Hugging Face's logo

czyhust
/
moss

Safetensors
Model card Files
xet
 Community
czyhust commited on
Commit
a75d462
·
verified ·
1 Parent(s): 49ff536

Initial upload (part 4)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +2 -0
  2. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/config.json +250 -0
  3. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/generation_config.json +9 -0
  4. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/processor_config.json +22 -0
  5. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/tokenizer_config.json +22 -0
  6. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/trainer_state.json +1714 -0
  7. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/chat_template.jinja +103 -0
  8. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/config.json +250 -0
  9. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/generation_config.json +9 -0
  10. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/processor_config.json +22 -0
  11. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/tokenizer_config.json +22 -0
  12. MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/trainer_state.json +1721 -0
  13. MOSS_Speaker-RoPE/output_lr2e4/config.json +250 -0
  14. MOSS_Speaker-RoPE/output_lr2e4/generation_config.json +9 -0
  15. MOSS_Speaker-RoPE/output_lr2e4/model.safetensors +3 -0
  16. MOSS_Speaker-RoPE/output_lr2e4/processor_config.json +22 -0
  17. MOSS_Speaker-RoPE/output_lr2e4/tokenizer.json +3 -0
  18. MOSS_Speaker-RoPE/output_lr2e4/tokenizer_config.json +22 -0
  19. MOSS_Speaker-RoPE/output_lr2e4/train_results.json +8 -0
  20. MOSS_Speaker-RoPE/output_lr2e4/trainer_state.json +1730 -0
  21. MOSS_Speaker-RoPE/output_lr2e4/training_args.bin +3 -0
  22. MOSS_Speaker-RoPE/output_spk_odd8_s3/all_results.json +8 -0
  23. MOSS_Speaker-RoPE/output_spk_odd8_s3/chat_template.jinja +103 -0
  24. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/chat_template.jinja +103 -0
  25. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/config.json +260 -0
  26. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/generation_config.json +9 -0
  27. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/processor_config.json +22 -0
  28. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/tokenizer_config.json +22 -0
  29. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/trainer_state.json +1714 -0
  30. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/chat_template.jinja +103 -0
  31. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/config.json +260 -0
  32. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/generation_config.json +9 -0
  33. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/processor_config.json +22 -0
  34. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/tokenizer_config.json +22 -0
  35. MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/trainer_state.json +1721 -0
  36. MOSS_Speaker-RoPE/output_spk_odd8_s3/config.json +260 -0
  37. MOSS_Speaker-RoPE/output_spk_odd8_s3/generation_config.json +9 -0
  38. MOSS_Speaker-RoPE/output_spk_odd8_s3/model.safetensors +3 -0
  39. MOSS_Speaker-RoPE/output_spk_odd8_s3/processor_config.json +22 -0
  40. MOSS_Speaker-RoPE/output_spk_odd8_s3/tokenizer.json +3 -0
  41. MOSS_Speaker-RoPE/output_spk_odd8_s3/tokenizer_config.json +22 -0
  42. MOSS_Speaker-RoPE/output_spk_odd8_s3/train_results.json +8 -0
  43. MOSS_Speaker-RoPE/output_spk_odd8_s3/trainer_state.json +1730 -0
  44. MOSS_Speaker-RoPE/output_spk_odd8_s3/training_args.bin +3 -0
  45. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/chat_template.jinja +103 -0
  46. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/config.json +260 -0
  47. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/generation_config.json +9 -0
  48. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/processor_config.json +22 -0
  49. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/tokenizer_config.json +22 -0
  50. MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/trainer_state.json +594 -0
.gitattributes CHANGED
@@ -53,3 +53,5 @@ MOSS_Speaker-RoPE/tests/speaker_embed_postnorm.png filter=lfs diff=lfs merge=lfs
53
  MOSS_Speaker-RoPE/tests/speaker_spkw5_heatmap.png filter=lfs diff=lfs merge=lfs -text
54
  MOSS_Speaker-RoPE/tests/speaker_embed_heatmap_norm.png filter=lfs diff=lfs merge=lfs -text
55
  MOSS_Speaker-RoPE/data/train_40min_new.jsonl filter=lfs diff=lfs merge=lfs -text
 
 
 
53
  MOSS_Speaker-RoPE/tests/speaker_spkw5_heatmap.png filter=lfs diff=lfs merge=lfs -text
54
  MOSS_Speaker-RoPE/tests/speaker_embed_heatmap_norm.png filter=lfs diff=lfs merge=lfs -text
55
  MOSS_Speaker-RoPE/data/train_40min_new.jsonl filter=lfs diff=lfs merge=lfs -text
56
+ MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_spkw5_back/checkpoint-1207/tokenizer.json filter=lfs diff=lfs merge=lfs -text
57
+ MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_spkw5_back/tokenizer.json filter=lfs diff=lfs merge=lfs -text
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/config.json ADDED
@@ -0,0 +1,250 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_rotary_dim": 64,
171
+ "speaker_rotary_inference_seed": 42,
172
+ "speaker_rotary_std": 1.0,
173
+ "speaker_sample_rate": 16000,
174
+ "speaker_tfmel_args": {
175
+ "do_preemph": true,
176
+ "do_spec_aug": false,
177
+ "eps": 1e-08,
178
+ "f_max": 7600,
179
+ "f_min": 20,
180
+ "hop_length": 160,
181
+ "n_fft": 512,
182
+ "n_mels": 72,
183
+ "norm_signal": true,
184
+ "win_length": 400
185
+ },
186
+ "speaker_window_seconds": 2.0,
187
+ "text_config": {
188
+ "attention_bias": false,
189
+ "attention_dropout": 0.0,
190
+ "bos_token_id": null,
191
+ "dtype": "bfloat16",
192
+ "eos_token_id": null,
193
+ "head_dim": 128,
194
+ "hidden_act": "silu",
195
+ "hidden_size": 1024,
196
+ "initializer_range": 0.02,
197
+ "intermediate_size": 3072,
198
+ "layer_types": [
199
+ "full_attention",
200
+ "full_attention",
201
+ "full_attention",
202
+ "full_attention",
203
+ "full_attention",
204
+ "full_attention",
205
+ "full_attention",
206
+ "full_attention",
207
+ "full_attention",
208
+ "full_attention",
209
+ "full_attention",
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention"
227
+ ],
228
+ "max_position_embeddings": 131072,
229
+ "max_window_layers": 28,
230
+ "model_type": "qwen3",
231
+ "num_attention_heads": 16,
232
+ "num_hidden_layers": 28,
233
+ "num_key_value_heads": 8,
234
+ "pad_token_id": 151643,
235
+ "partial_rotary_factor": 0.5,
236
+ "rms_norm_eps": 1e-06,
237
+ "rope_parameters": {
238
+ "rope_theta": 1000000,
239
+ "rope_type": "default"
240
+ },
241
+ "sliding_window": null,
242
+ "tie_word_embeddings": true,
243
+ "use_cache": false,
244
+ "use_sliding_window": false,
245
+ "vocab_size": 151936
246
+ },
247
+ "tie_word_embeddings": true,
248
+ "transformers_version": "5.14.0",
249
+ "use_cache": false
250
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1200/trainer_state.json ADDED
@@ -0,0 +1,1714 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.9942004971002486,
6
+ "eval_steps": 500,
7
+ "global_step": 1200,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 1147.55224609375,
15
+ "learning_rate": 6.611570247933885e-06,
16
+ "loss": 11.37186279296875,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 246.35504150390625,
22
+ "learning_rate": 1.487603305785124e-05,
23
+ "loss": 9.2022216796875,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 79.90882110595703,
29
+ "learning_rate": 2.3140495867768598e-05,
30
+ "loss": 7.6917167663574215,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 158.79872131347656,
36
+ "learning_rate": 3.1404958677685955e-05,
37
+ "loss": 6.138848114013672,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 48.520408630371094,
43
+ "learning_rate": 3.9669421487603306e-05,
44
+ "loss": 5.638393402099609,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 14.776691436767578,
50
+ "learning_rate": 4.793388429752066e-05,
51
+ "loss": 5.221096801757812,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 12.125936508178711,
57
+ "learning_rate": 5.619834710743802e-05,
58
+ "loss": 4.968892288208008,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 14.540782928466797,
64
+ "learning_rate": 6.446280991735537e-05,
65
+ "loss": 5.3284343719482425,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 13.903443336486816,
71
+ "learning_rate": 7.272727272727273e-05,
72
+ "loss": 5.018967819213867,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 12.485719680786133,
78
+ "learning_rate": 8.099173553719009e-05,
79
+ "loss": 4.600848770141601,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 44.14723205566406,
85
+ "learning_rate": 8.925619834710744e-05,
86
+ "loss": 4.335272979736328,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 9.01069164276123,
92
+ "learning_rate": 9.75206611570248e-05,
93
+ "loss": 4.316203308105469,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 16.169389724731445,
99
+ "learning_rate": 0.00010578512396694216,
100
+ "loss": 4.197259140014649,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 6.058357238769531,
106
+ "learning_rate": 0.0001140495867768595,
107
+ "loss": 4.562212371826172,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 5.126786231994629,
113
+ "learning_rate": 0.00012231404958677685,
114
+ "loss": 4.340535354614258,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 12.424266815185547,
120
+ "learning_rate": 0.00013057851239669423,
121
+ "loss": 3.9918304443359376,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 8.042577743530273,
127
+ "learning_rate": 0.0001388429752066116,
128
+ "loss": 4.087551879882812,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 8.642739295959473,
134
+ "learning_rate": 0.00014710743801652894,
135
+ "loss": 4.22486686706543,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 8.129426002502441,
141
+ "learning_rate": 0.00015537190082644627,
142
+ "loss": 3.8094669342041017,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 10.393760681152344,
148
+ "learning_rate": 0.00016363636363636366,
149
+ "loss": 3.687621307373047,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 23.797346115112305,
155
+ "learning_rate": 0.00017190082644628102,
156
+ "loss": 4.8999065399169925,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 7.251102924346924,
162
+ "learning_rate": 0.00018016528925619835,
163
+ "loss": 4.111215972900391,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 7.632305145263672,
169
+ "learning_rate": 0.0001884297520661157,
170
+ "loss": 3.88475341796875,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 2.708984851837158,
176
+ "learning_rate": 0.0001966942148760331,
177
+ "loss": 3.9735763549804686,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 3.5484111309051514,
183
+ "learning_rate": 0.00019999623426388962,
184
+ "loss": 3.911006546020508,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 3.580739974975586,
190
+ "learning_rate": 0.0001999732224591804,
191
+ "loss": 3.6748367309570313,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 3.1602091789245605,
197
+ "learning_rate": 0.0001999292957337329,
198
+ "loss": 4.212306213378906,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 2.242072582244873,
204
+ "learning_rate": 0.0001998644632772477,
205
+ "loss": 3.8557418823242187,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 3.2909867763519287,
211
+ "learning_rate": 0.00019977873865301363,
212
+ "loss": 3.953845977783203,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 2.695601224899292,
218
+ "learning_rate": 0.00019967213979507018,
219
+ "loss": 3.7311931610107423,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 3.831533670425415,
225
+ "learning_rate": 0.00019954468900445566,
226
+ "loss": 3.88598518371582,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 4.886938571929932,
232
+ "learning_rate": 0.00019939641294454172,
233
+ "loss": 3.8442028045654295,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 3.451791524887085,
239
+ "learning_rate": 0.00019922734263545517,
240
+ "loss": 3.9103469848632812,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 3.0034420490264893,
246
+ "learning_rate": 0.00019903751344758848,
247
+ "loss": 4.032486343383789,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 2.849339723587036,
253
+ "learning_rate": 0.00019882696509420007,
254
+ "loss": 3.407342529296875,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 2.095475435256958,
260
+ "learning_rate": 0.00019859574162310608,
261
+ "loss": 3.7116188049316405,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 3.148158073425293,
267
+ "learning_rate": 0.0001983438914074654,
268
+ "loss": 3.7624229431152343,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 2.4163601398468018,
274
+ "learning_rate": 0.00019807146713565955,
275
+ "loss": 3.439311218261719,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 2.7623701095581055,
281
+ "learning_rate": 0.00019777852580027042,
282
+ "loss": 3.8211971282958985,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 3.145005226135254,
288
+ "learning_rate": 0.00019746512868615656,
289
+ "loss": 3.5431480407714844,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 2.633315324783325,
295
+ "learning_rate": 0.00019713134135763256,
296
+ "loss": 3.784006881713867,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 2.0188381671905518,
302
+ "learning_rate": 0.00019677723364475237,
303
+ "loss": 3.7024520874023437,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 2.1343772411346436,
309
+ "learning_rate": 0.00019640287962870062,
310
+ "loss": 3.7561168670654297,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 1.9709585905075073,
316
+ "learning_rate": 0.0001960083576262943,
317
+ "loss": 3.6580154418945314,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 2.0804245471954346,
323
+ "learning_rate": 0.0001955937501735987,
324
+ "loss": 3.379087448120117,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 1.890079379081726,
330
+ "learning_rate": 0.0001951591440086602,
331
+ "loss": 3.5966243743896484,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 2.0973691940307617,
337
+ "learning_rate": 0.0001947046300533604,
338
+ "loss": 3.4583953857421874,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 2.4010658264160156,
344
+ "learning_rate": 0.00019423030339439464,
345
+ "loss": 3.3876201629638674,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 2.172189235687256,
351
+ "learning_rate": 0.00019373626326337946,
352
+ "loss": 3.7462398529052736,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 1.9334107637405396,
358
+ "learning_rate": 0.00019322261301609286,
359
+ "loss": 3.841419219970703,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 2.0680906772613525,
365
+ "learning_rate": 0.00019268946011085158,
366
+ "loss": 3.4502696990966797,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 2.5843758583068848,
372
+ "learning_rate": 0.00019213691608603047,
373
+ "loss": 3.404983139038086,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 1.888687252998352,
379
+ "learning_rate": 0.00019156509653672782,
380
+ "loss": 3.549253463745117,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 1.8593695163726807,
386
+ "learning_rate": 0.00019097412109058247,
387
+ "loss": 3.422663116455078,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 2.036330461502075,
393
+ "learning_rate": 0.00019036411338274703,
394
+ "loss": 3.7582332611083986,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 1.8483343124389648,
400
+ "learning_rate": 0.00018973520103002277,
401
+ "loss": 3.517898941040039,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 2.0232744216918945,
407
+ "learning_rate": 0.00018908751560416142,
408
+ "loss": 3.521917724609375,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 3.3193116188049316,
414
+ "learning_rate": 0.00018842119260433982,
415
+ "loss": 3.4880836486816404,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 2.5043585300445557,
421
+ "learning_rate": 0.0001877363714288126,
422
+ "loss": 3.446257400512695,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 1.9308786392211914,
428
+ "learning_rate": 0.00018703319534574956,
429
+ "loss": 3.3181568145751954,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 1.857397198677063,
435
+ "learning_rate": 0.00018631181146326305,
436
+ "loss": 3.219373321533203,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.9737612009048462,
442
+ "learning_rate": 0.00018557237069863222,
443
+ "loss": 3.5038528442382812,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 4.4995598793029785,
449
+ "learning_rate": 0.00018481502774673044,
450
+ "loss": 3.3514575958251953,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 2.948864221572876,
456
+ "learning_rate": 0.00018403994104766212,
457
+ "loss": 3.13072509765625,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 2.1394331455230713,
463
+ "learning_rate": 0.00018324727275361645,
464
+ "loss": 3.5682483673095704,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 3.813765287399292,
470
+ "learning_rate": 0.00018243718869494408,
471
+ "loss": 3.478915786743164,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.6351795196533203,
477
+ "learning_rate": 0.00018160985834546475,
478
+ "loss": 3.307242584228516,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 2.2016730308532715,
484
+ "learning_rate": 0.00018076545478701235,
485
+ "loss": 3.5482852935791014,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 2.1184046268463135,
491
+ "learning_rate": 0.00017990415467322534,
492
+ "loss": 3.51483039855957,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 1.704121470451355,
498
+ "learning_rate": 0.00017902613819258985,
499
+ "loss": 3.6264686584472656,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 1.926048994064331,
505
+ "learning_rate": 0.00017813158903074342,
506
+ "loss": 3.472669219970703,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 2.019219160079956,
512
+ "learning_rate": 0.00017722069433204687,
513
+ "loss": 3.2899856567382812,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 2.0369884967803955,
519
+ "learning_rate": 0.00017629364466043273,
520
+ "loss": 3.427472686767578,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 1.7189043760299683,
526
+ "learning_rate": 0.0001753506339595384,
527
+ "loss": 3.403704833984375,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 2.494591236114502,
533
+ "learning_rate": 0.00017439185951213202,
534
+ "loss": 3.5159603118896485,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 1.7678285837173462,
540
+ "learning_rate": 0.00017341752189883983,
541
+ "loss": 3.119024467468262,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 2.24629807472229,
547
+ "learning_rate": 0.0001724278249561838,
548
+ "loss": 3.525630569458008,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 1.8001821041107178,
554
+ "learning_rate": 0.0001714229757339379,
555
+ "loss": 3.541637420654297,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 2.8331096172332764,
561
+ "learning_rate": 0.0001704031844518121,
562
+ "loss": 3.3902435302734375,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 2.052628755569458,
568
+ "learning_rate": 0.00016936866445547353,
569
+ "loss": 3.218609619140625,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 2.2566065788269043,
575
+ "learning_rate": 0.00016831963217191334,
576
+ "loss": 3.5105533599853516,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 1.5952314138412476,
582
+ "learning_rate": 0.0001672563070641688,
583
+ "loss": 3.253592681884766,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 1.6297531127929688,
589
+ "learning_rate": 0.000166178911585411,
590
+ "loss": 3.176485061645508,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 1.6768420934677124,
596
+ "learning_rate": 0.00016508767113240598,
597
+ "loss": 3.4830841064453124,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 1.9300287961959839,
603
+ "learning_rate": 0.00016398281399836097,
604
+ "loss": 3.309653472900391,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 1.4318655729293823,
610
+ "learning_rate": 0.00016286457132516383,
611
+ "loss": 3.589862060546875,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 1.4944521188735962,
617
+ "learning_rate": 0.0001617331770550274,
618
+ "loss": 3.277871322631836,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.7870607376098633,
624
+ "learning_rate": 0.00016058886788154712,
625
+ "loss": 3.269728088378906,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 1.5668118000030518,
631
+ "learning_rate": 0.00015943188320018388,
632
+ "loss": 3.210224914550781,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 2.267331123352051,
638
+ "learning_rate": 0.00015826246505818112,
639
+ "loss": 3.3089523315429688,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 2.414400815963745,
645
+ "learning_rate": 0.0001570808581039271,
646
+ "loss": 3.1313955307006838,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 1.8084098100662231,
652
+ "learning_rate": 0.00015588730953577335,
653
+ "loss": 3.372232437133789,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 1.4967864751815796,
659
+ "learning_rate": 0.0001546820690503195,
660
+ "loss": 3.3817447662353515,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 1.7170981168746948,
666
+ "learning_rate": 0.0001534653887901754,
667
+ "loss": 3.1899662017822266,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 1.7449836730957031,
673
+ "learning_rate": 0.00015223752329121155,
674
+ "loss": 3.4185081481933595,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 2.0558719635009766,
680
+ "learning_rate": 0.00015099872942930887,
681
+ "loss": 3.4337844848632812,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 1.402202844619751,
687
+ "learning_rate": 0.0001497492663666189,
688
+ "loss": 3.000233268737793,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 1.575924038887024,
694
+ "learning_rate": 0.0001484893954973458,
695
+ "loss": 3.401926803588867,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.8477239608764648,
701
+ "learning_rate": 0.00014721938039306134,
702
+ "loss": 3.143557548522949,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 1.3538575172424316,
708
+ "learning_rate": 0.00014593948674756417,
709
+ "loss": 3.316628265380859,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 1.7879822254180908,
715
+ "learning_rate": 0.0001446499823212954,
716
+ "loss": 3.032166862487793,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.9181793928146362,
722
+ "learning_rate": 0.00014335113688532182,
723
+ "loss": 3.029685401916504,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 1.4723820686340332,
729
+ "learning_rate": 0.00014204322216489814,
730
+ "loss": 3.1593692779541014,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 1.5803518295288086,
736
+ "learning_rate": 0.00014072651178262096,
737
+ "loss": 3.1910980224609373,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 1.4509774446487427,
743
+ "learning_rate": 0.00013940128120118516,
744
+ "loss": 3.3486377716064455,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 1.2490512132644653,
750
+ "learning_rate": 0.00013806780766575588,
751
+ "loss": 3.127117919921875,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 1.3815803527832031,
757
+ "learning_rate": 0.0001367263701459674,
758
+ "loss": 3.3135940551757814,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 1.6695923805236816,
764
+ "learning_rate": 0.00013537724927756094,
765
+ "loss": 3.24918212890625,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 1.6700210571289062,
771
+ "learning_rate": 0.00013402072730367475,
772
+ "loss": 3.1998600006103515,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 1.458272099494934,
778
+ "learning_rate": 0.0001326570880157967,
779
+ "loss": 3.2691417694091798,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 1.6035007238388062,
785
+ "learning_rate": 0.00013128661669439416,
786
+ "loss": 3.191153717041016,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 1.3588290214538574,
792
+ "learning_rate": 0.00012990960004923154,
793
+ "loss": 3.177682113647461,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.3084040880203247,
799
+ "learning_rate": 0.0001285263261593893,
800
+ "loss": 3.0778690338134767,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 1.7576879262924194,
806
+ "learning_rate": 0.000127137084412996,
807
+ "loss": 3.063307189941406,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 1.61252760887146,
813
+ "learning_rate": 0.0001257421654466872,
814
+ "loss": 3.2951290130615236,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 1.5347871780395508,
820
+ "learning_rate": 0.0001243418610848024,
821
+ "loss": 3.1931894302368162,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 2.2056219577789307,
827
+ "learning_rate": 0.00012293646427833404,
828
+ "loss": 3.3431358337402344,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 1.6683447360992432,
834
+ "learning_rate": 0.00012152626904364067,
835
+ "loss": 3.057468795776367,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 1.455543875694275,
841
+ "learning_rate": 0.00012011157040093696,
842
+ "loss": 3.289105987548828,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 1.7347569465637207,
848
+ "learning_rate": 0.00011869266431257422,
849
+ "loss": 2.9250898361206055,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 1.3640249967575073,
855
+ "learning_rate": 0.00011726984762112328,
856
+ "loss": 3.325461578369141,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 1.4054839611053467,
862
+ "learning_rate": 0.00011584341798727365,
863
+ "loss": 3.168132019042969,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 1.2563236951828003,
869
+ "learning_rate": 0.0001144136738275612,
870
+ "loss": 3.237269973754883,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 2.893691062927246,
876
+ "learning_rate": 0.00011298091425193806,
877
+ "loss": 3.241581344604492,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 1.9952828884124756,
883
+ "learning_rate": 0.00011154543900119698,
884
+ "loss": 3.1403961181640625,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 1.573317050933838,
890
+ "learning_rate": 0.00011010754838426428,
891
+ "loss": 3.106332778930664,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 2.1935744285583496,
897
+ "learning_rate": 0.00010866754321537338,
898
+ "loss": 3.1031381607055666,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 1.5979467630386353,
904
+ "learning_rate": 0.00010722572475113315,
905
+ "loss": 3.1226659774780274,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 1.1873749494552612,
911
+ "learning_rate": 0.00010578239462750321,
912
+ "loss": 3.1140846252441405,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 1.8321475982666016,
918
+ "learning_rate": 0.00010433785479669038,
919
+ "loss": 3.0106977462768554,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 1.3288156986236572,
925
+ "learning_rate": 0.00010289240746397832,
926
+ "loss": 2.8980552673339846,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 1.416706919670105,
932
+ "learning_rate": 0.00010144635502450508,
933
+ "loss": 3.141121673583984,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 1.9086672067642212,
