_config.ASRConfig",
 
123
  "downsample_rate": 5,
124
  "dtype": "bfloat16",
125
  "encoder": {
126
+ "_name_or_path": "zai-org/GLM-ASR-Nano-2512",
 
 
 
127
  "architectures": [
128
+ "GlmAsrForConditionalGeneration"
 
 
 
 
 
129
  ],
130
+ "audio_config": {
131
+ "_name_or_path": "",
132
+ "architectures": null,
133
+ "attention_dropout": 0.0,
134
+ "chunk_size_feed_forward": 0,
135
+ "dtype": null,
136
+ "head_dim": 64,
137
+ "hidden_act": "gelu",
138
+ "hidden_size": 1280,
139
+ "id2label": {
140
+ "0": "LABEL_0",
141
+ "1": "LABEL_1"
142
+ },
143
+ "initializer_range": 0.02,
144
+ "intermediate_size": 5120,
145
+ "is_encoder_decoder": false,
146
+ "label2id": {
147
+ "LABEL_0": 0,
148
+ "LABEL_1": 1
149
+ },
150
+ "max_position_embeddings": 1500,
151
+ "model_type": "glmasr_encoder",
152
+ "num_attention_heads": 20,
153
+ "num_hidden_layers": 32,
154
+ "num_key_value_heads": 20,
155
+ "num_mel_bins": 128,
156
+ "output_attentions": false,
157
+ "output_hidden_states": false,
158
+ "partial_rotary_factor": 0.5,
159
+ "problem_type": null,
160
+ "return_dict": true,
161
+ "rope_parameters": {
162
+ "partial_rotary_factor": 0.5,
163
+ "rope_theta": 10000.0,
164
+ "rope_type": "default"
165
+ }
166
+ },
167
+ "audio_token_id": 59260,
168
  "dtype": "bfloat16",
169
+ "hidden_size": 2048,
170
+ "model_type": "glmasr",
171
+ "num_mel_bins": 128,
172
+ "projector_hidden_act": "gelu",
173
+ "text_config": {
174
+ "_name_or_path": "",
175
+ "architectures": null,
176
+ "attention_bias": false,
177
+ "attention_dropout": 0.0,
178
+ "bos_token_id": 1,
179
+ "chunk_size_feed_forward": 0,
180
+ "dtype": null,
181
+ "eos_token_id": [
182
+ 59246,
183
+ 59253,
184
+ 59255
185
  ],
186
+ "head_dim": 128,
187
+ "hidden_act": "silu",
188
+ "hidden_size": 2048,
189
+ "id2label": {
190
+ "0": "LABEL_0",
191
+ "1": "LABEL_1"
192
+ },
193
+ "initializer_range": 0.02,
194
+ "intermediate_size": 6144,
195
+ "is_encoder_decoder": false,
196
+ "label2id": {
197
+ "LABEL_0": 0,
198
+ "LABEL_1": 1
199
+ },
200
+ "max_position_embeddings": 8192,
201
+ "mlp_bias": false,
202
+ "model_type": "llama",
203
+ "num_attention_heads": 16,
204
+ "num_hidden_layers": 28,
205
+ "num_key_value_heads": 4,
206
+ "output_attentions": false,
207
+ "output_hidden_states": false,
208
+ "pad_token_id": null,
209
+ "pretraining_tp": 1,
210
+ "problem_type": null,
211
+ "return_dict": true,
212
+ "rms_norm_eps": 1e-05,
213
+ "rope_parameters": {
214
+ "rope_theta": 10000.0,
215
+ "rope_type": "default"
216
+ },
217
+ "tie_word_embeddings": false,
218
+ "use_cache": true,
219
+ "vocab_size": 59264
220
+ },
221
+ "vocab_size": 59264
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222
  },
223
  "encoder_conv_layers": [
224
  [
 
232
  2
233
  ]
234
  ],
235
+ "encoder_dim": 1280,
236
  "freeze_projector": false,
237
  "freq_mask_length": 27,
238
  "inference_warmup_tokens": 10,
239
  "label_smoothing": 0.0,
240
  "length_penalty": 1.0,
241
+ "llm_dim": 1024,
242
  "lora_alpha": 32,
243
  "lora_dropout": 0.0,
244
  "lora_rank": 8,
 