939
+ "learning_rate": 0.0001,
940
+ "loss": 3.282766342163086,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 1.4448398351669312,
946
+ "learning_rate": 9.855364497549496e-05,
947
+ "loss": 2.973546600341797,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 1.2777249813079834,
953
+ "learning_rate": 9.71075925360217e-05,
954
+ "loss": 3.145435905456543,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 1.276002287864685,
960
+ "learning_rate": 9.566214520330966e-05,
961
+ "loss": 3.2174652099609373,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 1.4613112211227417,
967
+ "learning_rate": 9.421760537249682e-05,
968
+ "loss": 3.17722110748291,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 1.1228781938552856,
974
+ "learning_rate": 9.277427524886689e-05,
975
+ "loss": 2.98958740234375,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 1.1255987882614136,
981
+ "learning_rate": 9.133245678462663e-05,
982
+ "loss": 3.1424510955810545,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.194167137145996,
988
+ "learning_rate": 8.989245161573576e-05,
989
+ "loss": 3.0195308685302735,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 1.864384412765503,
995
+ "learning_rate": 8.845456099880303e-05,
996
+ "loss": 2.997172737121582,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 1.241004467010498,
1002
+ "learning_rate": 8.701908574806197e-05,
1003
+ "loss": 2.9221866607666014,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 1.1507432460784912,
1009
+ "learning_rate": 8.55863261724388e-05,
1010
+ "loss": 2.9324596405029295,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.287448763847351,
1016
+ "learning_rate": 8.415658201272636e-05,
1017
+ "loss": 2.975849914550781,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 1.6067427396774292,
1023
+ "learning_rate": 8.273015237887673e-05,
1024
+ "loss": 2.980916213989258,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 1.4008774757385254,
1030
+ "learning_rate": 8.130733568742579e-05,
1031
+ "loss": 2.8944910049438475,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 1.2580070495605469,
1037
+ "learning_rate": 7.988842959906306e-05,
1038
+ "loss": 2.969981384277344,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 1.4962762594223022,
1044
+ "learning_rate": 7.847373095635937e-05,
1045
+ "loss": 3.1294715881347654,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 1.2535191774368286,
1051
+ "learning_rate": 7.706353572166597e-05,
1052
+ "loss": 3.066376876831055,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 0.8347238898277283,
1058
+ "learning_rate": 7.565813891519765e-05,
1059
+ "loss": 3.0157793045043944,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 1.2819799184799194,
1065
+ "learning_rate": 7.425783455331281e-05,
1066
+ "loss": 3.0389671325683594,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 1.1498677730560303,
1072
+ "learning_rate": 7.2862915587004e-05,
1073
+ "loss": 2.986423301696777,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 1.1965627670288086,
1079
+ "learning_rate": 7.147367384061074e-05,
1080
+ "loss": 2.941807746887207,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 1.8370661735534668,
1086
+ "learning_rate": 7.009039995076844e-05,
1087
+ "loss": 2.658160400390625,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.418321132659912,
1093
+ "learning_rate": 6.871338330560587e-05,
1094
+ "loss": 3.019854736328125,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 1.3217508792877197,
1100
+ "learning_rate": 6.734291198420333e-05,
1101
+ "loss": 2.8131311416625975,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.207661747932434,
1107
+ "learning_rate": 6.597927269632526e-05,
1108
+ "loss": 3.063412666320801,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 1.1648913621902466,
1114
+ "learning_rate": 6.462275072243908e-05,
1115
+ "loss": 3.0537391662597657,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 1.1196516752243042,
1121
+ "learning_rate": 6.327362985403265e-05,
1122
+ "loss": 3.0568498611450194,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 1.1875085830688477,
1128
+ "learning_rate": 6.193219233424414e-05,
1129
+ "loss": 2.8038040161132813,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 1.5340442657470703,
1135
+ "learning_rate": 6.059871879881487e-05,
1136
+ "loss": 2.922003746032715,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 1.4385885000228882,
1142
+ "learning_rate": 5.927348821737906e-05,
1143
+ "loss": 2.950858306884766,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 1.0835936069488525,
1149
+ "learning_rate": 5.795677783510187e-05,
1150
+ "loss": 2.898311996459961,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 1.4508082866668701,
1156
+ "learning_rate": 5.664886311467821e-05,
1157
+ "loss": 3.036098670959473,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 1.0718936920166016,
1163
+ "learning_rate": 5.535001767870463e-05,
1164
+ "loss": 2.8239120483398437,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 1.3246824741363525,
1170
+ "learning_rate": 5.406051325243586e-05,
1171
+ "loss": 2.9098350524902346,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 1.3559950590133667,
1177
+ "learning_rate": 5.278061960693865e-05,
1178
+ "loss": 2.7230411529541017,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 1.3093736171722412,
1184
+ "learning_rate": 5.151060450265419e-05,
1185
+ "loss": 2.998219871520996,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 1.1871122121810913,
1191
+ "learning_rate": 5.025073363338111e-05,
1192
+ "loss": 2.8021818161010743,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.5887531042099,
1198
+ "learning_rate": 4.900127057069116e-05,
1199
+ "loss": 3.138778305053711,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 1.0622819662094116,
1205
+ "learning_rate": 4.776247670878847e-05,
1206
+ "loss": 2.8708589553833006,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.0821974277496338,
1212
+ "learning_rate": 4.65346112098246e-05,
1213
+ "loss": 2.923636054992676,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 1.2138561010360718,
1219
+ "learning_rate": 4.5317930949680524e-05,
1220
+ "loss": 3.0319366455078125,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 1.3472957611083984,
1226
+ "learning_rate": 4.41126904642267e-05,
1227
+ "loss": 2.7567047119140624,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 1.985767126083374,
1233
+ "learning_rate": 4.291914189607297e-05,
1234
+ "loss": 3.056001663208008,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 1.1290528774261475,
1240
+ "learning_rate": 4.17375349418189e-05,
1241
+ "loss": 3.075904655456543,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 1.2921662330627441,
1247
+ "learning_rate": 4.056811679981609e-05,
1248
+ "loss": 2.8820091247558595,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 1.1284035444259644,
1254
+ "learning_rate": 3.9411132118452896e-05,
1255
+ "loss": 2.897430419921875,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1220782995224,
1261
+ "learning_rate": 3.826682294497265e-05,
1262
+ "loss": 2.782516860961914,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 1.0530436038970947,
1268
+ "learning_rate": 3.7135428674836184e-05,
1269
+ "loss": 2.7634408950805662,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 1.3937631845474243,
1275
+ "learning_rate": 3.6017186001639036e-05,
1276
+ "loss": 2.894866180419922,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 1.00092613697052,
1282
+ "learning_rate": 3.491232886759398e-05,
1283
+ "loss": 2.6879959106445312,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 1.1270594596862793,
1289
+ "learning_rate": 3.382108841458904e-05,
1290
+ "loss": 3.027115821838379,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 1.3118301630020142,
1296
+ "learning_rate": 3.274369293583121e-05,
1297
+ "loss": 2.8866352081298827,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 1.2168943881988525,
1303
+ "learning_rate": 3.1680367828086676e-05,
1304
+ "loss": 2.8968345642089846,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 1.0909639596939087,
1310
+ "learning_rate": 3.063133554452645e-05,
1311
+ "loss": 2.9398611068725584,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 1.1608167886734009,
1317
+ "learning_rate": 2.9596815548187908e-05,
1318
+ "loss": 2.842852783203125,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 1.511507511138916,
1324
+ "learning_rate": 2.8577024266062146e-05,
1325
+ "loss": 2.77429141998291,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 1.1318535804748535,
1331
+ "learning_rate": 2.7572175043816206e-05,
1332
+ "loss": 2.8206886291503905,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.9189174771308899,
1338
+ "learning_rate": 2.6582478101160167e-05,
1339
+ "loss": 2.9603801727294923,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 1.180462121963501,
1345
+ "learning_rate": 2.560814048786799e-05,
1346
+ "loss": 2.8037836074829103,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 1.1114805936813354,
1352
+ "learning_rate": 2.4649366040461597e-05,
1353
+ "loss": 2.941350746154785,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 1.31864595413208,
1359
+ "learning_rate": 2.3706355339567286e-05,
1360
+ "loss": 3.0531469345092774,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.9089719653129578,
1366
+ "learning_rate": 2.2779305667953154e-05,
1367
+ "loss": 2.746686553955078,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 1.0223416090011597,
1373
+ "learning_rate": 2.186841096925657e-05,
1374
+ "loss": 2.8307025909423826,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 1.27829909324646,
1380
+ "learning_rate": 2.097386180741019e-05,
1381
+ "loss": 2.928578758239746,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 1.3367505073547363,
1387
+ "learning_rate": 2.0095845326774697e-05,
1388
+ "loss": 2.8606313705444335,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 1.1823525428771973,
1394
+ "learning_rate": 1.9234545212987688e-05,
1395
+ "loss": 3.041524124145508,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 1.2242523431777954,
1401
+ "learning_rate": 1.8390141654535265e-05,
1402
+ "loss": 3.262685775756836,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.9626563191413879,
1408
+ "learning_rate": 1.756281130505595e-05,
1409
+ "loss": 2.824736976623535,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.9038745760917664,
1415
+ "learning_rate": 1.6752727246383582e-05,
1416
+ "loss": 3.0037527084350586,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 1.0726526975631714,
1422
+ "learning_rate": 1.5960058952337887e-05,
1423
+ "loss": 2.818869781494141,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 1.0472805500030518,
1429
+ "learning_rate": 1.518497225326958e-05,
1430
+ "loss": 2.9350055694580077,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 1.1116654872894287,
1436
+ "learning_rate": 1.4427629301367773e-05,
1437
+ "loss": 2.903761291503906,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.9501079320907593,
1443
+ "learning_rate": 1.3688188536736968e-05,
1444
+ "loss": 2.784720611572266,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 1.109456181526184,
1450
+ "learning_rate": 1.2966804654250465e-05,
1451
+ "loss": 3.1309579849243163,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 1.1933435201644897,
1457
+ "learning_rate": 1.2263628571187414e-05,
1458
+ "loss": 3.0474115371704102,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 1.4391895532608032,
1464
+ "learning_rate": 1.1578807395660207e-05,
1465
+ "loss": 3.0081947326660154,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.9554082155227661,
1471
+ "learning_rate": 1.0912484395838585e-05,
1472
+ "loss": 2.9179988861083985,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 1.1937369108200073,
1478
+ "learning_rate": 1.026479896997723e-05,
1479
+ "loss": 2.705255126953125,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.9862006306648254,
1485
+ "learning_rate": 9.635886617252975e-06,
1486
+ "loss": 2.745578384399414,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 1.1251375675201416,
1492
+ "learning_rate": 9.025878909417552e-06,
1493
+ "loss": 2.965798187255859,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 1.0850991010665894,
1499
+ "learning_rate": 8.434903463272215e-06,
1500
+ "loss": 2.9574241638183594,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 1.476346731185913,
1506
+ "learning_rate": 7.86308391396956e-06,
1507
+ "loss": 2.7245121002197266,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 1.181896686553955,
1513
+ "learning_rate": 7.310539889148404e-06,
1514
+ "loss": 3.0273662567138673,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.4082521200180054,
1520
+ "learning_rate": 6.777386983907152e-06,
1521
+ "loss": 2.9661121368408203,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 1.110843539237976,
1527
+ "learning_rate": 6.263736736620551e-06,
1528
+ "loss": 2.999568557739258,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 1.1253182888031006,
1534
+ "learning_rate": 5.769696605605379e-06,
1535
+ "loss": 2.998799705505371,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 1.089568853378296,
1541
+ "learning_rate": 5.295369946639628e-06,
1542
+ "loss": 2.6438697814941405,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 1.1276156902313232,
1548
+ "learning_rate": 4.840855991339799e-06,
1549
+ "loss": 2.9343143463134767,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 1.0737465620040894,
1555
+ "learning_rate": 4.406249826401309e-06,
1556
+ "loss": 3.080368995666504,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.8834953904151917,
1562
+ "learning_rate": 3.991642373705695e-06,
1563
+ "loss": 2.917354965209961,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 1.0340644121170044,
1569
+ "learning_rate": 3.5971203712993894e-06,
1570
+ "loss": 2.9122617721557615,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 1.1595566272735596,
1576
+ "learning_rate": 3.2227663552476194e-06,
1577
+ "loss": 2.817066764831543,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 1.0316624641418457,
1583
+ "learning_rate": 2.868658642367439e-06,
1584
+ "loss": 2.764511489868164,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 1.1504026651382446,
1590
+ "learning_rate": 2.5348713138434564e-06,
1591
+ "loss": 2.9506311416625977,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 1.0357000827789307,
1597
+ "learning_rate": 2.2214741997296117e-06,
1598
+ "loss": 2.817026901245117,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.8134933710098267,
1604
+ "learning_rate": 1.928532864340438e-06,
1605
+ "loss": 2.7651405334472656,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 1.3359462022781372,
1611
+ "learning_rate": 1.6561085925346332e-06,
1612
+ "loss": 2.8869171142578125,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 1.4075947999954224,
1618
+ "learning_rate": 1.4042583768939298e-06,
1619
+ "loss": 3.0473421096801756,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.7921971082687378,
1625
+ "learning_rate": 1.1730349057999412e-06,
1626
+ "loss": 2.8001834869384767,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 1.0570473670959473,
1632
+ "learning_rate": 9.624865524115346e-07,
1633
+ "loss": 2.9640541076660156,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 1.0891289710998535,
1639
+ "learning_rate": 7.726573645448487e-07,
1640
+ "loss": 3.0531845092773438,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 1.0408469438552856,
1646
+ "learning_rate": 6.03587055458299e-07,
1647
+ "loss": 2.9867210388183594,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.054602861404419,
1653
+ "learning_rate": 4.5531099554435576e-07,
1654
+ "loss": 2.8087078094482423,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 1.1407268047332764,
1660
+ "learning_rate": 3.278602049298418e-07,
1661
+ "loss": 3.1718997955322266,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.9369794130325317,
1667
+ "learning_rate": 2.2126134698637892e-07,
1668
+ "loss": 2.9225208282470705,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 1.6008250713348389,
1674
+ "learning_rate": 1.3553672275230523e-07,
1675
+ "loss": 3.1131832122802736,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.9389436841011047,
1681
+ "learning_rate": 7.070426626711868e-08,
1682
+ "loss": 2.8224658966064453,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.8056095242500305,
1688
+ "learning_rate": 2.677754081961137e-08,
1689
+ "loss": 2.772154998779297,
1690
+ "step": 1200
1691
+ }
1692
+ ],
1693
+ "logging_steps": 5,
1694
+ "max_steps": 1207,
1695
+ "num_input_tokens_seen": 0,
1696
+ "num_train_epochs": 1,
1697
+ "save_steps": 200,
1698
+ "stateful_callbacks": {
1699
+ "TrainerControl": {
1700
+ "args": {
1701
+ "should_epoch_stop": false,
1702
+ "should_evaluate": false,
1703
+ "should_log": false,
1704
+ "should_save": true,
1705
+ "should_training_stop": false
1706
+ },
1707
+ "attributes": {}
1708
+ }
1709
+ },
1710
+ "total_flos": 2.518116653061744e+17,
1711
+ "train_batch_size": 1,
1712
+ "trial_name": null,
1713
+ "trial_params": null
1714
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- macro render_content(content) -%}
2
+ {%- if content is string -%}
3
+ {{- content -}}
4
+ {%- else -%}
5
+ {%- for item in content -%}
6
+ {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
7
+ {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
8
+ {%- elif item.type == 'text' -%}
9
+ {{- item.text -}}
10
+ {%- endif -%}
11
+ {%- endfor -%}
12
+ {%- endif -%}
13
+ {%- endmacro -%}
14
+ {%- if tools %}
15
+ {{- '<|im_start|>system\n' }}
16
+ {%- if messages[0].role == 'system' %}
17
+ {{- render_content(messages[0].content) + '\n\n' }}
18
+ {%- else %}
19
+ {{- 'You are a helpful assistant.\n\n' }}
20
+ {%- endif %}
21
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
22
+ {%- for tool in tools %}
23
+ {{- "\n" }}
24
+ {{- tool | tojson }}
25
+ {%- endfor %}
26
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
27
+ {%- else %}
28
+ {%- if messages[0].role == 'system' %}
29
+ {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
30
+ {%- else %}
31
+ {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
32
+ {%- endif %}
33
+ {%- endif %}
34
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
35
+ {%- for message in messages[::-1] %}
36
+ {%- set index = (messages|length - 1) - loop.index0 %}
37
+ {%- set content = render_content(message.content) %}
38
+ {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
39
+ {%- set ns.multi_step_tool = false %}
40
+ {%- set ns.last_query_index = index %}
41
+ {%- endif %}
42
+ {%- endfor %}
43
+ {%- for message in messages %}
44
+ {%- set content = render_content(message.content) %}
45
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
47
+ {%- elif message.role == "assistant" %}
48
+ {%- set reasoning_content = '' %}
49
+ {%- if message.reasoning_content is string %}
50
+ {%- set reasoning_content = message.reasoning_content %}
51
+ {%- else %}
52
+ {%- if '</think>' in content %}
53
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
54
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
55
+ {%- endif %}
56
+ {%- endif %}
57
+ {%- if loop.index0 > ns.last_query_index %}
58
+ {%- if loop.last or (not loop.last and reasoning_content) %}
59
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
60
+ {%- else %}
61
+ {{- '<|im_start|>' + message.role + '\n' + content }}
62
+ {%- endif %}
63
+ {%- else %}
64
+ {{- '<|im_start|>' + message.role + '\n' + content }}
65
+ {%- endif %}
66
+ {%- if message.tool_calls %}
67
+ {%- for tool_call in message.tool_calls %}
68
+ {%- if (loop.first and content) or (not loop.first) %}
69
+ {{- '\n' }}
70
+ {%- endif %}
71
+ {%- if tool_call.function %}
72
+ {%- set tool_call = tool_call.function %}
73
+ {%- endif %}
74
+ {{- '<tool_call>\n{"name": "' }}
75
+ {{- tool_call.name }}
76
+ {{- '", "arguments": ' }}
77
+ {%- if tool_call.arguments is string %}
78
+ {{- tool_call.arguments }}
79
+ {%- else %}
80
+ {{- tool_call.arguments | tojson }}
81
+ {%- endif %}
82
+ {{- '}\n</tool_call>' }}
83
+ {%- endfor %}
84
+ {%- endif %}
85
+ {{- '<|im_end|>\n' }}
86
+ {%- elif message.role == "tool" %}
87
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
88
+ {{- '<|im_start|>user' }}
89
+ {%- endif %}
90
+ {{- '\n<tool_response>\n' }}
91
+ {{- content }}
92
+ {{- '\n</tool_response>' }}
93
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
94
+ {{- '<|im_end|>\n' }}
95
+ {%- endif %}
96
+ {%- endif %}
97
+ {%- endfor %}
98
+ {%- if add_generation_prompt %}
99
+ {{- '<|im_start|>assistant\n' }}
100
+ {%- if enable_thinking is defined and enable_thinking is false %}
101
+ {{- '<think>\n\n</think>\n\n' }}
102
+ {%- endif %}
103
+ {%- endif %}
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/config.json ADDED
@@ -0,0 +1,250 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_rotary_dim": 64,
171
+ "speaker_rotary_inference_seed": 42,
172
+ "speaker_rotary_std": 1.0,
173
+ "speaker_sample_rate": 16000,
174
+ "speaker_tfmel_args": {
175
+ "do_preemph": true,
176
+ "do_spec_aug": false,
177
+ "eps": 1e-08,
178
+ "f_max": 7600,
179
+ "f_min": 20,
180
+ "hop_length": 160,
181
+ "n_fft": 512,
182
+ "n_mels": 72,
183
+ "norm_signal": true,
184
+ "win_length": 400
185
+ },
186
+ "speaker_window_seconds": 2.0,
187
+ "text_config": {
188
+ "attention_bias": false,
189
+ "attention_dropout": 0.0,
190
+ "bos_token_id": null,
191
+ "dtype": "bfloat16",
192
+ "eos_token_id": null,
193
+ "head_dim": 128,
194
+ "hidden_act": "silu",
195
+ "hidden_size": 1024,
196
+ "initializer_range": 0.02,
197
+ "intermediate_size": 3072,
198
+ "layer_types": [
199
+ "full_attention",
200
+ "full_attention",
201
+ "full_attention",
202
+ "full_attention",
203
+ "full_attention",
204
+ "full_attention",
205
+ "full_attention",
206
+ "full_attention",
207
+ "full_attention",
208
+ "full_attention",
209
+ "full_attention",
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention"
227
+ ],
228
+ "max_position_embeddings": 131072,
229
+ "max_window_layers": 28,
230
+ "model_type": "qwen3",
231
+ "num_attention_heads": 16,
232
+ "num_hidden_layers": 28,
233
+ "num_key_value_heads": 8,
234
+ "pad_token_id": 151643,
235
+ "partial_rotary_factor": 0.5,
236
+ "rms_norm_eps": 1e-06,
237
+ "rope_parameters": {
238
+ "rope_theta": 1000000,
239
+ "rope_type": "default"
240
+ },
241
+ "sliding_window": null,
242
+ "tie_word_embeddings": true,
243
+ "use_cache": false,
244
+ "use_sliding_window": false,
245
+ "vocab_size": 151936
246
+ },
247
+ "tie_word_embeddings": true,
248
+ "transformers_version": "5.14.0",
249
+ "use_cache": false
250
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/checkpoint-1207/trainer_state.json ADDED
@@ -0,0 +1,1721 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1207,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 1147.55224609375,
15
+ "learning_rate": 6.611570247933885e-06,
16
+ "loss": 11.37186279296875,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 246.35504150390625,
22
+ "learning_rate": 1.487603305785124e-05,
23
+ "loss": 9.2022216796875,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 79.90882110595703,
29
+ "learning_rate": 2.3140495867768598e-05,
30
+ "loss": 7.6917167663574215,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 158.79872131347656,
36
+ "learning_rate": 3.1404958677685955e-05,
37
+ "loss": 6.138848114013672,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 48.520408630371094,
43
+ "learning_rate": 3.9669421487603306e-05,
44
+ "loss": 5.638393402099609,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 14.776691436767578,
50
+ "learning_rate": 4.793388429752066e-05,
51
+ "loss": 5.221096801757812,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 12.125936508178711,
57
+ "learning_rate": 5.619834710743802e-05,
58
+ "loss": 4.968892288208008,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 14.540782928466797,
64
+ "learning_rate": 6.446280991735537e-05,
65
+ "loss": 5.3284343719482425,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 13.903443336486816,
71
+ "learning_rate": 7.272727272727273e-05,
72
+ "loss": 5.018967819213867,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 12.485719680786133,
78
+ "learning_rate": 8.099173553719009e-05,
79
+ "loss": 4.600848770141601,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 44.14723205566406,
85
+ "learning_rate": 8.925619834710744e-05,
86
+ "loss": 4.335272979736328,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 9.01069164276123,
92
+ "learning_rate": 9.75206611570248e-05,
93
+ "loss": 4.316203308105469,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 16.169389724731445,
99
+ "learning_rate": 0.00010578512396694216,
100
+ "loss": 4.197259140014649,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 6.058357238769531,
106
+ "learning_rate": 0.0001140495867768595,
107
+ "loss": 4.562212371826172,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 5.126786231994629,
113
+ "learning_rate": 0.00012231404958677685,
114
+ "loss": 4.340535354614258,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 12.424266815185547,
120
+ "learning_rate": 0.00013057851239669423,
121
+ "loss": 3.9918304443359376,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 8.042577743530273,
127
+ "learning_rate": 0.0001388429752066116,
128
+ "loss": 4.087551879882812,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 8.642739295959473,
134
+ "learning_rate": 0.00014710743801652894,
135
+ "loss": 4.22486686706543,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 8.129426002502441,
141
+ "learning_rate": 0.00015537190082644627,
142
+ "loss": 3.8094669342041017,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 10.393760681152344,
148
+ "learning_rate": 0.00016363636363636366,
149
+ "loss": 3.687621307373047,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 23.797346115112305,
155
+ "learning_rate": 0.00017190082644628102,
156
+ "loss": 4.8999065399169925,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 7.251102924346924,
162
+ "learning_rate": 0.00018016528925619835,
163
+ "loss": 4.111215972900391,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 7.632305145263672,
169
+ "learning_rate": 0.0001884297520661157,
170
+ "loss": 3.88475341796875,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 2.708984851837158,
176
+ "learning_rate": 0.0001966942148760331,
177
+ "loss": 3.9735763549804686,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 3.5484111309051514,
183
+ "learning_rate": 0.00019999623426388962,
184
+ "loss": 3.911006546020508,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 3.580739974975586,
190
+ "learning_rate": 0.0001999732224591804,
191
+ "loss": 3.6748367309570313,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 3.1602091789245605,
197
+ "learning_rate": 0.0001999292957337329,
198
+ "loss": 4.212306213378906,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 2.242072582244873,
204
+ "learning_rate": 0.0001998644632772477,
205
+ "loss": 3.8557418823242187,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 3.2909867763519287,
211
+ "learning_rate": 0.00019977873865301363,
212
+ "loss": 3.953845977783203,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 2.695601224899292,
218
+ "learning_rate": 0.00019967213979507018,
219
+ "loss": 3.7311931610107423,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 3.831533670425415,
225
+ "learning_rate": 0.00019954468900445566,
226
+ "loss": 3.88598518371582,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 4.886938571929932,
232
+ "learning_rate": 0.00019939641294454172,
233
+ "loss": 3.8442028045654295,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 3.451791524887085,
239
+ "learning_rate": 0.00019922734263545517,
240
+ "loss": 3.9103469848632812,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 3.0034420490264893,
246
+ "learning_rate": 0.00019903751344758848,
247
+ "loss": 4.032486343383789,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 2.849339723587036,
253
+ "learning_rate": 0.00019882696509420007,
254
+ "loss": 3.407342529296875,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 2.095475435256958,
260
+ "learning_rate": 0.00019859574162310608,
261
+ "loss": 3.7116188049316405,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 3.148158073425293,
267
+ "learning_rate": 0.0001983438914074654,
268
+ "loss": 3.7624229431152343,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 2.4163601398468018,
274
+ "learning_rate": 0.00019807146713565955,
275
+ "loss": 3.439311218261719,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 2.7623701095581055,
281
+ "learning_rate": 0.00019777852580027042,
282
+ "loss": 3.8211971282958985,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 3.145005226135254,
288
+ "learning_rate": 0.00019746512868615656,
289
+ "loss": 3.5431480407714844,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 2.633315324783325,
295
+ "learning_rate": 0.00019713134135763256,
296
+ "loss": 3.784006881713867,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 2.0188381671905518,
302
+ "learning_rate": 0.00019677723364475237,
303
+ "loss": 3.7024520874023437,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 2.1343772411346436,
309
+ "learning_rate": 0.00019640287962870062,
310
+ "loss": 3.7561168670654297,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 1.9709585905075073,
316
+ "learning_rate": 0.0001960083576262943,
317
+ "loss": 3.6580154418945314,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 2.0804245471954346,
323
+ "learning_rate": 0.0001955937501735987,
324
+ "loss": 3.379087448120117,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 1.890079379081726,
330
+ "learning_rate": 0.0001951591440086602,
331
+ "loss": 3.5966243743896484,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 2.0973691940307617,
337
+ "learning_rate": 0.0001947046300533604,
338
+ "loss": 3.4583953857421874,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 2.4010658264160156,
344
+ "learning_rate": 0.00019423030339439464,
345
+ "loss": 3.3876201629638674,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 2.172189235687256,
351
+ "learning_rate": 0.00019373626326337946,
352
+ "loss": 3.7462398529052736,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 1.9334107637405396,
358
+ "learning_rate": 0.00019322261301609286,
359
+ "loss": 3.841419219970703,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 2.0680906772613525,
365
+ "learning_rate": 0.00019268946011085158,
366
+ "loss": 3.4502696990966797,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 2.5843758583068848,
372
+ "learning_rate": 0.00019213691608603047,
373
+ "loss": 3.404983139038086,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 1.888687252998352,
379
+ "learning_rate": 0.00019156509653672782,
380
+ "loss": 3.549253463745117,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 1.8593695163726807,
386
+ "learning_rate": 0.00019097412109058247,
387
+ "loss": 3.422663116455078,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 2.036330461502075,
393
+ "learning_rate": 0.00019036411338274703,
394
+ "loss": 3.7582332611083986,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 1.8483343124389648,
400
+ "learning_rate": 0.00018973520103002277,
401
+ "loss": 3.517898941040039,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 2.0232744216918945,
407
+ "learning_rate": 0.00018908751560416142,
408
+ "loss": 3.521917724609375,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 3.3193116188049316,
414
+ "learning_rate": 0.00018842119260433982,
415
+ "loss": 3.4880836486816404,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 2.5043585300445557,
421
+ "learning_rate": 0.0001877363714288126,
422
+ "loss": 3.446257400512695,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 1.9308786392211914,
428
+ "learning_rate": 0.00018703319534574956,
429
+ "loss": 3.3181568145751954,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 1.857397198677063,
435
+ "learning_rate": 0.00018631181146326305,
436
+ "loss": 3.219373321533203,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.9737612009048462,
442
+ "learning_rate": 0.00018557237069863222,
443
+ "loss": 3.5038528442382812,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 4.4995598793029785,
449
+ "learning_rate": 0.00018481502774673044,
450
+ "loss": 3.3514575958251953,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 2.948864221572876,
456
+ "learning_rate": 0.00018403994104766212,
457
+ "loss": 3.13072509765625,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 2.1394331455230713,
463
+ "learning_rate": 0.00018324727275361645,
464
+ "loss": 3.5682483673095704,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 3.813765287399292,
470
+ "learning_rate": 0.00018243718869494408,
471
+ "loss": 3.478915786743164,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.6351795196533203,
477
+ "learning_rate": 0.00018160985834546475,
478
+ "loss": 3.307242584228516,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 2.2016730308532715,
484
+ "learning_rate": 0.00018076545478701235,
485
+ "loss": 3.5482852935791014,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 2.1184046268463135,
491
+ "learning_rate": 0.00017990415467322534,
492
+ "loss": 3.51483039855957,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 1.704121470451355,
498
+ "learning_rate": 0.00017902613819258985,
499
+ "loss": 3.6264686584472656,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 1.926048994064331,
505
+ "learning_rate": 0.00017813158903074342,
506
+ "loss": 3.472669219970703,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 2.019219160079956,
512
+ "learning_rate": 0.00017722069433204687,
513
+ "loss": 3.2899856567382812,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 2.0369884967803955,
519
+ "learning_rate": 0.00017629364466043273,
520
+ "loss": 3.427472686767578,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 1.7189043760299683,
526
+ "learning_rate": 0.0001753506339595384,
527
+ "loss": 3.403704833984375,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 2.494591236114502,
533
+ "learning_rate": 0.00017439185951213202,
534
+ "loss": 3.5159603118896485,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 1.7678285837173462,
540
+ "learning_rate": 0.00017341752189883983,
541
+ "loss": 3.119024467468262,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 2.24629807472229,
547
+ "learning_rate": 0.0001724278249561838,
548
+ "loss": 3.525630569458008,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 1.8001821041107178,
554
+ "learning_rate": 0.0001714229757339379,
555
+ "loss": 3.541637420654297,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 2.8331096172332764,
561
+ "learning_rate": 0.0001704031844518121,
562
+ "loss": 3.3902435302734375,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 2.052628755569458,
568
+ "learning_rate": 0.00016936866445547353,
569
+ "loss": 3.218609619140625,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 2.2566065788269043,
575
+ "learning_rate": 0.00016831963217191334,
576
+ "loss": 3.5105533599853516,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 1.5952314138412476,
582
+ "learning_rate": 0.0001672563070641688,
583
+ "loss": 3.253592681884766,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 1.6297531127929688,
589
+ "learning_rate": 0.000166178911585411,
590
+ "loss": 3.176485061645508,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 1.6768420934677124,
596
+ "learning_rate": 0.00016508767113240598,
597
+ "loss": 3.4830841064453124,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 1.9300287961959839,
603
+ "learning_rate": 0.00016398281399836097,
604
+ "loss": 3.309653472900391,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 1.4318655729293823,
610
+ "learning_rate": 0.00016286457132516383,
611
+ "loss": 3.589862060546875,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 1.4944521188735962,
617
+ "learning_rate": 0.0001617331770550274,
618
+ "loss": 3.277871322631836,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.7870607376098633,
624
+ "learning_rate": 0.00016058886788154712,
625
+ "loss": 3.269728088378906,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 1.5668118000030518,
631
+ "learning_rate": 0.00015943188320018388,
632
+ "loss": 3.210224914550781,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 2.267331123352051,
638
+ "learning_rate": 0.00015826246505818112,
639
+ "loss": 3.3089523315429688,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 2.414400815963745,
645
+ "learning_rate": 0.0001570808581039271,
646
+ "loss": 3.1313955307006838,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 1.8084098100662231,
652
+ "learning_rate": 0.00015588730953577335,
653
+ "loss": 3.372232437133789,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 1.4967864751815796,
659
+ "learning_rate": 0.0001546820690503195,
660
+ "loss": 3.3817447662353515,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 1.7170981168746948,
666
+ "learning_rate": 0.0001534653887901754,
667
+ "loss": 3.1899662017822266,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 1.7449836730957031,
673
+ "learning_rate": 0.00015223752329121155,
674
+ "loss": 3.4185081481933595,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 2.0558719635009766,
680
+ "learning_rate": 0.00015099872942930887,
681
+ "loss": 3.4337844848632812,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 1.402202844619751,
687
+ "learning_rate": 0.0001497492663666189,
688
+ "loss": 3.000233268737793,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 1.575924038887024,
694
+ "learning_rate": 0.0001484893954973458,
695
+ "loss": 3.401926803588867,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.8477239608764648,
701
+ "learning_rate": 0.00014721938039306134,
702
+ "loss": 3.143557548522949,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 1.3538575172424316,
708
+ "learning_rate": 0.00014593948674756417,
709
+ "loss": 3.316628265380859,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 1.7879822254180908,
715
+ "learning_rate": 0.0001446499823212954,
716
+ "loss": 3.032166862487793,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.9181793928146362,
722
+ "learning_rate": 0.00014335113688532182,
723
+ "loss": 3.029685401916504,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 1.4723820686340332,
729
+ "learning_rate": 0.00014204322216489814,
730
+ "loss": 3.1593692779541014,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 1.5803518295288086,
736
+ "learning_rate": 0.00014072651178262096,
737
+ "loss": 3.1910980224609373,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 1.4509774446487427,
743
+ "learning_rate": 0.00013940128120118516,
744
+ "loss": 3.3486377716064455,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 1.2490512132644653,
750
+ "learning_rate": 0.00013806780766575588,
751
+ "loss": 3.127117919921875,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 1.3815803527832031,
757
+ "learning_rate": 0.0001367263701459674,
758
+ "loss": 3.3135940551757814,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 1.6695923805236816,
764
+ "learning_rate": 0.00013537724927756094,
765
+ "loss": 3.24918212890625,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 1.6700210571289062,
771
+ "learning_rate": 0.00013402072730367475,
772
+ "loss": 3.1998600006103515,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 1.458272099494934,
778
+ "learning_rate": 0.0001326570880157967,
779
+ "loss": 3.2691417694091798,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 1.6035007238388062,
785
+ "learning_rate": 0.00013128661669439416,
786
+ "loss": 3.191153717041016,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 1.3588290214538574,
792
+ "learning_rate": 0.00012990960004923154,
793
+ "loss": 3.177682113647461,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.3084040880203247,
799
+ "learning_rate": 0.0001285263261593893,
800
+ "loss": 3.0778690338134767,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 1.7576879262924194,
806
+ "learning_rate": 0.000127137084412996,
807
+ "loss": 3.063307189941406,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 1.61252760887146,
813
+ "learning_rate": 0.0001257421654466872,
814
+ "loss": 3.2951290130615236,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 1.5347871780395508,
820
+ "learning_rate": 0.0001243418610848024,
821
+ "loss": 3.1931894302368162,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 2.2056219577789307,
827
+ "learning_rate": 0.00012293646427833404,
828
+ "loss": 3.3431358337402344,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 1.6683447360992432,
834
+ "learning_rate": 0.00012152626904364067,
835
+ "loss": 3.057468795776367,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 1.455543875694275,
841
+ "learning_rate": 0.00012011157040093696,
842
+ "loss": 3.289105987548828,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 1.7347569465637207,
848
+ "learning_rate": 0.00011869266431257422,
849
+ "loss": 2.9250898361206055,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 1.3640249967575073,
855
+ "learning_rate": 0.00011726984762112328,
856
+ "loss": 3.325461578369141,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 1.4054839611053467,
862
+ "learning_rate": 0.00011584341798727365,
863
+ "loss": 3.168132019042969,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 1.2563236951828003,
869
+ "learning_rate": 0.0001144136738275612,
870
+ "loss": 3.237269973754883,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 2.893691062927246,
876
+ "learning_rate": 0.00011298091425193806,
877
+ "loss": 3.241581344604492,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 1.9952828884124756,
883
+ "learning_rate": 0.00011154543900119698,
884
+ "loss": 3.1403961181640625,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 1.573317050933838,
890
+ "learning_rate": 0.00011010754838426428,
891
+ "loss": 3.106332778930664,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 2.1935744285583496,
897
+ "learning_rate": 0.00010866754321537338,
898
+ "loss": 3.1031381607055666,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 1.5979467630386353,
904
+ "learning_rate": 0.00010722572475113315,
905
+ "loss": 3.1226659774780274,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 1.1873749494552612,
911
+ "learning_rate": 0.00010578239462750321,
912
+ "loss": 3.1140846252441405,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 1.8321475982666016,
918
+ "learning_rate": 0.00010433785479669038,
919
+ "loss": 3.0106977462768554,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 1.3288156986236572,
925
+ "learning_rate": 0.00010289240746397832,
926
+ "loss": 2.8980552673339846,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 1.416706919670105,
932
+ "learning_rate": 0.00010144635502450508,
933
+ "loss": 3.141121673583984,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 1.9086672067642212,
939
+ "learning_rate": 0.0001,
940
+ "loss": 3.282766342163086,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 1.4448398351669312,
946
+ "learning_rate": 9.855364497549496e-05,
947
+ "loss": 2.973546600341797,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 1.2777249813079834,
953
+ "learning_rate": 9.71075925360217e-05,
954
+ "loss": 3.145435905456543,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 1.276002287864685,
960
+ "learning_rate": 9.566214520330966e-05,
961
+ "loss": 3.2174652099609373,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 1.4613112211227417,
967
+ "learning_rate": 9.421760537249682e-05,
968
+ "loss": 3.17722110748291,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 1.1228781938552856,
974
+ "learning_rate": 9.277427524886689e-05,
975
+ "loss": 2.98958740234375,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 1.1255987882614136,
981
+ "learning_rate": 9.133245678462663e-05,
982
+ "loss": 3.1424510955810545,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.194167137145996,
988
+ "learning_rate": 8.989245161573576e-05,
989
+ "loss": 3.0195308685302735,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 1.864384412765503,
995
+ "learning_rate": 8.845456099880303e-05,
996
+ "loss": 2.997172737121582,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 1.241004467010498,
1002
+ "learning_rate": 8.701908574806197e-05,
1003
+ "loss": 2.9221866607666014,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 1.1507432460784912,
1009
+ "learning_rate": 8.55863261724388e-05,
1010
+ "loss": 2.9324596405029295,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.287448763847351,
1016
+ "learning_rate": 8.415658201272636e-05,
1017
+ "loss": 2.975849914550781,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 1.6067427396774292,
1023
+ "learning_rate": 8.273015237887673e-05,
1024
+ "loss": 2.980916213989258,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 1.4008774757385254,
1030
+ "learning_rate": 8.130733568742579e-05,
1031
+ "loss": 2.8944910049438475,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 1.2580070495605469,
1037
+ "learning_rate": 7.988842959906306e-05,
1038
+ "loss": 2.969981384277344,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 1.4962762594223022,
1044
+ "learning_rate": 7.847373095635937e-05,
1045
+ "loss": 3.1294715881347654,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 1.2535191774368286,
1051
+ "learning_rate": 7.706353572166597e-05,
1052
+ "loss": 3.066376876831055,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 0.8347238898277283,
1058
+ "learning_rate": 7.565813891519765e-05,
1059
+ "loss": 3.0157793045043944,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 1.2819799184799194,
1065
+ "learning_rate": 7.425783455331281e-05,
1066
+ "loss": 3.0389671325683594,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 1.1498677730560303,
1072
+ "learning_rate": 7.2862915587004e-05,
1073
+ "loss": 2.986423301696777,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 1.1965627670288086,
1079
+ "learning_rate": 7.147367384061074e-05,
1080
+ "loss": 2.941807746887207,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 1.8370661735534668,
1086
+ "learning_rate": 7.009039995076844e-05,
1087
+ "loss": 2.658160400390625,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.418321132659912,
1093
+ "learning_rate": 6.871338330560587e-05,
1094
+ "loss": 3.019854736328125,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 1.3217508792877197,
1100
+ "learning_rate": 6.734291198420333e-05,
1101
+ "loss": 2.8131311416625975,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.207661747932434,
1107
+ "learning_rate": 6.597927269632526e-05,
1108
+ "loss": 3.063412666320801,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 1.1648913621902466,
1114
+ "learning_rate": 6.462275072243908e-05,
1115
+ "loss": 3.0537391662597657,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 1.1196516752243042,
1121
+ "learning_rate": 6.327362985403265e-05,
1122
+ "loss": 3.0568498611450194,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 1.1875085830688477,
1128
+ "learning_rate": 6.193219233424414e-05,
1129
+ "loss": 2.8038040161132813,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 1.5340442657470703,
1135
+ "learning_rate": 6.059871879881487e-05,
1136
+ "loss": 2.922003746032715,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 1.4385885000228882,
1142
+ "learning_rate": 5.927348821737906e-05,
1143
+ "loss": 2.950858306884766,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 1.0835936069488525,
1149
+ "learning_rate": 5.795677783510187e-05,
1150
+ "loss": 2.898311996459961,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 1.4508082866668701,
1156
+ "learning_rate": 5.664886311467821e-05,
1157
+ "loss": 3.036098670959473,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 1.0718936920166016,
1163
+ "learning_rate": 5.535001767870463e-05,
1164
+ "loss": 2.8239120483398437,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 1.3246824741363525,
1170
+ "learning_rate": 5.406051325243586e-05,
1171
+ "loss": 2.9098350524902346,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 1.3559950590133667,
1177
+ "learning_rate": 5.278061960693865e-05,
1178
+ "loss": 2.7230411529541017,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 1.3093736171722412,
1184
+ "learning_rate": 5.151060450265419e-05,
1185
+ "loss": 2.998219871520996,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 1.1871122121810913,
1191
+ "learning_rate": 5.025073363338111e-05,
1192
+ "loss": 2.8021818161010743,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.5887531042099,
1198
+ "learning_rate": 4.900127057069116e-05,
1199
+ "loss": 3.138778305053711,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 1.0622819662094116,
1205
+ "learning_rate": 4.776247670878847e-05,
1206
+ "loss": 2.8708589553833006,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.0821974277496338,
1212
+ "learning_rate": 4.65346112098246e-05,
1213
+ "loss": 2.923636054992676,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 1.2138561010360718,
1219
+ "learning_rate": 4.5317930949680524e-05,
1220
+ "loss": 3.0319366455078125,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 1.3472957611083984,
1226
+ "learning_rate": 4.41126904642267e-05,
1227
+ "loss": 2.7567047119140624,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 1.985767126083374,
1233
+ "learning_rate": 4.291914189607297e-05,
1234
+ "loss": 3.056001663208008,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 1.1290528774261475,
1240
+ "learning_rate": 4.17375349418189e-05,
1241
+ "loss": 3.075904655456543,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 1.2921662330627441,
1247
+ "learning_rate": 4.056811679981609e-05,
1248
+ "loss": 2.8820091247558595,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 1.1284035444259644,
1254
+ "learning_rate": 3.9411132118452896e-05,
1255
+ "loss": 2.897430419921875,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1220782995224,
1261
+ "learning_rate": 3.826682294497265e-05,
1262
+ "loss": 2.782516860961914,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 1.0530436038970947,
1268
+ "learning_rate": 3.7135428674836184e-05,
1269
+ "loss": 2.7634408950805662,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 1.3937631845474243,
1275
+ "learning_rate": 3.6017186001639036e-05,
1276
+ "loss": 2.894866180419922,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 1.00092613697052,
1282
+ "learning_rate": 3.491232886759398e-05,
1283
+ "loss": 2.6879959106445312,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 1.1270594596862793,
1289
+ "learning_rate": 3.382108841458904e-05,
1290
+ "loss": 3.027115821838379,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 1.3118301630020142,
1296
+ "learning_rate": 3.274369293583121e-05,
1297
+ "loss": 2.8866352081298827,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 1.2168943881988525,
1303
+ "learning_rate": 3.1680367828086676e-05,
1304
+ "loss": 2.8968345642089846,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 1.0909639596939087,
1310
+ "learning_rate": 3.063133554452645e-05,
1311
+ "loss": 2.9398611068725584,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 1.1608167886734009,
1317
+ "learning_rate": 2.9596815548187908e-05,
1318
+ "loss": 2.842852783203125,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 1.511507511138916,
1324
+ "learning_rate": 2.8577024266062146e-05,
1325
+ "loss": 2.77429141998291,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 1.1318535804748535,
1331
+ "learning_rate": 2.7572175043816206e-05,
1332
+ "loss": 2.8206886291503905,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.9189174771308899,
1338
+ "learning_rate": 2.6582478101160167e-05,
1339
+ "loss": 2.9603801727294923,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 1.180462121963501,
1345
+ "learning_rate": 2.560814048786799e-05,
1346
+ "loss": 2.8037836074829103,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 1.1114805936813354,
1352
+ "learning_rate": 2.4649366040461597e-05,
1353
+ "loss": 2.941350746154785,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 1.31864595413208,
1359
+ "learning_rate": 2.3706355339567286e-05,
1360
+ "loss": 3.0531469345092774,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.9089719653129578,
1366
+ "learning_rate": 2.2779305667953154e-05,
1367
+ "loss": 2.746686553955078,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 1.0223416090011597,
1373
+ "learning_rate": 2.186841096925657e-05,
1374
+ "loss": 2.8307025909423826,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 1.27829909324646,
1380
+ "learning_rate": 2.097386180741019e-05,
1381
+ "loss": 2.928578758239746,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 1.3367505073547363,
1387
+ "learning_rate": 2.0095845326774697e-05,
1388
+ "loss": 2.8606313705444335,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 1.1823525428771973,
1394
+ "learning_rate": 1.9234545212987688e-05,
1395
+ "loss": 3.041524124145508,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 1.2242523431777954,
1401
+ "learning_rate": 1.8390141654535265e-05,
1402
+ "loss": 3.262685775756836,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.9626563191413879,
1408
+ "learning_rate": 1.756281130505595e-05,
1409
+ "loss": 2.824736976623535,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.9038745760917664,
1415
+ "learning_rate": 1.6752727246383582e-05,
1416
+ "loss": 3.0037527084350586,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 1.0726526975631714,
1422
+ "learning_rate": 1.5960058952337887e-05,
1423
+ "loss": 2.818869781494141,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 1.0472805500030518,
1429
+ "learning_rate": 1.518497225326958e-05,
1430
+ "loss": 2.9350055694580077,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 1.1116654872894287,
1436
+ "learning_rate": 1.4427629301367773e-05,
1437
+ "loss": 2.903761291503906,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.9501079320907593,
1443
+ "learning_rate": 1.3688188536736968e-05,
1444
+ "loss": 2.784720611572266,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 1.109456181526184,
1450
+ "learning_rate": 1.2966804654250465e-05,
1451
+ "loss": 3.1309579849243163,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 1.1933435201644897,
1457
+ "learning_rate": 1.2263628571187414e-05,