264
  "pipeline_tag": "automatic-speech-recognition",
265
  "pretrained_model_path": "mazesmazes/tiny-audio-embedded",
266
  "projector_dropout": 0.0,
267
+ "projector_hidden_dim": 1024,
268
  "projector_init_std": 0.02,
269
  "projector_num_layers": 2,
270
  "projector_pool_stride": 4,
 
279
  "system_prompt": "",
280
  "temperature": null,
281
  "text_config": {
282
+ "_name_or_path": "Qwen/Qwen3-0.6B",
283
  "architectures": [
284
+ "Qwen3ForCausalLM"
285
  ],
286
  "attention_bias": false,
287
  "attention_dropout": 0.0,
288
+ "bos_token_id": null,
289
  "dtype": "bfloat16",
290
+ "eos_token_id": 151645,
291
+ "head_dim": 128,
292
  "hidden_act": "silu",
293
+ "hidden_size": 1024,
294
  "initializer_range": 0.02,
295
+ "intermediate_size": 3072,
296
+ "layer_types": [
297
+ "full_attention",
298
+ "full_attention",
299
+ "full_attention",
300
+ "full_attention",
301
+ "full_attention",
302
+ "full_attention",
303
+ "full_attention",
304
+ "full_attention",
305
+ "full_attention",
306
+ "full_attention",
307
+ "full_attention",
308
+ "full_attention",
309
+ "full_attention",
310
+ "full_attention",
311
+ "full_attention",
312
+ "full_attention",
313
+ "full_attention",
314
+ "full_attention",
315
+ "full_attention",
316
+ "full_attention",
317
+ "full_attention",
318
+ "full_attention",
319
+ "full_attention",
320
+ "full_attention",
321
+ "full_attention",
322
+ "full_attention",
323
+ "full_attention",
324
+ "full_attention"
325
+ ],
326
+ "max_position_embeddings": 40960,
327
+ "max_window_layers": 28,
328
+ "model_type": "qwen3",
329
+ "num_attention_heads": 16,
330
+ "num_hidden_layers": 28,
331
+ "num_key_value_heads": 8,
332
+ "pad_token_id": 151643,
333
+ "rms_norm_eps": 1e-06,
334
  "rope_parameters": {
335
+ "rope_theta": 1000000,
336
  "rope_type": "default"
337
  },
338
+ "sliding_window": null,
339
  "tie_word_embeddings": true,
 
 
 
 
 
 
340
  "use_cache": true,
341
+ "use_sliding_window": false,
342
+ "vocab_size": 151670
343
  },
344
+ "text_model_id": "Qwen/Qwen3-0.6B",
345
  "time_mask_length": 100,
346
  "top_k": null,
347
  "top_p": null,
 
349
  "use_cache": false,
350
  "use_lora": false,
351
  "use_specaugment": true,
352
+ "vocab_size": 151670
353
  }
generation_config.json CHANGED
@@ -1,17 +1,16 @@
1
  {
2
- "_from_model_config": true,
3
- "bos_token_id": 1,
4
  "do_sample": false,
5
  "eos_token_id": [
6
- 2,
7
- 0
8
  ],
9
  "length_penalty": 1.0,
10
  "max_new_tokens": 128,
11
  "min_new_tokens": 0,
12
  "no_repeat_ngram_size": 0,
13
  "num_beams": 1,
14
- "pad_token_id": 2,
15
  "repetition_penalty": 1.0,
16
  "transformers_version": "5.6.0",
17
  "use_cache": true
 
1
  {
2
+ "bos_token_id": 151643,
 
3
  "do_sample": false,
4
  "eos_token_id": [
5
+ 151645,
6
+ 151643
7
  ],
8
  "length_penalty": 1.0,
9
  "max_new_tokens": 128,
10
  "min_new_tokens": 0,
11
  "no_repeat_ngram_size": 0,
12
  "num_beams": 1,
13
+ "pad_token_id": 151643,
14
  "repetition_penalty": 1.0,
15
  "transformers_version": "5.6.0",
16
  "use_cache": true
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8f8254c1fa042645b6f227181d85707442d3fd60855d11852e73d31111159f8e
3
- size 3081536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5ccfa44170cd0997b704c9a2c1f2851db65c958cc6b3fd85c53dcee85f0ef7a6
3
+ size 12585288
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d68647bfb17c549837ff4fcdc9552213db8622a91dfb51c29ff5d455023d851b
3
  size 5329
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bb18fa029dd1c46ae835bb3aa2c1aa8c1ada5d7dbb158a927ed0a51f7a0f96d2
3
  size 5329