1458
+ "loss": 3.0474115371704102,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 1.4391895532608032,
1464
+ "learning_rate": 1.1578807395660207e-05,
1465
+ "loss": 3.0081947326660154,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.9554082155227661,
1471
+ "learning_rate": 1.0912484395838585e-05,
1472
+ "loss": 2.9179988861083985,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 1.1937369108200073,
1478
+ "learning_rate": 1.026479896997723e-05,
1479
+ "loss": 2.705255126953125,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.9862006306648254,
1485
+ "learning_rate": 9.635886617252975e-06,
1486
+ "loss": 2.745578384399414,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 1.1251375675201416,
1492
+ "learning_rate": 9.025878909417552e-06,
1493
+ "loss": 2.965798187255859,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 1.0850991010665894,
1499
+ "learning_rate": 8.434903463272215e-06,
1500
+ "loss": 2.9574241638183594,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 1.476346731185913,
1506
+ "learning_rate": 7.86308391396956e-06,
1507
+ "loss": 2.7245121002197266,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 1.181896686553955,
1513
+ "learning_rate": 7.310539889148404e-06,
1514
+ "loss": 3.0273662567138673,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.4082521200180054,
1520
+ "learning_rate": 6.777386983907152e-06,
1521
+ "loss": 2.9661121368408203,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 1.110843539237976,
1527
+ "learning_rate": 6.263736736620551e-06,
1528
+ "loss": 2.999568557739258,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 1.1253182888031006,
1534
+ "learning_rate": 5.769696605605379e-06,
1535
+ "loss": 2.998799705505371,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 1.089568853378296,
1541
+ "learning_rate": 5.295369946639628e-06,
1542
+ "loss": 2.6438697814941405,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 1.1276156902313232,
1548
+ "learning_rate": 4.840855991339799e-06,
1549
+ "loss": 2.9343143463134767,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 1.0737465620040894,
1555
+ "learning_rate": 4.406249826401309e-06,
1556
+ "loss": 3.080368995666504,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.8834953904151917,
1562
+ "learning_rate": 3.991642373705695e-06,
1563
+ "loss": 2.917354965209961,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 1.0340644121170044,
1569
+ "learning_rate": 3.5971203712993894e-06,
1570
+ "loss": 2.9122617721557615,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 1.1595566272735596,
1576
+ "learning_rate": 3.2227663552476194e-06,
1577
+ "loss": 2.817066764831543,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 1.0316624641418457,
1583
+ "learning_rate": 2.868658642367439e-06,
1584
+ "loss": 2.764511489868164,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 1.1504026651382446,
1590
+ "learning_rate": 2.5348713138434564e-06,
1591
+ "loss": 2.9506311416625977,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 1.0357000827789307,
1597
+ "learning_rate": 2.2214741997296117e-06,
1598
+ "loss": 2.817026901245117,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.8134933710098267,
1604
+ "learning_rate": 1.928532864340438e-06,
1605
+ "loss": 2.7651405334472656,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 1.3359462022781372,
1611
+ "learning_rate": 1.6561085925346332e-06,
1612
+ "loss": 2.8869171142578125,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 1.4075947999954224,
1618
+ "learning_rate": 1.4042583768939298e-06,
1619
+ "loss": 3.0473421096801756,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.7921971082687378,
1625
+ "learning_rate": 1.1730349057999412e-06,
1626
+ "loss": 2.8001834869384767,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 1.0570473670959473,
1632
+ "learning_rate": 9.624865524115346e-07,
1633
+ "loss": 2.9640541076660156,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 1.0891289710998535,
1639
+ "learning_rate": 7.726573645448487e-07,
1640
+ "loss": 3.0531845092773438,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 1.0408469438552856,
1646
+ "learning_rate": 6.03587055458299e-07,
1647
+ "loss": 2.9867210388183594,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.054602861404419,
1653
+ "learning_rate": 4.5531099554435576e-07,
1654
+ "loss": 2.8087078094482423,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 1.1407268047332764,
1660
+ "learning_rate": 3.278602049298418e-07,
1661
+ "loss": 3.1718997955322266,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.9369794130325317,
1667
+ "learning_rate": 2.2126134698637892e-07,
1668
+ "loss": 2.9225208282470705,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 1.6008250713348389,
1674
+ "learning_rate": 1.3553672275230523e-07,
1675
+ "loss": 3.1131832122802736,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.9389436841011047,
1681
+ "learning_rate": 7.070426626711868e-08,
1682
+ "loss": 2.8224658966064453,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.8056095242500305,
1688
+ "learning_rate": 2.677754081961137e-08,
1689
+ "loss": 2.772154998779297,
1690
+ "step": 1200
1691
+ },
1692
+ {
1693
+ "epoch": 0.9983429991714996,
1694
+ "grad_norm": 1.3219596147537231,
1695
+ "learning_rate": 3.7657361103837776e-09,
1696
+ "loss": 3.1340938568115235,
1697
+ "step": 1205
1698
+ }
1699
+ ],
1700
+ "logging_steps": 5,
1701
+ "max_steps": 1207,
1702
+ "num_input_tokens_seen": 0,
1703
+ "num_train_epochs": 1,
1704
+ "save_steps": 200,
1705
+ "stateful_callbacks": {
1706
+ "TrainerControl": {
1707
+ "args": {
1708
+ "should_epoch_stop": false,
1709
+ "should_evaluate": false,
1710
+ "should_log": false,
1711
+ "should_save": true,
1712
+ "should_training_stop": true
1713
+ },
1714
+ "attributes": {}
1715
+ }
1716
+ },
1717
+ "total_flos": 2.532249769992768e+17,
1718
+ "train_batch_size": 1,
1719
+ "trial_name": null,
1720
+ "trial_params": null
1721
+ }
MOSS_Speaker-RoPE/output_lr2e4/config.json ADDED
@@ -0,0 +1,250 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_rotary_dim": 64,
171
+ "speaker_rotary_inference_seed": 42,
172
+ "speaker_rotary_std": 1.0,
173
+ "speaker_sample_rate": 16000,
174
+ "speaker_tfmel_args": {
175
+ "do_preemph": true,
176
+ "do_spec_aug": false,
177
+ "eps": 1e-08,
178
+ "f_max": 7600,
179
+ "f_min": 20,
180
+ "hop_length": 160,
181
+ "n_fft": 512,
182
+ "n_mels": 72,
183
+ "norm_signal": true,
184
+ "win_length": 400
185
+ },
186
+ "speaker_window_seconds": 2.0,
187
+ "text_config": {
188
+ "attention_bias": false,
189
+ "attention_dropout": 0.0,
190
+ "bos_token_id": null,
191
+ "dtype": "bfloat16",
192
+ "eos_token_id": null,
193
+ "head_dim": 128,
194
+ "hidden_act": "silu",
195
+ "hidden_size": 1024,
196
+ "initializer_range": 0.02,
197
+ "intermediate_size": 3072,
198
+ "layer_types": [
199
+ "full_attention",
200
+ "full_attention",
201
+ "full_attention",
202
+ "full_attention",
203
+ "full_attention",
204
+ "full_attention",
205
+ "full_attention",
206
+ "full_attention",
207
+ "full_attention",
208
+ "full_attention",
209
+ "full_attention",
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention"
227
+ ],
228
+ "max_position_embeddings": 131072,
229
+ "max_window_layers": 28,
230
+ "model_type": "qwen3",
231
+ "num_attention_heads": 16,
232
+ "num_hidden_layers": 28,
233
+ "num_key_value_heads": 8,
234
+ "pad_token_id": 151643,
235
+ "partial_rotary_factor": 0.5,
236
+ "rms_norm_eps": 1e-06,
237
+ "rope_parameters": {
238
+ "rope_theta": 1000000,
239
+ "rope_type": "default"
240
+ },
241
+ "sliding_window": null,
242
+ "tie_word_embeddings": true,
243
+ "use_cache": false,
244
+ "use_sliding_window": false,
245
+ "vocab_size": 151936
246
+ },
247
+ "tie_word_embeddings": true,
248
+ "transformers_version": "5.14.0",
249
+ "use_cache": false
250
+ }
MOSS_Speaker-RoPE/output_lr2e4/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_lr2e4/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac31de076c8011f5f18fb1486b6c05829615dafc57c289a656f1384146d966ed
3
+ size 1867849776
MOSS_Speaker-RoPE/output_lr2e4/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb21895752d2380b32ae2e2c40905c7aa6c8e55c5e6ba33cd065bf77d4d53f74
3
+ size 11423344
MOSS_Speaker-RoPE/output_lr2e4/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_lr2e4/train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "total_flos": 2.532249769992768e+17,
4
+ "train_loss": 3.365502968245335,
5
+ "train_runtime": 20256.5023,
6
+ "train_samples_per_second": 0.06,
7
+ "train_steps_per_second": 0.06
8
+ }
MOSS_Speaker-RoPE/output_lr2e4/trainer_state.json ADDED
@@ -0,0 +1,1730 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1207,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 1147.55224609375,
15
+ "learning_rate": 6.611570247933885e-06,
16
+ "loss": 11.37186279296875,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 246.35504150390625,
22
+ "learning_rate": 1.487603305785124e-05,
23
+ "loss": 9.2022216796875,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 79.90882110595703,
29
+ "learning_rate": 2.3140495867768598e-05,
30
+ "loss": 7.6917167663574215,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 158.79872131347656,
36
+ "learning_rate": 3.1404958677685955e-05,
37
+ "loss": 6.138848114013672,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 48.520408630371094,
43
+ "learning_rate": 3.9669421487603306e-05,
44
+ "loss": 5.638393402099609,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 14.776691436767578,
50
+ "learning_rate": 4.793388429752066e-05,
51
+ "loss": 5.221096801757812,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 12.125936508178711,
57
+ "learning_rate": 5.619834710743802e-05,
58
+ "loss": 4.968892288208008,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 14.540782928466797,
64
+ "learning_rate": 6.446280991735537e-05,
65
+ "loss": 5.3284343719482425,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 13.903443336486816,
71
+ "learning_rate": 7.272727272727273e-05,
72
+ "loss": 5.018967819213867,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 12.485719680786133,
78
+ "learning_rate": 8.099173553719009e-05,
79
+ "loss": 4.600848770141601,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 44.14723205566406,
85
+ "learning_rate": 8.925619834710744e-05,
86
+ "loss": 4.335272979736328,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 9.01069164276123,
92
+ "learning_rate": 9.75206611570248e-05,
93
+ "loss": 4.316203308105469,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 16.169389724731445,
99
+ "learning_rate": 0.00010578512396694216,
100
+ "loss": 4.197259140014649,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 6.058357238769531,
106
+ "learning_rate": 0.0001140495867768595,
107
+ "loss": 4.562212371826172,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 5.126786231994629,
113
+ "learning_rate": 0.00012231404958677685,
114
+ "loss": 4.340535354614258,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 12.424266815185547,
120
+ "learning_rate": 0.00013057851239669423,
121
+ "loss": 3.9918304443359376,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 8.042577743530273,
127
+ "learning_rate": 0.0001388429752066116,
128
+ "loss": 4.087551879882812,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 8.642739295959473,
134
+ "learning_rate": 0.00014710743801652894,
135
+ "loss": 4.22486686706543,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 8.129426002502441,
141
+ "learning_rate": 0.00015537190082644627,
142
+ "loss": 3.8094669342041017,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 10.393760681152344,
148
+ "learning_rate": 0.00016363636363636366,
149
+ "loss": 3.687621307373047,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 23.797346115112305,
155
+ "learning_rate": 0.00017190082644628102,
156
+ "loss": 4.8999065399169925,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 7.251102924346924,
162
+ "learning_rate": 0.00018016528925619835,
163
+ "loss": 4.111215972900391,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 7.632305145263672,
169
+ "learning_rate": 0.0001884297520661157,
170
+ "loss": 3.88475341796875,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 2.708984851837158,
176
+ "learning_rate": 0.0001966942148760331,
177
+ "loss": 3.9735763549804686,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 3.5484111309051514,
183
+ "learning_rate": 0.00019999623426388962,
184
+ "loss": 3.911006546020508,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 3.580739974975586,
190
+ "learning_rate": 0.0001999732224591804,
191
+ "loss": 3.6748367309570313,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 3.1602091789245605,
197
+ "learning_rate": 0.0001999292957337329,
198
+ "loss": 4.212306213378906,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 2.242072582244873,
204
+ "learning_rate": 0.0001998644632772477,
205
+ "loss": 3.8557418823242187,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 3.2909867763519287,
211
+ "learning_rate": 0.00019977873865301363,
212
+ "loss": 3.953845977783203,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 2.695601224899292,
218
+ "learning_rate": 0.00019967213979507018,
219
+ "loss": 3.7311931610107423,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 3.831533670425415,
225
+ "learning_rate": 0.00019954468900445566,
226
+ "loss": 3.88598518371582,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 4.886938571929932,
232
+ "learning_rate": 0.00019939641294454172,
233
+ "loss": 3.8442028045654295,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 3.451791524887085,
239
+ "learning_rate": 0.00019922734263545517,
240
+ "loss": 3.9103469848632812,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 3.0034420490264893,
246
+ "learning_rate": 0.00019903751344758848,
247
+ "loss": 4.032486343383789,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 2.849339723587036,
253
+ "learning_rate": 0.00019882696509420007,
254
+ "loss": 3.407342529296875,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 2.095475435256958,
260
+ "learning_rate": 0.00019859574162310608,
261
+ "loss": 3.7116188049316405,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 3.148158073425293,
267
+ "learning_rate": 0.0001983438914074654,
268
+ "loss": 3.7624229431152343,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 2.4163601398468018,
274
+ "learning_rate": 0.00019807146713565955,
275
+ "loss": 3.439311218261719,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 2.7623701095581055,
281
+ "learning_rate": 0.00019777852580027042,
282
+ "loss": 3.8211971282958985,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 3.145005226135254,
288
+ "learning_rate": 0.00019746512868615656,
289
+ "loss": 3.5431480407714844,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 2.633315324783325,
295
+ "learning_rate": 0.00019713134135763256,
296
+ "loss": 3.784006881713867,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 2.0188381671905518,
302
+ "learning_rate": 0.00019677723364475237,
303
+ "loss": 3.7024520874023437,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 2.1343772411346436,
309
+ "learning_rate": 0.00019640287962870062,
310
+ "loss": 3.7561168670654297,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 1.9709585905075073,
316
+ "learning_rate": 0.0001960083576262943,
317
+ "loss": 3.6580154418945314,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 2.0804245471954346,
323
+ "learning_rate": 0.0001955937501735987,
324
+ "loss": 3.379087448120117,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 1.890079379081726,
330
+ "learning_rate": 0.0001951591440086602,
331
+ "loss": 3.5966243743896484,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 2.0973691940307617,
337
+ "learning_rate": 0.0001947046300533604,
338
+ "loss": 3.4583953857421874,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 2.4010658264160156,
344
+ "learning_rate": 0.00019423030339439464,
345
+ "loss": 3.3876201629638674,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 2.172189235687256,
351
+ "learning_rate": 0.00019373626326337946,
352
+ "loss": 3.7462398529052736,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 1.9334107637405396,
358
+ "learning_rate": 0.00019322261301609286,
359
+ "loss": 3.841419219970703,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 2.0680906772613525,
365
+ "learning_rate": 0.00019268946011085158,
366
+ "loss": 3.4502696990966797,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 2.5843758583068848,
372
+ "learning_rate": 0.00019213691608603047,
373
+ "loss": 3.404983139038086,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 1.888687252998352,
379
+ "learning_rate": 0.00019156509653672782,
380
+ "loss": 3.549253463745117,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 1.8593695163726807,
386
+ "learning_rate": 0.00019097412109058247,
387
+ "loss": 3.422663116455078,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 2.036330461502075,
393
+ "learning_rate": 0.00019036411338274703,
394
+ "loss": 3.7582332611083986,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 1.8483343124389648,
400
+ "learning_rate": 0.00018973520103002277,
401
+ "loss": 3.517898941040039,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 2.0232744216918945,
407
+ "learning_rate": 0.00018908751560416142,
408
+ "loss": 3.521917724609375,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 3.3193116188049316,
414
+ "learning_rate": 0.00018842119260433982,
415
+ "loss": 3.4880836486816404,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 2.5043585300445557,
421
+ "learning_rate": 0.0001877363714288126,
422
+ "loss": 3.446257400512695,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 1.9308786392211914,
428
+ "learning_rate": 0.00018703319534574956,
429
+ "loss": 3.3181568145751954,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 1.857397198677063,
435
+ "learning_rate": 0.00018631181146326305,
436
+ "loss": 3.219373321533203,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.9737612009048462,
442
+ "learning_rate": 0.00018557237069863222,
443
+ "loss": 3.5038528442382812,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 4.4995598793029785,
449
+ "learning_rate": 0.00018481502774673044,
450
+ "loss": 3.3514575958251953,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 2.948864221572876,
456
+ "learning_rate": 0.00018403994104766212,
457
+ "loss": 3.13072509765625,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 2.1394331455230713,
463
+ "learning_rate": 0.00018324727275361645,
464
+ "loss": 3.5682483673095704,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 3.813765287399292,
470
+ "learning_rate": 0.00018243718869494408,
471
+ "loss": 3.478915786743164,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.6351795196533203,
477
+ "learning_rate": 0.00018160985834546475,
478
+ "loss": 3.307242584228516,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 2.2016730308532715,
484
+ "learning_rate": 0.00018076545478701235,
485
+ "loss": 3.5482852935791014,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 2.1184046268463135,
491
+ "learning_rate": 0.00017990415467322534,
492
+ "loss": 3.51483039855957,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 1.704121470451355,
498
+ "learning_rate": 0.00017902613819258985,
499
+ "loss": 3.6264686584472656,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 1.926048994064331,
505
+ "learning_rate": 0.00017813158903074342,
506
+ "loss": 3.472669219970703,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 2.019219160079956,
512
+ "learning_rate": 0.00017722069433204687,
513
+ "loss": 3.2899856567382812,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 2.0369884967803955,
519
+ "learning_rate": 0.00017629364466043273,
520
+ "loss": 3.427472686767578,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 1.7189043760299683,
526
+ "learning_rate": 0.0001753506339595384,
527
+ "loss": 3.403704833984375,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 2.494591236114502,
533
+ "learning_rate": 0.00017439185951213202,
534
+ "loss": 3.5159603118896485,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 1.7678285837173462,
540
+ "learning_rate": 0.00017341752189883983,
541
+ "loss": 3.119024467468262,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 2.24629807472229,
547
+ "learning_rate": 0.0001724278249561838,
548
+ "loss": 3.525630569458008,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 1.8001821041107178,
554
+ "learning_rate": 0.0001714229757339379,
555
+ "loss": 3.541637420654297,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 2.8331096172332764,
561
+ "learning_rate": 0.0001704031844518121,
562
+ "loss": 3.3902435302734375,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 2.052628755569458,
568
+ "learning_rate": 0.00016936866445547353,
569
+ "loss": 3.218609619140625,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 2.2566065788269043,
575
+ "learning_rate": 0.00016831963217191334,
576
+ "loss": 3.5105533599853516,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 1.5952314138412476,
582
+ "learning_rate": 0.0001672563070641688,
583
+ "loss": 3.253592681884766,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 1.6297531127929688,
589
+ "learning_rate": 0.000166178911585411,
590
+ "loss": 3.176485061645508,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 1.6768420934677124,
596
+ "learning_rate": 0.00016508767113240598,
597
+ "loss": 3.4830841064453124,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 1.9300287961959839,
603
+ "learning_rate": 0.00016398281399836097,
604
+ "loss": 3.309653472900391,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 1.4318655729293823,
610
+ "learning_rate": 0.00016286457132516383,
611
+ "loss": 3.589862060546875,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 1.4944521188735962,
617
+ "learning_rate": 0.0001617331770550274,
618
+ "loss": 3.277871322631836,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.7870607376098633,
624
+ "learning_rate": 0.00016058886788154712,
625
+ "loss": 3.269728088378906,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 1.5668118000030518,
631
+ "learning_rate": 0.00015943188320018388,
632
+ "loss": 3.210224914550781,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 2.267331123352051,
638
+ "learning_rate": 0.00015826246505818112,
639
+ "loss": 3.3089523315429688,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 2.414400815963745,
645
+ "learning_rate": 0.0001570808581039271,
646
+ "loss": 3.1313955307006838,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 1.8084098100662231,
652
+ "learning_rate": 0.00015588730953577335,
653
+ "loss": 3.372232437133789,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 1.4967864751815796,
659
+ "learning_rate": 0.0001546820690503195,
660
+ "loss": 3.3817447662353515,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 1.7170981168746948,
666
+ "learning_rate": 0.0001534653887901754,
667
+ "loss": 3.1899662017822266,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 1.7449836730957031,
673
+ "learning_rate": 0.00015223752329121155,
674
+ "loss": 3.4185081481933595,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 2.0558719635009766,
680
+ "learning_rate": 0.00015099872942930887,
681
+ "loss": 3.4337844848632812,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 1.402202844619751,
687
+ "learning_rate": 0.0001497492663666189,
688
+ "loss": 3.000233268737793,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 1.575924038887024,
694
+ "learning_rate": 0.0001484893954973458,
695
+ "loss": 3.401926803588867,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.8477239608764648,
701
+ "learning_rate": 0.00014721938039306134,
702
+ "loss": 3.143557548522949,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 1.3538575172424316,
708
+ "learning_rate": 0.00014593948674756417,
709
+ "loss": 3.316628265380859,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 1.7879822254180908,
715
+ "learning_rate": 0.0001446499823212954,
716
+ "loss": 3.032166862487793,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.9181793928146362,
722
+ "learning_rate": 0.00014335113688532182,
723
+ "loss": 3.029685401916504,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 1.4723820686340332,
729
+ "learning_rate": 0.00014204322216489814,
730
+ "loss": 3.1593692779541014,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 1.5803518295288086,
736
+ "learning_rate": 0.00014072651178262096,
737
+ "loss": 3.1910980224609373,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 1.4509774446487427,
743
+ "learning_rate": 0.00013940128120118516,
744
+ "loss": 3.3486377716064455,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 1.2490512132644653,
750
+ "learning_rate": 0.00013806780766575588,
751
+ "loss": 3.127117919921875,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 1.3815803527832031,
757
+ "learning_rate": 0.0001367263701459674,
758
+ "loss": 3.3135940551757814,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 1.6695923805236816,
764
+ "learning_rate": 0.00013537724927756094,
765
+ "loss": 3.24918212890625,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 1.6700210571289062,
771
+ "learning_rate": 0.00013402072730367475,
772
+ "loss": 3.1998600006103515,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 1.458272099494934,
778
+ "learning_rate": 0.0001326570880157967,
779
+ "loss": 3.2691417694091798,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 1.6035007238388062,
785
+ "learning_rate": 0.00013128661669439416,
786
+ "loss": 3.191153717041016,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 1.3588290214538574,
792
+ "learning_rate": 0.00012990960004923154,
793
+ "loss": 3.177682113647461,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.3084040880203247,
799
+ "learning_rate": 0.0001285263261593893,
800
+ "loss": 3.0778690338134767,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 1.7576879262924194,
806
+ "learning_rate": 0.000127137084412996,
807
+ "loss": 3.063307189941406,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 1.61252760887146,
813
+ "learning_rate": 0.0001257421654466872,
814
+ "loss": 3.2951290130615236,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 1.5347871780395508,
820
+ "learning_rate": 0.0001243418610848024,
821
+ "loss": 3.1931894302368162,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 2.2056219577789307,
827
+ "learning_rate": 0.00012293646427833404,
828
+ "loss": 3.3431358337402344,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 1.6683447360992432,
834
+ "learning_rate": 0.00012152626904364067,
835
+ "loss": 3.057468795776367,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 1.455543875694275,
841
+ "learning_rate": 0.00012011157040093696,
842
+ "loss": 3.289105987548828,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 1.7347569465637207,
848
+ "learning_rate": 0.00011869266431257422,
849
+ "loss": 2.9250898361206055,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 1.3640249967575073,
855
+ "learning_rate": 0.00011726984762112328,
856
+ "loss": 3.325461578369141,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 1.4054839611053467,
862
+ "learning_rate": 0.00011584341798727365,
863
+ "loss": 3.168132019042969,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 1.2563236951828003,
869
+ "learning_rate": 0.0001144136738275612,
870
+ "loss": 3.237269973754883,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 2.893691062927246,
876
+ "learning_rate": 0.00011298091425193806,
877
+ "loss": 3.241581344604492,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 1.9952828884124756,
883
+ "learning_rate": 0.00011154543900119698,
884
+ "loss": 3.1403961181640625,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 1.573317050933838,
890
+ "learning_rate": 0.00011010754838426428,
891
+ "loss": 3.106332778930664,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 2.1935744285583496,
897
+ "learning_rate": 0.00010866754321537338,
898
+ "loss": 3.1031381607055666,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 1.5979467630386353,
904
+ "learning_rate": 0.00010722572475113315,
905
+ "loss": 3.1226659774780274,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 1.1873749494552612,
911
+ "learning_rate": 0.00010578239462750321,
912
+ "loss": 3.1140846252441405,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 1.8321475982666016,
918
+ "learning_rate": 0.00010433785479669038,
919
+ "loss": 3.0106977462768554,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 1.3288156986236572,
925
+ "learning_rate": 0.00010289240746397832,
926
+ "loss": 2.8980552673339846,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 1.416706919670105,
932
+ "learning_rate": 0.00010144635502450508,
933
+ "loss": 3.141121673583984,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 1.9086672067642212,
939
+ "learning_rate": 0.0001,
940
+ "loss": 3.282766342163086,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 1.4448398351669312,
946
+ "learning_rate": 9.855364497549496e-05,
947
+ "loss": 2.973546600341797,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 1.2777249813079834,
953
+ "learning_rate": 9.71075925360217e-05,
954
+ "loss": 3.145435905456543,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 1.276002287864685,
960
+ "learning_rate": 9.566214520330966e-05,
961
+ "loss": 3.2174652099609373,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 1.4613112211227417,
967
+ "learning_rate": 9.421760537249682e-05,
968
+ "loss": 3.17722110748291,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 1.1228781938552856,
974
+ "learning_rate": 9.277427524886689e-05,
975
+ "loss": 2.98958740234375,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 1.1255987882614136,
981
+ "learning_rate": 9.133245678462663e-05,
982
+ "loss": 3.1424510955810545,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.194167137145996,
988
+ "learning_rate": 8.989245161573576e-05,
989
+ "loss": 3.0195308685302735,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 1.864384412765503,
995
+ "learning_rate": 8.845456099880303e-05,
996
+ "loss": 2.997172737121582,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 1.241004467010498,
1002
+ "learning_rate": 8.701908574806197e-05,
1003
+ "loss": 2.9221866607666014,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 1.1507432460784912,
1009
+ "learning_rate": 8.55863261724388e-05,
1010
+ "loss": 2.9324596405029295,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.287448763847351,
1016
+ "learning_rate": 8.415658201272636e-05,
1017
+ "loss": 2.975849914550781,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 1.6067427396774292,
1023
+ "learning_rate": 8.273015237887673e-05,
1024
+ "loss": 2.980916213989258,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 1.4008774757385254,
1030
+ "learning_rate": 8.130733568742579e-05,
1031
+ "loss": 2.8944910049438475,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 1.2580070495605469,
1037
+ "learning_rate": 7.988842959906306e-05,
1038
+ "loss": 2.969981384277344,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 1.4962762594223022,
1044
+ "learning_rate": 7.847373095635937e-05,
1045
+ "loss": 3.1294715881347654,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 1.2535191774368286,
1051
+ "learning_rate": 7.706353572166597e-05,
1052
+ "loss": 3.066376876831055,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 0.8347238898277283,
1058
+ "learning_rate": 7.565813891519765e-05,
1059
+ "loss": 3.0157793045043944,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 1.2819799184799194,
1065
+ "learning_rate": 7.425783455331281e-05,
1066
+ "loss": 3.0389671325683594,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 1.1498677730560303,
1072
+ "learning_rate": 7.2862915587004e-05,
1073
+ "loss": 2.986423301696777,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 1.1965627670288086,
1079
+ "learning_rate": 7.147367384061074e-05,
1080
+ "loss": 2.941807746887207,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 1.8370661735534668,
1086
+ "learning_rate": 7.009039995076844e-05,
1087
+ "loss": 2.658160400390625,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.418321132659912,
1093
+ "learning_rate": 6.871338330560587e-05,
1094
+ "loss": 3.019854736328125,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 1.3217508792877197,
1100
+ "learning_rate": 6.734291198420333e-05,
1101
+ "loss": 2.8131311416625975,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.207661747932434,
1107
+ "learning_rate": 6.597927269632526e-05,
1108
+ "loss": 3.063412666320801,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 1.1648913621902466,
1114
+ "learning_rate": 6.462275072243908e-05,
1115
+ "loss": 3.0537391662597657,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 1.1196516752243042,
1121
+ "learning_rate": 6.327362985403265e-05,
1122
+ "loss": 3.0568498611450194,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 1.1875085830688477,
1128
+ "learning_rate": 6.193219233424414e-05,
1129
+ "loss": 2.8038040161132813,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 1.5340442657470703,
1135
+ "learning_rate": 6.059871879881487e-05,
1136
+ "loss": 2.922003746032715,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 1.4385885000228882,
1142
+ "learning_rate": 5.927348821737906e-05,
1143
+ "loss": 2.950858306884766,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 1.0835936069488525,
1149
+ "learning_rate": 5.795677783510187e-05,
1150
+ "loss": 2.898311996459961,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 1.4508082866668701,
1156
+ "learning_rate": 5.664886311467821e-05,
1157
+ "loss": 3.036098670959473,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 1.0718936920166016,
1163
+ "learning_rate": 5.535001767870463e-05,
1164
+ "loss": 2.8239120483398437,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 1.3246824741363525,
1170
+ "learning_rate": 5.406051325243586e-05,
1171
+ "loss": 2.9098350524902346,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 1.3559950590133667,
1177
+ "learning_rate": 5.278061960693865e-05,
1178
+ "loss": 2.7230411529541017,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 1.3093736171722412,
1184
+ "learning_rate": 5.151060450265419e-05,
1185
+ "loss": 2.998219871520996,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 1.1871122121810913,
1191
+ "learning_rate": 5.025073363338111e-05,
1192
+ "loss": 2.8021818161010743,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.5887531042099,
1198
+ "learning_rate": 4.900127057069116e-05,
1199
+ "loss": 3.138778305053711,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 1.0622819662094116,
1205
+ "learning_rate": 4.776247670878847e-05,
1206
+ "loss": 2.8708589553833006,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.0821974277496338,
1212
+ "learning_rate": 4.65346112098246e-05,
1213
+ "loss": 2.923636054992676,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 1.2138561010360718,
1219
+ "learning_rate": 4.5317930949680524e-05,
1220
+ "loss": 3.0319366455078125,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 1.3472957611083984,
1226
+ "learning_rate": 4.41126904642267e-05,
1227
+ "loss": 2.7567047119140624,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 1.985767126083374,
1233
+ "learning_rate": 4.291914189607297e-05,
1234
+ "loss": 3.056001663208008,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 1.1290528774261475,
1240
+ "learning_rate": 4.17375349418189e-05,
1241
+ "loss": 3.075904655456543,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 1.2921662330627441,
1247
+ "learning_rate": 4.056811679981609e-05,
1248
+ "loss": 2.8820091247558595,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 1.1284035444259644,
1254
+ "learning_rate": 3.9411132118452896e-05,
1255
+ "loss": 2.897430419921875,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1220782995224,
1261
+ "learning_rate": 3.826682294497265e-05,
1262
+ "loss": 2.782516860961914,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 1.0530436038970947,
1268
+ "learning_rate": 3.7135428674836184e-05,
1269
+ "loss": 2.7634408950805662,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 1.3937631845474243,
1275
+ "learning_rate": 3.6017186001639036e-05,
1276
+ "loss": 2.894866180419922,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 1.00092613697052,
1282
+ "learning_rate": 3.491232886759398e-05,
1283
+ "loss": 2.6879959106445312,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 1.1270594596862793,
1289
+ "learning_rate": 3.382108841458904e-05,
1290
+ "loss": 3.027115821838379,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 1.3118301630020142,
1296
+ "learning_rate": 3.274369293583121e-05,
1297
+ "loss": 2.8866352081298827,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 1.2168943881988525,
1303
+ "learning_rate": 3.1680367828086676e-05,
1304
+ "loss": 2.8968345642089846,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 1.0909639596939087,
1310
+ "learning_rate": 3.063133554452645e-05,
1311
+ "loss": 2.9398611068725584,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 1.1608167886734009,
1317
+ "learning_rate": 2.9596815548187908e-05,
1318
+ "loss": 2.842852783203125,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 1.511507511138916,
1324
+ "learning_rate": 2.8577024266062146e-05,
1325
+ "loss": 2.77429141998291,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 1.1318535804748535,
1331
+ "learning_rate": 2.7572175043816206e-05,
1332
+ "loss": 2.8206886291503905,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.9189174771308899,
1338
+ "learning_rate": 2.6582478101160167e-05,
1339
+ "loss": 2.9603801727294923,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 1.180462121963501,
1345
+ "learning_rate": 2.560814048786799e-05,
1346
+ "loss": 2.8037836074829103,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 1.1114805936813354,
1352
+ "learning_rate": 2.4649366040461597e-05,
1353
+ "loss": 2.941350746154785,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 1.31864595413208,
1359
+ "learning_rate": 2.3706355339567286e-05,
1360
+ "loss": 3.0531469345092774,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.9089719653129578,
1366
+ "learning_rate": 2.2779305667953154e-05,
1367
+ "loss": 2.746686553955078,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 1.0223416090011597,
1373
+ "learning_rate": 2.186841096925657e-05,
1374
+ "loss": 2.8307025909423826,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 1.27829909324646,
1380
+ "learning_rate": 2.097386180741019e-05,
1381
+ "loss": 2.928578758239746,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 1.3367505073547363,
1387
+ "learning_rate": 2.0095845326774697e-05,
1388
+ "loss": 2.8606313705444335,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 1.1823525428771973,
1394
+ "learning_rate": 1.9234545212987688e-05,
1395
+ "loss": 3.041524124145508,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 1.2242523431777954,
1401
+ "learning_rate": 1.8390141654535265e-05,
1402
+ "loss": 3.262685775756836,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.9626563191413879,
1408
+ "learning_rate": 1.756281130505595e-05,
1409
+ "loss": 2.824736976623535,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.9038745760917664,
1415
+ "learning_rate": 1.6752727246383582e-05,
1416
+ "loss": 3.0037527084350586,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 1.0726526975631714,
1422
+ "learning_rate": 1.5960058952337887e-05,
1423
+ "loss": 2.818869781494141,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 1.0472805500030518,
1429
+ "learning_rate": 1.518497225326958e-05,
1430
+ "loss": 2.9350055694580077,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 1.1116654872894287,
1436
+ "learning_rate": 1.4427629301367773e-05,
1437
+ "loss": 2.903761291503906,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.9501079320907593,
1443
+ "learning_rate": 1.3688188536736968e-05,
1444
+ "loss": 2.784720611572266,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 1.109456181526184,
1450
+ "learning_rate": 1.2966804654250465e-05,
1451
+ "loss": 3.1309579849243163,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 1.1933435201644897,
1457
+ "learning_rate": 1.2263628571187414e-05,
1458
+ "loss": 3.0474115371704102,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 1.4391895532608032,
1464
+ "learning_rate": 1.1578807395660207e-05,
1465
+ "loss": 3.0081947326660154,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.9554082155227661,
1471
+ "learning_rate": 1.0912484395838585e-05,
1472
+ "loss": 2.9179988861083985,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 1.1937369108200073,
1478
+ "learning_rate": 1.026479896997723e-05,
1479
+ "loss": 2.705255126953125,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.9862006306648254,
1485
+ "learning_rate": 9.635886617252975e-06,
1486
+ "loss": 2.745578384399414,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 1.1251375675201416,
1492
+ "learning_rate": 9.025878909417552e-06,
1493
+ "loss": 2.965798187255859,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 1.0850991010665894,
1499
+ "learning_rate": 8.434903463272215e-06,
1500
+ "loss": 2.9574241638183594,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 1.476346731185913,
1506
+ "learning_rate": 7.86308391396956e-06,
1507
+ "loss": 2.7245121002197266,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 1.181896686553955,
1513
+ "learning_rate": 7.310539889148404e-06,
1514
+ "loss": 3.0273662567138673,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.4082521200180054,
1520
+ "learning_rate": 6.777386983907152e-06,
1521
+ "loss": 2.9661121368408203,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 1.110843539237976,
1527
+ "learning_rate": 6.263736736620551e-06,
1528
+ "loss": 2.999568557739258,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 1.1253182888031006,
1534
+ "learning_rate": 5.769696605605379e-06,
1535
+ "loss": 2.998799705505371,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 1.089568853378296,
1541
+ "learning_rate": 5.295369946639628e-06,
1542
+ "loss": 2.6438697814941405,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 1.1276156902313232,
1548
+ "learning_rate": 4.840855991339799e-06,
1549
+ "loss": 2.9343143463134767,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 1.0737465620040894,
1555
+ "learning_rate": 4.406249826401309e-06,
1556
+ "loss": 3.080368995666504,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.8834953904151917,
1562
+ "learning_rate": 3.991642373705695e-06,
1563
+ "loss": 2.917354965209961,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 1.0340644121170044,
1569
+ "learning_rate": 3.5971203712993894e-06,
1570
+ "loss": 2.9122617721557615,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 1.1595566272735596,
1576
+ "learning_rate": 3.2227663552476194e-06,
1577
+ "loss": 2.817066764831543,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 1.0316624641418457,
1583
+ "learning_rate": 2.868658642367439e-06,
1584
+ "loss": 2.764511489868164,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 1.1504026651382446,
1590
+ "learning_rate": 2.5348713138434564e-06,
1591
+ "loss": 2.9506311416625977,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 1.0357000827789307,
1597
+ "learning_rate": 2.2214741997296117e-06,
1598
+ "loss": 2.817026901245117,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.8134933710098267,
1604
+ "learning_rate": 1.928532864340438e-06,
1605
+ "loss": 2.7651405334472656,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 1.3359462022781372,
1611
+ "learning_rate": 1.6561085925346332e-06,
1612
+ "loss": 2.8869171142578125,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 1.4075947999954224,
1618
+ "learning_rate": 1.4042583768939298e-06,
1619
+ "loss": 3.0473421096801756,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.7921971082687378,
1625
+ "learning_rate": 1.1730349057999412e-06,
1626
+ "loss": 2.8001834869384767,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 1.0570473670959473,
1632
+ "learning_rate": 9.624865524115346e-07,
1633
+ "loss": 2.9640541076660156,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 1.0891289710998535,
1639
+ "learning_rate": 7.726573645448487e-07,
1640
+ "loss": 3.0531845092773438,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 1.0408469438552856,
1646
+ "learning_rate": 6.03587055458299e-07,
1647
+ "loss": 2.9867210388183594,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.054602861404419,
1653
+ "learning_rate": 4.5531099554435576e-07,
1654
+ "loss": 2.8087078094482423,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 1.1407268047332764,
1660
+ "learning_rate": 3.278602049298418e-07,
1661
+ "loss": 3.1718997955322266,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.9369794130325317,
1667
+ "learning_rate": 2.2126134698637892e-07,
1668
+ "loss": 2.9225208282470705,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 1.6008250713348389,
1674
+ "learning_rate": 1.3553672275230523e-07,
1675
+ "loss": 3.1131832122802736,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.9389436841011047,
1681
+ "learning_rate": 7.070426626711868e-08,
1682
+ "loss": 2.8224658966064453,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.8056095242500305,
1688
+ "learning_rate": 2.677754081961137e-08,
1689
+ "loss": 2.772154998779297,
1690
+ "step": 1200
1691
+ },
1692
+ {
1693
+ "epoch": 0.9983429991714996,
1694
+ "grad_norm": 1.3219596147537231,
1695
+ "learning_rate": 3.7657361103837776e-09,
1696
+ "loss": 3.1340938568115235,
1697
+ "step": 1205
1698
+ },
1699
+ {
1700
+ "epoch": 1.0,
1701
+ "step": 1207,
1702
+ "total_flos": 2.532249769992768e+17,
1703
+ "train_loss": 3.365502968245335,
1704
+ "train_runtime": 20256.5023,
1705
+ "train_samples_per_second": 0.06,
1706
+ "train_steps_per_second": 0.06
1707
+ }
1708
+ ],
1709
+ "logging_steps": 5,
1710
+ "max_steps": 1207,
1711
+ "num_input_tokens_seen": 0,
1712
+ "num_train_epochs": 1,
1713
+ "save_steps": 200,
1714
+ "stateful_callbacks": {
1715
+ "TrainerControl": {
1716
+ "args": {
1717
+ "should_epoch_stop": false,
1718
+ "should_evaluate": false,
1719
+ "should_log": false,
1720
+ "should_save": true,
1721
+ "should_training_stop": true
1722
+ },
1723
+ "attributes": {}
1724
+ }
1725
+ },
1726
+ "total_flos": 2.532249769992768e+17,
1727
+ "train_batch_size": 1,
1728
+ "trial_name": null,
1729
+ "trial_params": null
1730
+ }
MOSS_Speaker-RoPE/output_lr2e4/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c303d1b2cafcf35829f18c90c85ffa34cd0d19313509785e9189f8bee8b2e432
3
+ size 5201
MOSS_Speaker-RoPE/output_spk_odd8_s3/all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "total_flos": 2.5323923576872166e+17,
4
+ "train_loss": 0.43766265026083045,
5
+ "train_runtime": 20328.9815,
6
+ "train_samples_per_second": 0.059,
7
+ "train_steps_per_second": 0.059
8
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- macro render_content(content) -%}
2
+ {%- if content is string -%}
3
+ {{- content -}}
4
+ {%- else -%}
5
+ {%- for item in content -%}
6
+ {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
7
+ {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
8
+ {%- elif item.type == 'text' -%}
9
+ {{- item.text -}}
10
+ {%- endif -%}
11
+ {%- endfor -%}
12
+ {%- endif -%}
13
+ {%- endmacro -%}
14
+ {%- if tools %}
15
+ {{- '<|im_start|>system\n' }}
16
+ {%- if messages[0].role == 'system' %}
17
+ {{- render_content(messages[0].content) + '\n\n' }}
18
+ {%- else %}
19
+ {{- 'You are a helpful assistant.\n\n' }}
20
+ {%- endif %}
21
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
22
+ {%- for tool in tools %}
23
+ {{- "\n" }}
24
+ {{- tool | tojson }}
25
+ {%- endfor %}
26
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
27
+ {%- else %}
28
+ {%- if messages[0].role == 'system' %}
29
+ {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
30
+ {%- else %}
31
+ {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
32
+ {%- endif %}
33
+ {%- endif %}
34
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
35
+ {%- for message in messages[::-1] %}
36
+ {%- set index = (messages|length - 1) - loop.index0 %}
37
+ {%- set content = render_content(message.content) %}
38
+ {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
39
+ {%- set ns.multi_step_tool = false %}
40
+ {%- set ns.last_query_index = index %}
41
+ {%- endif %}
42
+ {%- endfor %}
43
+ {%- for message in messages %}
44
+ {%- set content = render_content(message.content) %}
45
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
47
+ {%- elif message.role == "assistant" %}
48
+ {%- set reasoning_content = '' %}
49
+ {%- if message.reasoning_content is string %}
50
+ {%- set reasoning_content = message.reasoning_content %}
51
+ {%- else %}
52
+ {%- if '</think>' in content %}
53
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
54
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
55
+ {%- endif %}
56
+ {%- endif %}
57
+ {%- if loop.index0 > ns.last_query_index %}
58
+ {%- if loop.last or (not loop.last and reasoning_content) %}
59
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
60
+ {%- else %}
61
+ {{- '<|im_start|>' + message.role + '\n' + content }}
62
+ {%- endif %}
63
+ {%- else %}
64
+ {{- '<|im_start|>' + message.role + '\n' + content }}
65
+ {%- endif %}
66
+ {%- if message.tool_calls %}
67
+ {%- for tool_call in message.tool_calls %}
68
+ {%- if (loop.first and content) or (not loop.first) %}
69
+ {{- '\n' }}
70
+ {%- endif %}
71
+ {%- if tool_call.function %}
72
+ {%- set tool_call = tool_call.function %}
73
+ {%- endif %}
74
+ {{- '<tool_call>\n{"name": "' }}
75
+ {{- tool_call.name }}
76
+ {{- '", "arguments": ' }}
77
+ {%- if tool_call.arguments is string %}
78
+ {{- tool_call.arguments }}
79
+ {%- else %}
80
+ {{- tool_call.arguments | tojson }}
81
+ {%- endif %}
82
+ {{- '}\n</tool_call>' }}
83
+ {%- endfor %}
84
+ {%- endif %}
85
+ {{- '<|im_end|>\n' }}
86
+ {%- elif message.role == "tool" %}
87
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
88
+ {{- '<|im_start|>user' }}
89
+ {%- endif %}
90
+ {{- '\n<tool_response>\n' }}
91
+ {{- content }}
92
+ {{- '\n</tool_response>' }}
93
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
94
+ {{- '<|im_end|>\n' }}
95
+ {%- endif %}
96
+ {%- endif %}
97
+ {%- endfor %}
98
+ {%- if add_generation_prompt %}
99
+ {{- '<|im_start|>assistant\n' }}
100
+ {%- if enable_thinking is defined and enable_thinking is false %}
101
+ {{- '<think>\n\n</think>\n\n' }}
102
+ {%- endif %}
103
+ {%- endif %}
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- macro render_content(content) -%}
2
+ {%- if content is string -%}
3
+ {{- content -}}
4
+ {%- else -%}
5
+ {%- for item in content -%}
6
+ {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
7
+ {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
8
+ {%- elif item.type == 'text' -%}
9
+ {{- item.text -}}
10
+ {%- endif -%}
11
+ {%- endfor -%}
12
+ {%- endif -%}
13
+ {%- endmacro -%}
14
+ {%- if tools %}
15
+ {{- '<|im_start|>system\n' }}
16
+ {%- if messages[0].role == 'system' %}
17
+ {{- render_content(messages[0].content) + '\n\n' }}
18
+ {%- else %}
19
+ {{- 'You are a helpful assistant.\n\n' }}
20
+ {%- endif %}
21
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
22
+ {%- for tool in tools %}
23
+ {{- "\n" }}
24
+ {{- tool | tojson }}
25
+ {%- endfor %}
26
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
27
+ {%- else %}
28
+ {%- if messages[0].role == 'system' %}
29
+ {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
30
+ {%- else %}
31
+ {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
32
+ {%- endif %}
33
+ {%- endif %}
34
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
35
+ {%- for message in messages[::-1] %}
36
+ {%- set index = (messages|length - 1) - loop.index0 %}
37
+ {%- set content = render_content(message.content) %}
38
+ {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
39
+ {%- set ns.multi_step_tool = false %}
40
+ {%- set ns.last_query_index = index %}
41
+ {%- endif %}
42
+ {%- endfor %}
43
+ {%- for message in messages %}
44
+ {%- set content = render_content(message.content) %}
45
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
47
+ {%- elif message.role == "assistant" %}
48
+ {%- set reasoning_content = '' %}
49
+ {%- if message.reasoning_content is string %}
50
+ {%- set reasoning_content = message.reasoning_content %}
51
+ {%- else %}
52
+ {%- if '</think>' in content %}
53
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
54
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
55
+ {%- endif %}
56
+ {%- endif %}
57
+ {%- if loop.index0 > ns.last_query_index %}
58
+ {%- if loop.last or (not loop.last and reasoning_content) %}
59
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
60
+ {%- else %}
61
+ {{- '<|im_start|>' + message.role + '\n' + content }}
62
+ {%- endif %}
63
+ {%- else %}
64
+ {{- '<|im_start|>' + message.role + '\n' + content }}
65
+ {%- endif %}
66
+ {%- if message.tool_calls %}
67
+ {%- for tool_call in message.tool_calls %}
68
+ {%- if (loop.first and content) or (not loop.first) %}
69
+ {{- '\n' }}
70
+ {%- endif %}
71
+ {%- if tool_call.function %}
72
+ {%- set tool_call = tool_call.function %}
73
+ {%- endif %}
74
+ {{- '<tool_call>\n{"name": "' }}
75
+ {{- tool_call.name }}
76
+ {{- '", "arguments": ' }}
77
+ {%- if tool_call.arguments is string %}
78
+ {{- tool_call.arguments }}
79
+ {%- else %}
80
+ {{- tool_call.arguments | tojson }}
81
+ {%- endif %}
82
+ {{- '}\n</tool_call>' }}
83
+ {%- endfor %}
84
+ {%- endif %}
85
+ {{- '<|im_end|>\n' }}
86
+ {%- elif message.role == "tool" %}
87
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
88
+ {{- '<|im_start|>user' }}
89
+ {%- endif %}
90
+ {{- '\n<tool_response>\n' }}
91
+ {{- content }}
92
+ {{- '\n</tool_response>' }}
93
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
94
+ {{- '<|im_end|>\n' }}
95
+ {%- endif %}
96
+ {%- endif %}
97
+ {%- endfor %}
98
+ {%- if add_generation_prompt %}
99
+ {{- '<|im_start|>assistant\n' }}
100
+ {%- if enable_thinking is defined and enable_thinking is false %}
101
+ {{- '<think>\n\n</think>\n\n' }}
102
+ {%- endif %}
103
+ {%- endif %}
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/config.json ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_plane_indices": [
171
+ 3,
172
+ 11,
173
+ 19,
174
+ 27,
175
+ 35,
176
+ 43,
177
+ 51,
178
+ 59
179
+ ],
180
+ "speaker_plane_order": "interleaved",
181
+ "speaker_rotary_dim": 16,
182
+ "speaker_rotary_inference_seed": 42,
183
+ "speaker_rotary_std": 0.02,
184
+ "speaker_sample_rate": 16000,
185
+ "speaker_tfmel_args": {
186
+ "do_preemph": true,
187
+ "do_spec_aug": false,
188
+ "eps": 1e-08,
189
+ "f_max": 7600,
190
+ "f_min": 20,
191
+ "hop_length": 160,
192
+ "n_fft": 512,
193
+ "n_mels": 72,
194
+ "norm_signal": true,
195
+ "win_length": 400
196
+ },
197
+ "speaker_window_seconds": 2.0,
198
+ "text_config": {
199
+ "attention_bias": false,
200
+ "attention_dropout": 0.0,
201
+ "bos_token_id": null,
202
+ "dtype": "bfloat16",
203
+ "eos_token_id": null,
204
+ "head_dim": 128,
205
+ "hidden_act": "silu",
206
+ "hidden_size": 1024,
207
+ "initializer_range": 0.02,
208
+ "intermediate_size": 3072,
209
+ "layer_types": [
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention",
227
+ "full_attention",
228
+ "full_attention",
229
+ "full_attention",
230
+ "full_attention",
231
+ "full_attention",
232
+ "full_attention",
233
+ "full_attention",
234
+ "full_attention",
235
+ "full_attention",
236
+ "full_attention",
237
+ "full_attention"
238
+ ],
239
+ "max_position_embeddings": 131072,
240
+ "max_window_layers": 28,
241
+ "model_type": "qwen3",
242
+ "num_attention_heads": 16,
243
+ "num_hidden_layers": 28,
244
+ "num_key_value_heads": 8,
245
+ "pad_token_id": 151643,
246
+ "rms_norm_eps": 1e-06,
247
+ "rope_parameters": {
248
+ "rope_theta": 1000000,
249
+ "rope_type": "default"
250
+ },
251
+ "sliding_window": null,
252
+ "tie_word_embeddings": true,
253
+ "use_cache": false,
254
+ "use_sliding_window": false,
255
+ "vocab_size": 151936
256
+ },
257
+ "tie_word_embeddings": true,
258
+ "transformers_version": "5.14.0",
259
+ "use_cache": false
260
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1200/trainer_state.json ADDED
@@ -0,0 +1,1714 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.9942004971002486,
6
+ "eval_steps": 500,
7
+ "global_step": 1200,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 20.629674911499023,
15
+ "learning_rate": 3.3057851239669424e-06,
16
+ "loss": 0.8916114807128906,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 12.812925338745117,
22
+ "learning_rate": 7.43801652892562e-06,
23
+ "loss": 0.558007287979126,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 7.200536251068115,
29
+ "learning_rate": 1.1570247933884299e-05,
30
+ "loss": 0.6662204742431641,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 1.363626480102539,
36
+ "learning_rate": 1.5702479338842978e-05,
37
+ "loss": 0.45300092697143557,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 3.3842880725860596,
43
+ "learning_rate": 1.9834710743801653e-05,
44
+ "loss": 0.494948148727417,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 1.388489842414856,
50
+ "learning_rate": 2.396694214876033e-05,
51
+ "loss": 0.4781633377075195,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 0.5891152024269104,
57
+ "learning_rate": 2.809917355371901e-05,
58
+ "loss": 0.48348569869995117,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 0.9562617540359497,
64
+ "learning_rate": 3.2231404958677685e-05,
65
+ "loss": 0.4969147205352783,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 0.6433483958244324,
71
+ "learning_rate": 3.6363636363636364e-05,
72
+ "loss": 0.4836277008056641,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 0.898301362991333,
78
+ "learning_rate": 4.049586776859504e-05,
79
+ "loss": 0.4123377799987793,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 0.8125119805335999,
85
+ "learning_rate": 4.462809917355372e-05,
86
+ "loss": 0.45714097023010253,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 0.5890400409698486,
92
+ "learning_rate": 4.87603305785124e-05,
93
+ "loss": 0.4525331497192383,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 0.6006086468696594,
99
+ "learning_rate": 5.289256198347108e-05,
100
+ "loss": 0.3964966297149658,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 1.017815113067627,
106
+ "learning_rate": 5.702479338842975e-05,
107
+ "loss": 0.37221901416778563,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 0.9333603978157043,
113
+ "learning_rate": 6.115702479338842e-05,
114
+ "loss": 0.4029701232910156,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 1.0475515127182007,
120
+ "learning_rate": 6.528925619834711e-05,
121
+ "loss": 0.45854949951171875,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 1.0967167615890503,
127
+ "learning_rate": 6.94214876033058e-05,
128
+ "loss": 0.455504035949707,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 0.7258841395378113,
134
+ "learning_rate": 7.355371900826447e-05,
135
+ "loss": 0.7798455238342286,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 1.0904923677444458,
141
+ "learning_rate": 7.768595041322314e-05,
142
+ "loss": 0.46855831146240234,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 0.924197256565094,
148
+ "learning_rate": 8.181818181818183e-05,
149
+ "loss": 0.4880524635314941,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 1.9587433338165283,
155
+ "learning_rate": 8.595041322314051e-05,
156
+ "loss": 0.463060998916626,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 0.6440144181251526,
162
+ "learning_rate": 9.008264462809917e-05,
163
+ "loss": 0.48012351989746094,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 0.8251569867134094,
169
+ "learning_rate": 9.421487603305785e-05,
170
+ "loss": 0.4483522415161133,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 0.8789721727371216,
176
+ "learning_rate": 9.834710743801654e-05,
177
+ "loss": 0.4580700397491455,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 0.8313543796539307,
183
+ "learning_rate": 9.999811713194481e-05,
184
+ "loss": 0.46596560478210447,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 0.6774608492851257,
190
+ "learning_rate": 9.99866112295902e-05,
191
+ "loss": 0.4806799411773682,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 0.6783014535903931,
197
+ "learning_rate": 9.996464786686644e-05,
198
+ "loss": 0.4560887336730957,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 0.7160524129867554,
204
+ "learning_rate": 9.993223163862385e-05,
205
+ "loss": 0.4927361011505127,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 0.5270054340362549,
211
+ "learning_rate": 9.988936932650681e-05,
212
+ "loss": 0.42853679656982424,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 1.0818812847137451,
218
+ "learning_rate": 9.983606989753509e-05,
219
+ "loss": 0.5378652572631836,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 0.968137800693512,
225
+ "learning_rate": 9.977234450222783e-05,
226
+ "loss": 0.48653225898742675,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 0.8479903936386108,
232
+ "learning_rate": 9.969820647227086e-05,
233
+ "loss": 0.5200669765472412,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 0.7051109075546265,
239
+ "learning_rate": 9.961367131772759e-05,
240
+ "loss": 0.4588489532470703,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 2.205345630645752,
246
+ "learning_rate": 9.951875672379424e-05,
247
+ "loss": 0.5904168128967285,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 0.8751126527786255,
253
+ "learning_rate": 9.941348254710003e-05,
254
+ "loss": 0.43442707061767577,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 0.671466052532196,
260
+ "learning_rate": 9.929787081155304e-05,
261
+ "loss": 0.3809401512145996,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 1.235303282737732,
267
+ "learning_rate": 9.91719457037327e-05,
268
+ "loss": 0.4351661205291748,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 0.5449215769767761,
274
+ "learning_rate": 9.903573356782977e-05,
275
+ "loss": 0.41109681129455566,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 0.6575606465339661,
281
+ "learning_rate": 9.888926290013521e-05,
282
+ "loss": 0.46851186752319335,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 0.528086245059967,
288
+ "learning_rate": 9.873256434307828e-05,
289
+ "loss": 0.40171051025390625,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 0.44805943965911865,
295
+ "learning_rate": 9.856567067881628e-05,
296
+ "loss": 0.3575094223022461,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 0.6913977861404419,
302
+ "learning_rate": 9.838861682237618e-05,
303
+ "loss": 0.4208680152893066,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 0.6070751547813416,
309
+ "learning_rate": 9.820143981435031e-05,
310
+ "loss": 0.4100965976715088,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 0.7453505396842957,
316
+ "learning_rate": 9.800417881314716e-05,
317
+ "loss": 0.5092792510986328,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 0.718826949596405,
323
+ "learning_rate": 9.779687508679935e-05,
324
+ "loss": 0.5120072841644288,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 0.8064695596694946,
330
+ "learning_rate": 9.75795720043301e-05,
331
+ "loss": 0.43071684837341306,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 0.6657348275184631,
337
+ "learning_rate": 9.73523150266802e-05,
338
+ "loss": 0.43650312423706056,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 0.6227285861968994,
344
+ "learning_rate": 9.711515169719732e-05,
345
+ "loss": 0.4965176582336426,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 0.6354798078536987,
351
+ "learning_rate": 9.686813163168973e-05,
352
+ "loss": 0.4135477542877197,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 0.8802199959754944,
358
+ "learning_rate": 9.661130650804643e-05,
359
+ "loss": 0.4657131671905518,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 0.6120206713676453,
365
+ "learning_rate": 9.634473005542579e-05,
366
+ "loss": 0.5105055332183838,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 0.6608315706253052,
372
+ "learning_rate": 9.606845804301524e-05,
373
+ "loss": 0.48976764678955076,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 0.6136434078216553,
379
+ "learning_rate": 9.578254826836391e-05,
380
+ "loss": 0.39600679874420164,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 0.7732479572296143,
386
+ "learning_rate": 9.548706054529124e-05,
387
+ "loss": 0.43799166679382323,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 0.8353448510169983,
393
+ "learning_rate": 9.518205669137352e-05,
394
+ "loss": 0.39786520004272463,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 0.5929117798805237,
400
+ "learning_rate": 9.486760051501139e-05,
401
+ "loss": 0.4511585235595703,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 0.7609494924545288,
407
+ "learning_rate": 9.454375780208071e-05,
408
+ "loss": 0.460142993927002,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 0.4482679069042206,
414
+ "learning_rate": 9.421059630216991e-05,
415
+ "loss": 0.42333049774169923,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 0.5880385637283325,
421
+ "learning_rate": 9.38681857144063e-05,
422
+ "loss": 0.39257287979125977,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 0.4800879955291748,
428
+ "learning_rate": 9.351659767287478e-05,
429
+ "loss": 0.46072850227355955,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 0.5711155533790588,
435
+ "learning_rate": 9.315590573163153e-05,
436
+ "loss": 0.4340324401855469,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.6473472118377686,
442
+ "learning_rate": 9.278618534931611e-05,
443
+ "loss": 0.4806147575378418,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 0.9981988668441772,
449
+ "learning_rate": 9.240751387336522e-05,
450
+ "loss": 0.45314898490905764,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 0.9140734076499939,
456
+ "learning_rate": 9.201997052383106e-05,
457
+ "loss": 0.48151440620422364,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 0.7532357573509216,
463
+ "learning_rate": 9.162363637680822e-05,
464
+ "loss": 0.4129056453704834,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 0.9811023473739624,
470
+ "learning_rate": 9.121859434747204e-05,
471
+ "loss": 0.484523344039917,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.4147282838821411,
477
+ "learning_rate": 9.080492917273237e-05,
478
+ "loss": 0.5427152633666992,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 1.0163606405258179,
484
+ "learning_rate": 9.038272739350618e-05,
485
+ "loss": 0.4134209156036377,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 0.6049103140830994,
491
+ "learning_rate": 8.995207733661267e-05,
492
+ "loss": 0.40175676345825195,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 0.7045316696166992,
498
+ "learning_rate": 8.951306909629492e-05,
499
+ "loss": 0.40282230377197265,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 0.6261049509048462,
505
+ "learning_rate": 8.906579451537171e-05,
506
+ "loss": 0.3981405973434448,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 1.287496566772461,
512
+ "learning_rate": 8.861034716602343e-05,
513
+ "loss": 0.4517494201660156,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 0.6869074702262878,
519
+ "learning_rate": 8.814682233021636e-05,
520
+ "loss": 0.44563779830932615,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 0.6493797302246094,
526
+ "learning_rate": 8.76753169797692e-05,
527
+ "loss": 0.4556128978729248,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 0.5839359760284424,
533
+ "learning_rate": 8.719592975606601e-05,
534
+ "loss": 0.43351993560791013,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 0.3983668088912964,
540
+ "learning_rate": 8.670876094941991e-05,
541
+ "loss": 0.4440763473510742,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 0.6596688032150269,
547
+ "learning_rate": 8.62139124780919e-05,
548
+ "loss": 0.5530457973480225,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 0.6001258492469788,
554
+ "learning_rate": 8.571148786696894e-05,
555
+ "loss": 0.4752801418304443,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 0.405813604593277,
561
+ "learning_rate": 8.520159222590605e-05,
562
+ "loss": 0.38340303897857664,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 0.8015041351318359,
568
+ "learning_rate": 8.468433222773677e-05,
569
+ "loss": 0.4637146949768066,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 0.6381627321243286,
575
+ "learning_rate": 8.415981608595667e-05,
576
+ "loss": 0.42714495658874513,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 0.5829190015792847,
582
+ "learning_rate": 8.36281535320844e-05,
583
+ "loss": 0.3956311702728271,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 0.8698112964630127,
589
+ "learning_rate": 8.30894557927055e-05,
590
+ "loss": 0.4846199989318848,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 0.6768237352371216,
596
+ "learning_rate": 8.254383556620299e-05,
597
+ "loss": 0.47880539894104,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 0.5710030198097229,
603
+ "learning_rate": 8.199140699918049e-05,
604
+ "loss": 0.4479952335357666,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 0.6229125261306763,
610
+ "learning_rate": 8.143228566258192e-05,
611
+ "loss": 0.433013916015625,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 0.6174870133399963,
617
+ "learning_rate": 8.08665885275137e-05,
618
+ "loss": 0.4820831298828125,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.9886339902877808,
624
+ "learning_rate": 8.029443394077356e-05,
625
+ "loss": 0.4312156677246094,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 0.7721105217933655,
631
+ "learning_rate": 7.971594160009194e-05,
632
+ "loss": 0.4745286464691162,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 0.8215830326080322,
638
+ "learning_rate": 7.913123252909056e-05,
639
+ "loss": 0.43494420051574706,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 0.8584961295127869,
645
+ "learning_rate": 7.854042905196354e-05,
646
+ "loss": 0.46215004920959474,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 0.483641117811203,
652
+ "learning_rate": 7.794365476788667e-05,
653
+ "loss": 0.37451136112213135,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 0.567133903503418,
659
+ "learning_rate": 7.734103452515975e-05,
660
+ "loss": 0.4840177059173584,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 0.6834204196929932,
666
+ "learning_rate": 7.67326943950877e-05,
667
+ "loss": 0.46247282028198244,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 0.8395087122917175,
673
+ "learning_rate": 7.611876164560578e-05,
674
+ "loss": 0.36043124198913573,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 0.3639385998249054,
680
+ "learning_rate": 7.549936471465444e-05,
681
+ "loss": 0.42089052200317384,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 0.4404556453227997,
687
+ "learning_rate": 7.487463318330946e-05,
688
+ "loss": 0.3542492866516113,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 0.6046218276023865,
694
+ "learning_rate": 7.42446977486729e-05,
695
+ "loss": 0.38021628856658934,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.6285887956619263,
701
+ "learning_rate": 7.360969019653067e-05,
702
+ "loss": 0.6484775066375732,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 0.9363073110580444,
708
+ "learning_rate": 7.296974337378209e-05,
709
+ "loss": 0.41137142181396485,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 0.4143483340740204,
715
+ "learning_rate": 7.23249911606477e-05,
716
+ "loss": 0.3924715518951416,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.2242801189422607,
722
+ "learning_rate": 7.167556844266091e-05,
723
+ "loss": 0.4372579574584961,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 0.49445945024490356,
729
+ "learning_rate": 7.102161108244907e-05,
730
+ "loss": 0.4041281700134277,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 0.5578067898750305,
736
+ "learning_rate": 7.036325589131048e-05,
737
+ "loss": 0.44759469032287597,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 0.6338609457015991,
743
+ "learning_rate": 6.970064060059258e-05,
744
+ "loss": 0.3942832946777344,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 0.6305315494537354,
750
+ "learning_rate": 6.903390383287794e-05,
751
+ "loss": 0.3744609832763672,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 0.6225669980049133,
757
+ "learning_rate": 6.83631850729837e-05,
758
+ "loss": 0.4644033432006836,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 0.4533715844154358,
764
+ "learning_rate": 6.768862463878047e-05,
765
+ "loss": 0.43013477325439453,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 0.668293297290802,
771
+ "learning_rate": 6.701036365183737e-05,
772
+ "loss": 0.5000359058380127,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 0.6384305357933044,
778
+ "learning_rate": 6.632854400789835e-05,
779
+ "loss": 0.4234946250915527,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 0.35919633507728577,
785
+ "learning_rate": 6.564330834719708e-05,
786
+ "loss": 0.4122346878051758,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 0.5707134008407593,
792
+ "learning_rate": 6.495480002461577e-05,
793
+ "loss": 0.4068869113922119,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.7894991636276245,
799
+ "learning_rate": 6.426316307969465e-05,
800
+ "loss": 0.4764272689819336,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 0.7096588611602783,
806
+ "learning_rate": 6.3568542206498e-05,
807
+ "loss": 0.40894289016723634,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 0.6279922723770142,
813
+ "learning_rate": 6.28710827233436e-05,
814
+ "loss": 0.4551478385925293,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 0.9392326474189758,
820
+ "learning_rate": 6.21709305424012e-05,
821
+ "loss": 0.42044363021850584,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 0.8099265098571777,
827
+ "learning_rate": 6.146823213916702e-05,
828
+ "loss": 0.421197509765625,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 0.407980740070343,
834
+ "learning_rate": 6.076313452182033e-05,
835
+ "loss": 0.45879392623901366,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 0.6435403823852539,
841
+ "learning_rate": 6.005578520046848e-05,
842
+ "loss": 0.7919346332550049,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 0.5578521490097046,
848
+ "learning_rate": 5.934633215628711e-05,
849
+ "loss": 0.4143537044525146,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 0.5052011013031006,
855
+ "learning_rate": 5.863492381056164e-05,
856
+ "loss": 0.38072867393493653,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 0.6519150733947754,
862
+ "learning_rate": 5.792170899363683e-05,
863
+ "loss": 0.4297952175140381,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 0.9843361377716064,
869
+ "learning_rate": 5.72068369137806e-05,
870
+ "loss": 0.3838667869567871,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 0.7498537302017212,
876
+ "learning_rate": 5.649045712596903e-05,
877
+ "loss": 0.40416569709777833,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 0.6527976393699646,
883
+ "learning_rate": 5.577271950059849e-05,
884
+ "loss": 0.4134161949157715,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 0.8321830630302429,
890
+ "learning_rate": 5.505377419213214e-05,
891
+ "loss": 0.43797922134399414,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 0.6492611169815063,
897
+ "learning_rate": 5.433377160768669e-05,
898
+ "loss": 0.4532886028289795,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 0.6094037294387817,
904
+ "learning_rate": 5.3612862375566574e-05,
905
+ "loss": 0.43567543029785155,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 0.551053524017334,
911
+ "learning_rate": 5.2891197313751606e-05,
912
+ "loss": 0.4400018692016602,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 0.6376934051513672,
918
+ "learning_rate": 5.216892739834519e-05,
919
+ "loss": 0.42303104400634767,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 0.9007910490036011,
925
+ "learning_rate": 5.144620373198916e-05,
926
+ "loss": 0.45666775703430174,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 0.650941014289856,
932
+ "learning_rate": 5.072317751225254e-05,
933
+ "loss": 0.4667072296142578,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 0.48627233505249023,
939
+ "learning_rate": 5e-05,
940
+ "loss": 0.3643670558929443,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 0.7920517325401306,
946
+ "learning_rate": 4.927682248774748e-05,
947
+ "loss": 0.44395933151245115,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 0.5550063252449036,
953
+ "learning_rate": 4.855379626801085e-05,
954
+ "loss": 0.36930572986602783,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 0.6230601668357849,
960
+ "learning_rate": 4.783107260165483e-05,
961
+ "loss": 0.4727597236633301,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 0.6426054835319519,
967
+ "learning_rate": 4.710880268624841e-05,
968
+ "loss": 0.460523796081543,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 0.5865359902381897,
974
+ "learning_rate": 4.6387137624433444e-05,
975
+ "loss": 0.4252904415130615,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 0.5453794598579407,
981
+ "learning_rate": 4.5666228392313315e-05,
982
+ "loss": 0.4039918422698975,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.0684459209442139,
988
+ "learning_rate": 4.494622580786788e-05,
989
+ "loss": 0.42879571914672854,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 0.8819708824157715,
995
+ "learning_rate": 4.422728049940151e-05,
996
+ "loss": 0.35405232906341555,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 0.5523223280906677,
1002
+ "learning_rate": 4.3509542874030986e-05,
1003
+ "loss": 0.36108903884887694,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 0.49861323833465576,
1009
+ "learning_rate": 4.27931630862194e-05,
1010
+ "loss": 0.4754959583282471,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.2883014678955078,
1016
+ "learning_rate": 4.207829100636318e-05,
1017
+ "loss": 0.38763861656188964,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 0.5406184792518616,
1023
+ "learning_rate": 4.1365076189438365e-05,
1024
+ "loss": 0.39906909465789797,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 0.8991988301277161,
1030
+ "learning_rate": 4.065366784371289e-05,
1031
+ "loss": 0.418521785736084,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 0.5573598742485046,
1037
+ "learning_rate": 3.994421479953153e-05,
1038
+ "loss": 0.43770723342895507,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 0.7751765847206116,
1044
+ "learning_rate": 3.9236865478179685e-05,
1045
+ "loss": 0.36180825233459474,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 0.5557352304458618,
1051
+ "learning_rate": 3.853176786083298e-05,
1052
+ "loss": 0.4191638946533203,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 1.012902021408081,
1058
+ "learning_rate": 3.7829069457598825e-05,
1059
+ "loss": 0.43454732894897463,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 0.8307779431343079,
1065
+ "learning_rate": 3.7128917276656406e-05,
1066
+ "loss": 0.3497183799743652,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 0.6312240958213806,
1072
+ "learning_rate": 3.6431457793502e-05,
1073
+ "loss": 0.43721756935119627,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 0.9830090403556824,
1079
+ "learning_rate": 3.573683692030537e-05,
1080
+ "loss": 0.48459711074829104,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 0.4116923213005066,
1086
+ "learning_rate": 3.504519997538422e-05,
1087
+ "loss": 0.31733293533325196,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.0353782176971436,
1093
+ "learning_rate": 3.435669165280294e-05,
1094
+ "loss": 0.43854188919067383,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 0.7071158289909363,
1100
+ "learning_rate": 3.3671455992101666e-05,
1101
+ "loss": 0.4189001560211182,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.6019755601882935,
1107
+ "learning_rate": 3.298963634816263e-05,
1108
+ "loss": 0.7713140487670899,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 0.5350252389907837,
1114
+ "learning_rate": 3.231137536121954e-05,
1115
+ "loss": 0.3926787137985229,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 0.37985295057296753,
1121
+ "learning_rate": 3.1636814927016324e-05,
1122
+ "loss": 0.4014743328094482,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 0.6357853412628174,
1128
+ "learning_rate": 3.096609616712207e-05,
1129
+ "loss": 0.4581602096557617,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 0.5001391768455505,
1135
+ "learning_rate": 3.0299359399407434e-05,
1136
+ "loss": 0.3605159044265747,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 0.9979525804519653,
1142
+ "learning_rate": 2.963674410868953e-05,
1143
+ "loss": 0.5704981803894043,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 0.6091923713684082,
1149
+ "learning_rate": 2.8978388917550935e-05,
1150
+ "loss": 0.3738471269607544,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 0.4976308345794678,
1156
+ "learning_rate": 2.8324431557339105e-05,
1157
+ "loss": 0.43538880348205566,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 0.9942322969436646,
1163
+ "learning_rate": 2.7675008839352314e-05,
1164
+ "loss": 0.3554206609725952,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 0.6866982579231262,
1170
+ "learning_rate": 2.703025662621793e-05,
1171
+ "loss": 0.4553103446960449,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 0.89406818151474,
1177
+ "learning_rate": 2.6390309803469326e-05,
1178
+ "loss": 0.43181838989257815,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 0.5093903541564941,
1184
+ "learning_rate": 2.5755302251327096e-05,
1185
+ "loss": 0.4385825157165527,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 0.5549860596656799,
1191
+ "learning_rate": 2.5125366816690555e-05,
1192
+ "loss": 0.329516339302063,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.4509713649749756,
1198
+ "learning_rate": 2.450063528534558e-05,
1199
+ "loss": 0.3463990926742554,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 0.5200580358505249,
1205
+ "learning_rate": 2.3881238354394236e-05,
1206
+ "loss": 0.40448341369628904,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.1131764650344849,
1212
+ "learning_rate": 2.32673056049123e-05,
1213
+ "loss": 0.426401424407959,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 0.7889172434806824,
1219
+ "learning_rate": 2.2658965474840262e-05,
1220
+ "loss": 0.38459455966949463,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 0.517167866230011,
1226
+ "learning_rate": 2.205634523211335e-05,
1227
+ "loss": 0.36908698081970215,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 0.5234260559082031,
1233
+ "learning_rate": 2.1459570948036485e-05,
1234
+ "loss": 0.3863369941711426,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 0.7236451506614685,
1240
+ "learning_rate": 2.086876747090945e-05,
1241
+ "loss": 0.40941739082336426,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 0.4438631236553192,
1247
+ "learning_rate": 2.0284058399908045e-05,
1248
+ "loss": 0.34200191497802734,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 0.39247673749923706,
1254
+ "learning_rate": 1.9705566059226448e-05,
1255
+ "loss": 0.4146144390106201,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1701165437698364,
1261
+ "learning_rate": 1.9133411472486324e-05,
1262
+ "loss": 0.4748171329498291,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 0.5402036309242249,
1268
+ "learning_rate": 1.8567714337418092e-05,
1269
+ "loss": 0.3901249170303345,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 0.6763427257537842,
1275
+ "learning_rate": 1.8008593000819518e-05,
1276
+ "loss": 0.4359886646270752,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 0.5417786240577698,
1282
+ "learning_rate": 1.745616443379699e-05,
1283
+ "loss": 0.39662876129150393,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 0.7444621920585632,
1289
+ "learning_rate": 1.691054420729452e-05,
1290
+ "loss": 0.3708503723144531,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 0.7745838165283203,
1296
+ "learning_rate": 1.6371846467915603e-05,
1297
+ "loss": 0.3678457736968994,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 0.4944072961807251,
1303
+ "learning_rate": 1.5840183914043338e-05,
1304
+ "loss": 0.3787250280380249,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 0.7562234997749329,
1310
+ "learning_rate": 1.5315667772263224e-05,
1311
+ "loss": 0.4543753623962402,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 0.5324202179908752,
1317
+ "learning_rate": 1.4798407774093954e-05,
1318
+ "loss": 0.34689984321594236,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 0.6181265711784363,
1324
+ "learning_rate": 1.4288512133031073e-05,
1325
+ "loss": 0.396760082244873,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 0.5689008831977844,
1331
+ "learning_rate": 1.3786087521908103e-05,
1332
+ "loss": 0.3692843198776245,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.631606936454773,
1338
+ "learning_rate": 1.3291239050580084e-05,
1339
+ "loss": 0.683953332901001,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 0.7201916575431824,
1345
+ "learning_rate": 1.2804070243933996e-05,
1346
+ "loss": 0.37404820919036863,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 0.5721992254257202,
1352
+ "learning_rate": 1.2324683020230798e-05,
1353
+ "loss": 0.7489796638488769,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 0.7852861881256104,
1359
+ "learning_rate": 1.1853177669783643e-05,
1360
+ "loss": 0.4024354457855225,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.5761847496032715,
1366
+ "learning_rate": 1.1389652833976577e-05,
1367
+ "loss": 0.4159713268280029,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 0.4608430564403534,
1373
+ "learning_rate": 1.0934205484628284e-05,
1374
+ "loss": 0.38759047985076905,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 0.4075900614261627,
1380
+ "learning_rate": 1.0486930903705094e-05,
1381
+ "loss": 0.38979687690734866,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 0.6070588231086731,
1387
+ "learning_rate": 1.0047922663387349e-05,
1388
+ "loss": 0.43787498474121095,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 0.5205054879188538,
1394
+ "learning_rate": 9.617272606493844e-06,
1395
+ "loss": 0.39637553691864014,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 0.8867539763450623,
1401
+ "learning_rate": 9.195070827267632e-06,
1402
+ "loss": 0.4684460639953613,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.7465023398399353,
1408
+ "learning_rate": 8.781405652527974e-06,
1409
+ "loss": 0.3949731826782227,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.6426427960395813,
1415
+ "learning_rate": 8.376363623191791e-06,
1416
+ "loss": 0.4073126792907715,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 0.6218783259391785,
1422
+ "learning_rate": 7.980029476168944e-06,
1423
+ "loss": 0.43186631202697756,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 0.6543543934822083,
1429
+ "learning_rate": 7.59248612663479e-06,
1430
+ "loss": 0.35069923400878905,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 0.8917542099952698,
1436
+ "learning_rate": 7.213814650683887e-06,
1437
+ "loss": 0.47125835418701173,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.4148687422275543,
1443
+ "learning_rate": 6.844094268368484e-06,
1444
+ "loss": 0.38721065521240233,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 0.6250177621841431,
1450
+ "learning_rate": 6.483402327125232e-06,
1451
+ "loss": 0.44888858795166015,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 0.4487530291080475,
1457
+ "learning_rate": 6.131814285593707e-06,
1458
+ "loss": 0.37155201435089114,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 0.450059711933136,
1464
+ "learning_rate": 5.7894036978301035e-06,
1465
+ "loss": 0.44730415344238283,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.858016312122345,
1471
+ "learning_rate": 5.4562421979192925e-06,
1472
+ "loss": 0.4445347309112549,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 0.5695084929466248,
1478
+ "learning_rate": 5.132399484988615e-06,
1479
+ "loss": 0.4245354175567627,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.4118499159812927,
1485
+ "learning_rate": 4.817943308626488e-06,
1486
+ "loss": 0.39943842887878417,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 0.7165275812149048,
1492
+ "learning_rate": 4.512939454708776e-06,
1493
+ "loss": 0.43923196792602537,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 0.5231788754463196,
1499
+ "learning_rate": 4.2174517316361075e-06,
1500
+ "loss": 0.41583733558654784,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 0.4489861726760864,
1506
+ "learning_rate": 3.93154195698478e-06,
1507
+ "loss": 0.4420301914215088,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 0.7916709184646606,
1513
+ "learning_rate": 3.655269944574202e-06,
1514
+ "loss": 0.415742826461792,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.1566158533096313,
1520
+ "learning_rate": 3.388693491953576e-06,
1521
+ "loss": 0.4660985469818115,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 0.5106304287910461,
1527
+ "learning_rate": 3.1318683683102754e-06,
1528
+ "loss": 0.36864919662475587,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 0.6080639362335205,
1534
+ "learning_rate": 2.8848483028026895e-06,
1535
+ "loss": 0.44466662406921387,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 0.5853272080421448,
1541
+ "learning_rate": 2.647684973319814e-06,
1542
+ "loss": 0.3511647701263428,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 0.41179347038269043,
1548
+ "learning_rate": 2.4204279956698995e-06,
1549
+ "loss": 0.4063854217529297,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 0.5927942991256714,
1555
+ "learning_rate": 2.2031249132006547e-06,
1556
+ "loss": 0.42436866760253905,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.7885688543319702,
1562
+ "learning_rate": 1.9958211868528474e-06,
1563
+ "loss": 0.3807806968688965,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 0.5567767024040222,
1569
+ "learning_rate": 1.7985601856496947e-06,
1570
+ "loss": 0.4070634365081787,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 0.7330124378204346,
1576
+ "learning_rate": 1.6113831776238097e-06,
1577
+ "loss": 0.41397967338562014,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 0.45662936568260193,
1583
+ "learning_rate": 1.4343293211837195e-06,
1584
+ "loss": 0.3888988494873047,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 0.48038941621780396,
1590
+ "learning_rate": 1.2674356569217282e-06,
1591
+ "loss": 0.3652131795883179,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 0.6213480830192566,
1597
+ "learning_rate": 1.1107370998648058e-06,
1598
+ "loss": 0.3998314142227173,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.7687107920646667,
1604
+ "learning_rate": 9.64266432170219e-07,
1605
+ "loss": 0.45170745849609373,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 0.6509118676185608,
1611
+ "learning_rate": 8.280542962673166e-07,
1612
+ "loss": 0.42131795883178713,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 0.6040301322937012,
1618
+ "learning_rate": 7.021291884469649e-07,
1619
+ "loss": 0.3819763422012329,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.8355879783630371,
1625
+ "learning_rate": 5.865174528999706e-07,
1626
+ "loss": 0.42635231018066405,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 0.548719048500061,
1632
+ "learning_rate": 4.812432762057673e-07,
1633
+ "loss": 0.35939807891845704,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 0.5332063436508179,
1639
+ "learning_rate": 3.8632868227242435e-07,
1640
+ "loss": 0.4174611568450928,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 0.5104030966758728,
1646
+ "learning_rate": 3.017935277291495e-07,
1647
+ "loss": 0.35938918590545654,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.0479686260223389,
1653
+ "learning_rate": 2.2765549777217788e-07,
1654
+ "loss": 0.353275465965271,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 0.6005430817604065,
1660
+ "learning_rate": 1.639301024649209e-07,
1661
+ "loss": 0.44223575592041015,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.6831442713737488,
1667
+ "learning_rate": 1.1063067349318946e-07,
1668
+ "loss": 0.36453771591186523,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 0.6081545948982239,
1674
+ "learning_rate": 6.776836137615262e-08,
1675
+ "loss": 0.35844419002532957,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.4227214455604553,
1681
+ "learning_rate": 3.535213313355934e-08,
1682
+ "loss": 0.399729585647583,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.9676080942153931,
1688
+ "learning_rate": 1.3388770409805685e-08,
1689
+ "loss": 0.43811402320861814,
1690
+ "step": 1200
1691
+ }
1692
+ ],
1693
+ "logging_steps": 5,
1694
+ "max_steps": 1207,
1695
+ "num_input_tokens_seen": 0,
1696
+ "num_train_epochs": 1,
1697
+ "save_steps": 200,
1698
+ "stateful_callbacks": {
1699
+ "TrainerControl": {
1700
+ "args": {
1701
+ "should_epoch_stop": false,
1702
+ "should_evaluate": false,
1703
+ "should_log": false,
1704
+ "should_save": true,
1705
+ "should_training_stop": false
1706
+ },
1707
+ "attributes": {}
1708
+ }
1709
+ },
1710
+ "total_flos": 2.518258444938741e+17,
1711
+ "train_batch_size": 1,
1712
+ "trial_name": null,
1713
+ "trial_params": null
1714
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- macro render_content(content) -%}
2
+ {%- if content is string -%}
3
+ {{- content -}}
4
+ {%- else -%}
5
+ {%- for item in content -%}
6
+ {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
7
+ {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
8
+ {%- elif item.type == 'text' -%}
9
+ {{- item.text -}}
10
+ {%- endif -%}
11
+ {%- endfor -%}
12
+ {%- endif -%}
13
+ {%- endmacro -%}
14
+ {%- if tools %}
15
+ {{- '<|im_start|>system\n' }}
16
+ {%- if messages[0].role == 'system' %}
17
+ {{- render_content(messages[0].content) + '\n\n' }}
18
+ {%- else %}
19
+ {{- 'You are a helpful assistant.\n\n' }}
20
+ {%- endif %}
21
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
22
+ {%- for tool in tools %}
23
+ {{- "\n" }}
24
+ {{- tool | tojson }}
25
+ {%- endfor %}
26
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
27
+ {%- else %}
28
+ {%- if messages[0].role == 'system' %}
29
+ {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
30
+ {%- else %}
31
+ {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
32
+ {%- endif %}
33
+ {%- endif %}
34
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
35
+ {%- for message in messages[::-1] %}
36
+ {%- set index = (messages|length - 1) - loop.index0 %}
37
+ {%- set content = render_content(message.content) %}
38
+ {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
39
+ {%- set ns.multi_step_tool = false %}
40
+ {%- set ns.last_query_index = index %}
41
+ {%- endif %}
42
+ {%- endfor %}
43
+ {%- for message in messages %}
44
+ {%- set content = render_content(message.content) %}
45
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
47
+ {%- elif message.role == "assistant" %}
48
+ {%- set reasoning_content = '' %}
49
+ {%- if message.reasoning_content is string %}
50
+ {%- set reasoning_content = message.reasoning_content %}
51
+ {%- else %}
52
+ {%- if '</think>' in content %}
53
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
54
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
55
+ {%- endif %}
56
+ {%- endif %}
57
+ {%- if loop.index0 > ns.last_query_index %}
58
+ {%- if loop.last or (not loop.last and reasoning_content) %}
59
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
60
+ {%- else %}
61
+ {{- '<|im_start|>' + message.role + '\n' + content }}
62
+ {%- endif %}
63
+ {%- else %}
64
+ {{- '<|im_start|>' + message.role + '\n' + content }}
65
+ {%- endif %}
66
+ {%- if message.tool_calls %}
67
+ {%- for tool_call in message.tool_calls %}
68
+ {%- if (loop.first and content) or (not loop.first) %}
69
+ {{- '\n' }}
70
+ {%- endif %}
71
+ {%- if tool_call.function %}
72
+ {%- set tool_call = tool_call.function %}
73
+ {%- endif %}
74
+ {{- '<tool_call>\n{"name": "' }}
75
+ {{- tool_call.name }}
76
+ {{- '", "arguments": ' }}
77
+ {%- if tool_call.arguments is string %}
78
+ {{- tool_call.arguments }}
79
+ {%- else %}
80
+ {{- tool_call.arguments | tojson }}
81
+ {%- endif %}
82
+ {{- '}\n</tool_call>' }}
83
+ {%- endfor %}
84
+ {%- endif %}
85
+ {{- '<|im_end|>\n' }}
86
+ {%- elif message.role == "tool" %}
87
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
88
+ {{- '<|im_start|>user' }}
89
+ {%- endif %}
90
+ {{- '\n<tool_response>\n' }}
91
+ {{- content }}
92
+ {{- '\n</tool_response>' }}
93
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
94
+ {{- '<|im_end|>\n' }}
95
+ {%- endif %}
96
+ {%- endif %}
97
+ {%- endfor %}
98
+ {%- if add_generation_prompt %}
99
+ {{- '<|im_start|>assistant\n' }}
100
+ {%- if enable_thinking is defined and enable_thinking is false %}
101
+ {{- '<think>\n\n</think>\n\n' }}
102
+ {%- endif %}
103
+ {%- endif %}
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/config.json ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_plane_indices": [
171
+ 3,
172
+ 11,
173
+ 19,
174
+ 27,
175
+ 35,
176
+ 43,
177
+ 51,
178
+ 59
179
+ ],
180
+ "speaker_plane_order": "interleaved",
181
+ "speaker_rotary_dim": 16,
182
+ "speaker_rotary_inference_seed": 42,
183
+ "speaker_rotary_std": 0.02,
184
+ "speaker_sample_rate": 16000,
185
+ "speaker_tfmel_args": {
186
+ "do_preemph": true,
187
+ "do_spec_aug": false,
188
+ "eps": 1e-08,
189
+ "f_max": 7600,
190
+ "f_min": 20,
191
+ "hop_length": 160,
192
+ "n_fft": 512,
193
+ "n_mels": 72,
194
+ "norm_signal": true,
195
+ "win_length": 400
196
+ },
197
+ "speaker_window_seconds": 2.0,
198
+ "text_config": {
199
+ "attention_bias": false,
200
+ "attention_dropout": 0.0,
201
+ "bos_token_id": null,
202
+ "dtype": "bfloat16",
203
+ "eos_token_id": null,
204
+ "head_dim": 128,
205
+ "hidden_act": "silu",
206
+ "hidden_size": 1024,
207
+ "initializer_range": 0.02,
208
+ "intermediate_size": 3072,
209
+ "layer_types": [
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention",
227
+ "full_attention",
228
+ "full_attention",
229
+ "full_attention",
230
+ "full_attention",
231
+ "full_attention",
232
+ "full_attention",
233
+ "full_attention",
234
+ "full_attention",
235
+ "full_attention",
236
+ "full_attention",
237
+ "full_attention"
238
+ ],
239
+ "max_position_embeddings": 131072,
240
+ "max_window_layers": 28,
241
+ "model_type": "qwen3",
242
+ "num_attention_heads": 16,
243
+ "num_hidden_layers": 28,
244
+ "num_key_value_heads": 8,
245
+ "pad_token_id": 151643,
246
+ "rms_norm_eps": 1e-06,
247
+ "rope_parameters": {
248
+ "rope_theta": 1000000,
249
+ "rope_type": "default"
250
+ },
251
+ "sliding_window": null,
252
+ "tie_word_embeddings": true,
253
+ "use_cache": false,
254
+ "use_sliding_window": false,
255
+ "vocab_size": 151936
256
+ },
257
+ "tie_word_embeddings": true,
258
+ "transformers_version": "5.14.0",
259
+ "use_cache": false
260
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/checkpoint-1207/trainer_state.json ADDED
@@ -0,0 +1,1721 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1207,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 20.629674911499023,
15
+ "learning_rate": 3.3057851239669424e-06,
16
+ "loss": 0.8916114807128906,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 12.812925338745117,
22
+ "learning_rate": 7.43801652892562e-06,
23
+ "loss": 0.558007287979126,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 7.200536251068115,
29
+ "learning_rate": 1.1570247933884299e-05,
30
+ "loss": 0.6662204742431641,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 1.363626480102539,
36
+ "learning_rate": 1.5702479338842978e-05,
37
+ "loss": 0.45300092697143557,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 3.3842880725860596,
43
+ "learning_rate": 1.9834710743801653e-05,
44
+ "loss": 0.494948148727417,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 1.388489842414856,
50
+ "learning_rate": 2.396694214876033e-05,
51
+ "loss": 0.4781633377075195,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 0.5891152024269104,
57
+ "learning_rate": 2.809917355371901e-05,
58
+ "loss": 0.48348569869995117,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 0.9562617540359497,
64
+ "learning_rate": 3.2231404958677685e-05,
65
+ "loss": 0.4969147205352783,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 0.6433483958244324,
71
+ "learning_rate": 3.6363636363636364e-05,
72
+ "loss": 0.4836277008056641,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 0.898301362991333,
78
+ "learning_rate": 4.049586776859504e-05,
79
+ "loss": 0.4123377799987793,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 0.8125119805335999,
85
+ "learning_rate": 4.462809917355372e-05,
86
+ "loss": 0.45714097023010253,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 0.5890400409698486,
92
+ "learning_rate": 4.87603305785124e-05,
93
+ "loss": 0.4525331497192383,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 0.6006086468696594,
99
+ "learning_rate": 5.289256198347108e-05,
100
+ "loss": 0.3964966297149658,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 1.017815113067627,
106
+ "learning_rate": 5.702479338842975e-05,
107
+ "loss": 0.37221901416778563,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 0.9333603978157043,
113
+ "learning_rate": 6.115702479338842e-05,
114
+ "loss": 0.4029701232910156,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 1.0475515127182007,
120
+ "learning_rate": 6.528925619834711e-05,
121
+ "loss": 0.45854949951171875,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 1.0967167615890503,
127
+ "learning_rate": 6.94214876033058e-05,
128
+ "loss": 0.455504035949707,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 0.7258841395378113,
134
+ "learning_rate": 7.355371900826447e-05,
135
+ "loss": 0.7798455238342286,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 1.0904923677444458,
141
+ "learning_rate": 7.768595041322314e-05,
142
+ "loss": 0.46855831146240234,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 0.924197256565094,
148
+ "learning_rate": 8.181818181818183e-05,
149
+ "loss": 0.4880524635314941,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 1.9587433338165283,
155
+ "learning_rate": 8.595041322314051e-05,
156
+ "loss": 0.463060998916626,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 0.6440144181251526,
162
+ "learning_rate": 9.008264462809917e-05,
163
+ "loss": 0.48012351989746094,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 0.8251569867134094,
169
+ "learning_rate": 9.421487603305785e-05,
170
+ "loss": 0.4483522415161133,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 0.8789721727371216,
176
+ "learning_rate": 9.834710743801654e-05,
177
+ "loss": 0.4580700397491455,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 0.8313543796539307,
183
+ "learning_rate": 9.999811713194481e-05,
184
+ "loss": 0.46596560478210447,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 0.6774608492851257,
190
+ "learning_rate": 9.99866112295902e-05,
191
+ "loss": 0.4806799411773682,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 0.6783014535903931,
197
+ "learning_rate": 9.996464786686644e-05,
198
+ "loss": 0.4560887336730957,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 0.7160524129867554,
204
+ "learning_rate": 9.993223163862385e-05,
205
+ "loss": 0.4927361011505127,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 0.5270054340362549,
211
+ "learning_rate": 9.988936932650681e-05,
212
+ "loss": 0.42853679656982424,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 1.0818812847137451,
218
+ "learning_rate": 9.983606989753509e-05,
219
+ "loss": 0.5378652572631836,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 0.968137800693512,
225
+ "learning_rate": 9.977234450222783e-05,
226
+ "loss": 0.48653225898742675,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 0.8479903936386108,
232
+ "learning_rate": 9.969820647227086e-05,
233
+ "loss": 0.5200669765472412,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 0.7051109075546265,
239
+ "learning_rate": 9.961367131772759e-05,
240
+ "loss": 0.4588489532470703,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 2.205345630645752,
246
+ "learning_rate": 9.951875672379424e-05,
247
+ "loss": 0.5904168128967285,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 0.8751126527786255,
253
+ "learning_rate": 9.941348254710003e-05,
254
+ "loss": 0.43442707061767577,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 0.671466052532196,
260
+ "learning_rate": 9.929787081155304e-05,
261
+ "loss": 0.3809401512145996,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 1.235303282737732,
267
+ "learning_rate": 9.91719457037327e-05,
268
+ "loss": 0.4351661205291748,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 0.5449215769767761,
274
+ "learning_rate": 9.903573356782977e-05,
275
+ "loss": 0.41109681129455566,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 0.6575606465339661,
281
+ "learning_rate": 9.888926290013521e-05,
282
+ "loss": 0.46851186752319335,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 0.528086245059967,
288
+ "learning_rate": 9.873256434307828e-05,
289
+ "loss": 0.40171051025390625,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 0.44805943965911865,
295
+ "learning_rate": 9.856567067881628e-05,
296
+ "loss": 0.3575094223022461,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 0.6913977861404419,
302
+ "learning_rate": 9.838861682237618e-05,
303
+ "loss": 0.4208680152893066,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 0.6070751547813416,
309
+ "learning_rate": 9.820143981435031e-05,
310
+ "loss": 0.4100965976715088,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 0.7453505396842957,
316
+ "learning_rate": 9.800417881314716e-05,
317
+ "loss": 0.5092792510986328,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 0.718826949596405,
323
+ "learning_rate": 9.779687508679935e-05,
324
+ "loss": 0.5120072841644288,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 0.8064695596694946,
330
+ "learning_rate": 9.75795720043301e-05,
331
+ "loss": 0.43071684837341306,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 0.6657348275184631,
337
+ "learning_rate": 9.73523150266802e-05,
338
+ "loss": 0.43650312423706056,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 0.6227285861968994,
344
+ "learning_rate": 9.711515169719732e-05,
345
+ "loss": 0.4965176582336426,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 0.6354798078536987,
351
+ "learning_rate": 9.686813163168973e-05,
352
+ "loss": 0.4135477542877197,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 0.8802199959754944,
358
+ "learning_rate": 9.661130650804643e-05,
359
+ "loss": 0.4657131671905518,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 0.6120206713676453,
365
+ "learning_rate": 9.634473005542579e-05,
366
+ "loss": 0.5105055332183838,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 0.6608315706253052,
372
+ "learning_rate": 9.606845804301524e-05,
373
+ "loss": 0.48976764678955076,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 0.6136434078216553,
379
+ "learning_rate": 9.578254826836391e-05,
380
+ "loss": 0.39600679874420164,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 0.7732479572296143,
386
+ "learning_rate": 9.548706054529124e-05,
387
+ "loss": 0.43799166679382323,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 0.8353448510169983,
393
+ "learning_rate": 9.518205669137352e-05,
394
+ "loss": 0.39786520004272463,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 0.5929117798805237,
400
+ "learning_rate": 9.486760051501139e-05,
401
+ "loss": 0.4511585235595703,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 0.7609494924545288,
407
+ "learning_rate": 9.454375780208071e-05,
408
+ "loss": 0.460142993927002,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 0.4482679069042206,
414
+ "learning_rate": 9.421059630216991e-05,
415
+ "loss": 0.42333049774169923,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 0.5880385637283325,
421
+ "learning_rate": 9.38681857144063e-05,
422
+ "loss": 0.39257287979125977,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 0.4800879955291748,
428
+ "learning_rate": 9.351659767287478e-05,
429
+ "loss": 0.46072850227355955,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 0.5711155533790588,
435
+ "learning_rate": 9.315590573163153e-05,
436
+ "loss": 0.4340324401855469,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.6473472118377686,
442
+ "learning_rate": 9.278618534931611e-05,
443
+ "loss": 0.4806147575378418,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 0.9981988668441772,
449
+ "learning_rate": 9.240751387336522e-05,
450
+ "loss": 0.45314898490905764,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 0.9140734076499939,
456
+ "learning_rate": 9.201997052383106e-05,
457
+ "loss": 0.48151440620422364,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 0.7532357573509216,
463
+ "learning_rate": 9.162363637680822e-05,
464
+ "loss": 0.4129056453704834,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 0.9811023473739624,
470
+ "learning_rate": 9.121859434747204e-05,
471
+ "loss": 0.484523344039917,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.4147282838821411,
477
+ "learning_rate": 9.080492917273237e-05,
478
+ "loss": 0.5427152633666992,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 1.0163606405258179,
484
+ "learning_rate": 9.038272739350618e-05,
485
+ "loss": 0.4134209156036377,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 0.6049103140830994,
491
+ "learning_rate": 8.995207733661267e-05,
492
+ "loss": 0.40175676345825195,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 0.7045316696166992,
498
+ "learning_rate": 8.951306909629492e-05,
499
+ "loss": 0.40282230377197265,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 0.6261049509048462,
505
+ "learning_rate": 8.906579451537171e-05,
506
+ "loss": 0.3981405973434448,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 1.287496566772461,
512
+ "learning_rate": 8.861034716602343e-05,
513
+ "loss": 0.4517494201660156,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 0.6869074702262878,
519
+ "learning_rate": 8.814682233021636e-05,
520
+ "loss": 0.44563779830932615,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 0.6493797302246094,
526
+ "learning_rate": 8.76753169797692e-05,
527
+ "loss": 0.4556128978729248,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 0.5839359760284424,
533
+ "learning_rate": 8.719592975606601e-05,
534
+ "loss": 0.43351993560791013,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 0.3983668088912964,
540
+ "learning_rate": 8.670876094941991e-05,
541
+ "loss": 0.4440763473510742,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 0.6596688032150269,
547
+ "learning_rate": 8.62139124780919e-05,
548
+ "loss": 0.5530457973480225,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 0.6001258492469788,
554
+ "learning_rate": 8.571148786696894e-05,
555
+ "loss": 0.4752801418304443,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 0.405813604593277,
561
+ "learning_rate": 8.520159222590605e-05,
562
+ "loss": 0.38340303897857664,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 0.8015041351318359,
568
+ "learning_rate": 8.468433222773677e-05,
569
+ "loss": 0.4637146949768066,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 0.6381627321243286,
575
+ "learning_rate": 8.415981608595667e-05,
576
+ "loss": 0.42714495658874513,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 0.5829190015792847,
582
+ "learning_rate": 8.36281535320844e-05,
583
+ "loss": 0.3956311702728271,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 0.8698112964630127,
589
+ "learning_rate": 8.30894557927055e-05,
590
+ "loss": 0.4846199989318848,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 0.6768237352371216,
596
+ "learning_rate": 8.254383556620299e-05,
597
+ "loss": 0.47880539894104,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 0.5710030198097229,
603
+ "learning_rate": 8.199140699918049e-05,
604
+ "loss": 0.4479952335357666,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 0.6229125261306763,
610
+ "learning_rate": 8.143228566258192e-05,
611
+ "loss": 0.433013916015625,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 0.6174870133399963,
617
+ "learning_rate": 8.08665885275137e-05,
618
+ "loss": 0.4820831298828125,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.9886339902877808,
624
+ "learning_rate": 8.029443394077356e-05,
625
+ "loss": 0.4312156677246094,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 0.7721105217933655,
631
+ "learning_rate": 7.971594160009194e-05,
632
+ "loss": 0.4745286464691162,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 0.8215830326080322,
638
+ "learning_rate": 7.913123252909056e-05,
639
+ "loss": 0.43494420051574706,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 0.8584961295127869,
645
+ "learning_rate": 7.854042905196354e-05,
646
+ "loss": 0.46215004920959474,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 0.483641117811203,
652
+ "learning_rate": 7.794365476788667e-05,
653
+ "loss": 0.37451136112213135,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 0.567133903503418,
659
+ "learning_rate": 7.734103452515975e-05,
660
+ "loss": 0.4840177059173584,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 0.6834204196929932,
666
+ "learning_rate": 7.67326943950877e-05,
667
+ "loss": 0.46247282028198244,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 0.8395087122917175,
673
+ "learning_rate": 7.611876164560578e-05,
674
+ "loss": 0.36043124198913573,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 0.3639385998249054,
680
+ "learning_rate": 7.549936471465444e-05,
681
+ "loss": 0.42089052200317384,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 0.4404556453227997,
687
+ "learning_rate": 7.487463318330946e-05,
688
+ "loss": 0.3542492866516113,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 0.6046218276023865,
694
+ "learning_rate": 7.42446977486729e-05,
695
+ "loss": 0.38021628856658934,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.6285887956619263,
701
+ "learning_rate": 7.360969019653067e-05,
702
+ "loss": 0.6484775066375732,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 0.9363073110580444,
708
+ "learning_rate": 7.296974337378209e-05,
709
+ "loss": 0.41137142181396485,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 0.4143483340740204,
715
+ "learning_rate": 7.23249911606477e-05,
716
+ "loss": 0.3924715518951416,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.2242801189422607,
722
+ "learning_rate": 7.167556844266091e-05,
723
+ "loss": 0.4372579574584961,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 0.49445945024490356,
729
+ "learning_rate": 7.102161108244907e-05,
730
+ "loss": 0.4041281700134277,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 0.5578067898750305,
736
+ "learning_rate": 7.036325589131048e-05,
737
+ "loss": 0.44759469032287597,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 0.6338609457015991,
743
+ "learning_rate": 6.970064060059258e-05,
744
+ "loss": 0.3942832946777344,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 0.6305315494537354,
750
+ "learning_rate": 6.903390383287794e-05,
751
+ "loss": 0.3744609832763672,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 0.6225669980049133,
757
+ "learning_rate": 6.83631850729837e-05,
758
+ "loss": 0.4644033432006836,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 0.4533715844154358,
764
+ "learning_rate": 6.768862463878047e-05,
765
+ "loss": 0.43013477325439453,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 0.668293297290802,
771
+ "learning_rate": 6.701036365183737e-05,
772
+ "loss": 0.5000359058380127,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 0.6384305357933044,
778
+ "learning_rate": 6.632854400789835e-05,
779
+ "loss": 0.4234946250915527,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 0.35919633507728577,
785
+ "learning_rate": 6.564330834719708e-05,
786
+ "loss": 0.4122346878051758,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 0.5707134008407593,
792
+ "learning_rate": 6.495480002461577e-05,
793
+ "loss": 0.4068869113922119,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.7894991636276245,
799
+ "learning_rate": 6.426316307969465e-05,
800
+ "loss": 0.4764272689819336,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 0.7096588611602783,
806
+ "learning_rate": 6.3568542206498e-05,
807
+ "loss": 0.40894289016723634,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 0.6279922723770142,
813
+ "learning_rate": 6.28710827233436e-05,
814
+ "loss": 0.4551478385925293,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 0.9392326474189758,
820
+ "learning_rate": 6.21709305424012e-05,
821
+ "loss": 0.42044363021850584,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 0.8099265098571777,
827
+ "learning_rate": 6.146823213916702e-05,
828
+ "loss": 0.421197509765625,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 0.407980740070343,
834
+ "learning_rate": 6.076313452182033e-05,
835
+ "loss": 0.45879392623901366,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 0.6435403823852539,
841
+ "learning_rate": 6.005578520046848e-05,
842
+ "loss": 0.7919346332550049,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 0.5578521490097046,
848
+ "learning_rate": 5.934633215628711e-05,
849
+ "loss": 0.4143537044525146,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 0.5052011013031006,
855
+ "learning_rate": 5.863492381056164e-05,
856
+ "loss": 0.38072867393493653,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 0.6519150733947754,
862
+ "learning_rate": 5.792170899363683e-05,
863
+ "loss": 0.4297952175140381,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 0.9843361377716064,
869
+ "learning_rate": 5.72068369137806e-05,
870
+ "loss": 0.3838667869567871,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 0.7498537302017212,
876
+ "learning_rate": 5.649045712596903e-05,
877
+ "loss": 0.40416569709777833,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 0.6527976393699646,
883
+ "learning_rate": 5.577271950059849e-05,
884
+ "loss": 0.4134161949157715,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 0.8321830630302429,
890
+ "learning_rate": 5.505377419213214e-05,
891
+ "loss": 0.43797922134399414,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 0.6492611169815063,
897
+ "learning_rate": 5.433377160768669e-05,
898
+ "loss": 0.4532886028289795,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 0.6094037294387817,
904
+ "learning_rate": 5.3612862375566574e-05,
905
+ "loss": 0.43567543029785155,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 0.551053524017334,
911
+ "learning_rate": 5.2891197313751606e-05,
912
+ "loss": 0.4400018692016602,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 0.6376934051513672,
918
+ "learning_rate": 5.216892739834519e-05,
919
+ "loss": 0.42303104400634767,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 0.9007910490036011,
925
+ "learning_rate": 5.144620373198916e-05,
926
+ "loss": 0.45666775703430174,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 0.650941014289856,
932
+ "learning_rate": 5.072317751225254e-05,
933
+ "loss": 0.4667072296142578,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 0.48627233505249023,
939
+ "learning_rate": 5e-05,
940
+ "loss": 0.3643670558929443,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 0.7920517325401306,
946
+ "learning_rate": 4.927682248774748e-05,
947
+ "loss": 0.44395933151245115,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 0.5550063252449036,
953
+ "learning_rate": 4.855379626801085e-05,
954
+ "loss": 0.36930572986602783,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 0.6230601668357849,
960
+ "learning_rate": 4.783107260165483e-05,
961
+ "loss": 0.4727597236633301,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 0.6426054835319519,
967
+ "learning_rate": 4.710880268624841e-05,
968
+ "loss": 0.460523796081543,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 0.5865359902381897,
974
+ "learning_rate": 4.6387137624433444e-05,
975
+ "loss": 0.4252904415130615,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 0.5453794598579407,
981
+ "learning_rate": 4.5666228392313315e-05,
982
+ "loss": 0.4039918422698975,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.0684459209442139,
988
+ "learning_rate": 4.494622580786788e-05,
989
+ "loss": 0.42879571914672854,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 0.8819708824157715,
995
+ "learning_rate": 4.422728049940151e-05,
996
+ "loss": 0.35405232906341555,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 0.5523223280906677,
1002
+ "learning_rate": 4.3509542874030986e-05,
1003
+ "loss": 0.36108903884887694,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 0.49861323833465576,
1009
+ "learning_rate": 4.27931630862194e-05,
1010
+ "loss": 0.4754959583282471,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.2883014678955078,
1016
+ "learning_rate": 4.207829100636318e-05,
1017
+ "loss": 0.38763861656188964,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 0.5406184792518616,
1023
+ "learning_rate": 4.1365076189438365e-05,
1024
+ "loss": 0.39906909465789797,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 0.8991988301277161,
1030
+ "learning_rate": 4.065366784371289e-05,
1031
+ "loss": 0.418521785736084,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 0.5573598742485046,
1037
+ "learning_rate": 3.994421479953153e-05,
1038
+ "loss": 0.43770723342895507,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 0.7751765847206116,
1044
+ "learning_rate": 3.9236865478179685e-05,
1045
+ "loss": 0.36180825233459474,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 0.5557352304458618,
1051
+ "learning_rate": 3.853176786083298e-05,
1052
+ "loss": 0.4191638946533203,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 1.012902021408081,
1058
+ "learning_rate": 3.7829069457598825e-05,
1059
+ "loss": 0.43454732894897463,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 0.8307779431343079,
1065
+ "learning_rate": 3.7128917276656406e-05,
1066
+ "loss": 0.3497183799743652,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 0.6312240958213806,
1072
+ "learning_rate": 3.6431457793502e-05,
1073
+ "loss": 0.43721756935119627,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 0.9830090403556824,
1079
+ "learning_rate": 3.573683692030537e-05,
1080
+ "loss": 0.48459711074829104,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 0.4116923213005066,
1086
+ "learning_rate": 3.504519997538422e-05,
1087
+ "loss": 0.31733293533325196,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.0353782176971436,
1093
+ "learning_rate": 3.435669165280294e-05,
1094
+ "loss": 0.43854188919067383,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 0.7071158289909363,
1100
+ "learning_rate": 3.3671455992101666e-05,
1101
+ "loss": 0.4189001560211182,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.6019755601882935,
1107
+ "learning_rate": 3.298963634816263e-05,
1108
+ "loss": 0.7713140487670899,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 0.5350252389907837,
1114
+ "learning_rate": 3.231137536121954e-05,
1115
+ "loss": 0.3926787137985229,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 0.37985295057296753,
1121
+ "learning_rate": 3.1636814927016324e-05,
1122
+ "loss": 0.4014743328094482,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 0.6357853412628174,
1128
+ "learning_rate": 3.096609616712207e-05,
1129
+ "loss": 0.4581602096557617,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 0.5001391768455505,
1135
+ "learning_rate": 3.0299359399407434e-05,
1136
+ "loss": 0.3605159044265747,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 0.9979525804519653,
1142
+ "learning_rate": 2.963674410868953e-05,
1143
+ "loss": 0.5704981803894043,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 0.6091923713684082,
1149
+ "learning_rate": 2.8978388917550935e-05,
1150
+ "loss": 0.3738471269607544,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 0.4976308345794678,
1156
+ "learning_rate": 2.8324431557339105e-05,
1157
+ "loss": 0.43538880348205566,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 0.9942322969436646,
1163
+ "learning_rate": 2.7675008839352314e-05,
1164
+ "loss": 0.3554206609725952,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 0.6866982579231262,
1170
+ "learning_rate": 2.703025662621793e-05,
1171
+ "loss": 0.4553103446960449,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 0.89406818151474,
1177
+ "learning_rate": 2.6390309803469326e-05,
1178
+ "loss": 0.43181838989257815,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 0.5093903541564941,
1184
+ "learning_rate": 2.5755302251327096e-05,
1185
+ "loss": 0.4385825157165527,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 0.5549860596656799,
1191
+ "learning_rate": 2.5125366816690555e-05,
1192
+ "loss": 0.329516339302063,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.4509713649749756,
1198
+ "learning_rate": 2.450063528534558e-05,
1199
+ "loss": 0.3463990926742554,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 0.5200580358505249,
1205
+ "learning_rate": 2.3881238354394236e-05,
1206
+ "loss": 0.40448341369628904,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.1131764650344849,
1212
+ "learning_rate": 2.32673056049123e-05,
1213
+ "loss": 0.426401424407959,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 0.7889172434806824,
1219
+ "learning_rate": 2.2658965474840262e-05,
1220
+ "loss": 0.38459455966949463,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 0.517167866230011,
1226
+ "learning_rate": 2.205634523211335e-05,
1227
+ "loss": 0.36908698081970215,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 0.5234260559082031,
1233
+ "learning_rate": 2.1459570948036485e-05,
1234
+ "loss": 0.3863369941711426,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 0.7236451506614685,
1240
+ "learning_rate": 2.086876747090945e-05,
1241
+ "loss": 0.40941739082336426,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 0.4438631236553192,
1247
+ "learning_rate": 2.0284058399908045e-05,
1248
+ "loss": 0.34200191497802734,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 0.39247673749923706,
1254
+ "learning_rate": 1.9705566059226448e-05,
1255
+ "loss": 0.4146144390106201,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1701165437698364,
1261
+ "learning_rate": 1.9133411472486324e-05,
1262
+ "loss": 0.4748171329498291,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 0.5402036309242249,
1268
+ "learning_rate": 1.8567714337418092e-05,
1269
+ "loss": 0.3901249170303345,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 0.6763427257537842,
1275
+ "learning_rate": 1.8008593000819518e-05,
1276
+ "loss": 0.4359886646270752,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 0.5417786240577698,
1282
+ "learning_rate": 1.745616443379699e-05,
1283
+ "loss": 0.39662876129150393,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 0.7444621920585632,
1289
+ "learning_rate": 1.691054420729452e-05,
1290
+ "loss": 0.3708503723144531,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 0.7745838165283203,
1296
+ "learning_rate": 1.6371846467915603e-05,
1297
+ "loss": 0.3678457736968994,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 0.4944072961807251,
1303
+ "learning_rate": 1.5840183914043338e-05,
1304
+ "loss": 0.3787250280380249,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 0.7562234997749329,
1310
+ "learning_rate": 1.5315667772263224e-05,
1311
+ "loss": 0.4543753623962402,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 0.5324202179908752,
1317
+ "learning_rate": 1.4798407774093954e-05,
1318
+ "loss": 0.34689984321594236,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 0.6181265711784363,
1324
+ "learning_rate": 1.4288512133031073e-05,
1325
+ "loss": 0.396760082244873,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 0.5689008831977844,
1331
+ "learning_rate": 1.3786087521908103e-05,
1332
+ "loss": 0.3692843198776245,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.631606936454773,
1338
+ "learning_rate": 1.3291239050580084e-05,
1339
+ "loss": 0.683953332901001,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 0.7201916575431824,
1345
+ "learning_rate": 1.2804070243933996e-05,
1346
+ "loss": 0.37404820919036863,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 0.5721992254257202,
1352
+ "learning_rate": 1.2324683020230798e-05,
1353
+ "loss": 0.7489796638488769,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 0.7852861881256104,
1359
+ "learning_rate": 1.1853177669783643e-05,
1360
+ "loss": 0.4024354457855225,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.5761847496032715,
1366
+ "learning_rate": 1.1389652833976577e-05,
1367
+ "loss": 0.4159713268280029,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 0.4608430564403534,
1373
+ "learning_rate": 1.0934205484628284e-05,
1374
+ "loss": 0.38759047985076905,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 0.4075900614261627,
1380
+ "learning_rate": 1.0486930903705094e-05,
1381
+ "loss": 0.38979687690734866,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 0.6070588231086731,
1387
+ "learning_rate": 1.0047922663387349e-05,
1388
+ "loss": 0.43787498474121095,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 0.5205054879188538,
1394
+ "learning_rate": 9.617272606493844e-06,
1395
+ "loss": 0.39637553691864014,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 0.8867539763450623,
1401
+ "learning_rate": 9.195070827267632e-06,
1402
+ "loss": 0.4684460639953613,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.7465023398399353,
1408
+ "learning_rate": 8.781405652527974e-06,
1409
+ "loss": 0.3949731826782227,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.6426427960395813,
1415
+ "learning_rate": 8.376363623191791e-06,
1416
+ "loss": 0.4073126792907715,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 0.6218783259391785,
1422
+ "learning_rate": 7.980029476168944e-06,
1423
+ "loss": 0.43186631202697756,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 0.6543543934822083,
1429
+ "learning_rate": 7.59248612663479e-06,
1430
+ "loss": 0.35069923400878905,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 0.8917542099952698,
1436
+ "learning_rate": 7.213814650683887e-06,
1437
+ "loss": 0.47125835418701173,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.4148687422275543,
1443
+ "learning_rate": 6.844094268368484e-06,
1444
+ "loss": 0.38721065521240233,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 0.6250177621841431,
1450
+ "learning_rate": 6.483402327125232e-06,
1451
+ "loss": 0.44888858795166015,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 0.4487530291080475,
1457
+ "learning_rate": 6.131814285593707e-06,
1458
+ "loss": 0.37155201435089114,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 0.450059711933136,
1464
+ "learning_rate": 5.7894036978301035e-06,
1465
+ "loss": 0.44730415344238283,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.858016312122345,
1471
+ "learning_rate": 5.4562421979192925e-06,
1472
+ "loss": 0.4445347309112549,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 0.5695084929466248,
1478
+ "learning_rate": 5.132399484988615e-06,
1479
+ "loss": 0.4245354175567627,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.4118499159812927,
1485
+ "learning_rate": 4.817943308626488e-06,
1486
+ "loss": 0.39943842887878417,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 0.7165275812149048,
1492
+ "learning_rate": 4.512939454708776e-06,
1493
+ "loss": 0.43923196792602537,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 0.5231788754463196,
1499
+ "learning_rate": 4.2174517316361075e-06,
1500
+ "loss": 0.41583733558654784,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 0.4489861726760864,
1506
+ "learning_rate": 3.93154195698478e-06,
1507
+ "loss": 0.4420301914215088,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 0.7916709184646606,
1513
+ "learning_rate": 3.655269944574202e-06,
1514
+ "loss": 0.415742826461792,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.1566158533096313,
1520
+ "learning_rate": 3.388693491953576e-06,
1521
+ "loss": 0.4660985469818115,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 0.5106304287910461,
1527
+ "learning_rate": 3.1318683683102754e-06,
1528
+ "loss": 0.36864919662475587,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 0.6080639362335205,
1534
+ "learning_rate": 2.8848483028026895e-06,
1535
+ "loss": 0.44466662406921387,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 0.5853272080421448,
1541
+ "learning_rate": 2.647684973319814e-06,
1542
+ "loss": 0.3511647701263428,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 0.41179347038269043,
1548
+ "learning_rate": 2.4204279956698995e-06,
1549
+ "loss": 0.4063854217529297,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 0.5927942991256714,
1555
+ "learning_rate": 2.2031249132006547e-06,
1556
+ "loss": 0.42436866760253905,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.7885688543319702,
1562
+ "learning_rate": 1.9958211868528474e-06,
1563
+ "loss": 0.3807806968688965,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 0.5567767024040222,
1569
+ "learning_rate": 1.7985601856496947e-06,
1570
+ "loss": 0.4070634365081787,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 0.7330124378204346,
1576
+ "learning_rate": 1.6113831776238097e-06,
1577
+ "loss": 0.41397967338562014,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 0.45662936568260193,
1583
+ "learning_rate": 1.4343293211837195e-06,
1584
+ "loss": 0.3888988494873047,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 0.48038941621780396,
1590
+ "learning_rate": 1.2674356569217282e-06,
1591
+ "loss": 0.3652131795883179,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 0.6213480830192566,
1597
+ "learning_rate": 1.1107370998648058e-06,
1598
+ "loss": 0.3998314142227173,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.7687107920646667,
1604
+ "learning_rate": 9.64266432170219e-07,
1605
+ "loss": 0.45170745849609373,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 0.6509118676185608,
1611
+ "learning_rate": 8.280542962673166e-07,
1612
+ "loss": 0.42131795883178713,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 0.6040301322937012,
1618
+ "learning_rate": 7.021291884469649e-07,
1619
+ "loss": 0.3819763422012329,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.8355879783630371,
1625
+ "learning_rate": 5.865174528999706e-07,
1626
+ "loss": 0.42635231018066405,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 0.548719048500061,
1632
+ "learning_rate": 4.812432762057673e-07,
1633
+ "loss": 0.35939807891845704,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 0.5332063436508179,
1639
+ "learning_rate": 3.8632868227242435e-07,
1640
+ "loss": 0.4174611568450928,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 0.5104030966758728,
1646
+ "learning_rate": 3.017935277291495e-07,
1647
+ "loss": 0.35938918590545654,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.0479686260223389,
1653
+ "learning_rate": 2.2765549777217788e-07,
1654
+ "loss": 0.353275465965271,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 0.6005430817604065,
1660
+ "learning_rate": 1.639301024649209e-07,
1661
+ "loss": 0.44223575592041015,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.6831442713737488,
1667
+ "learning_rate": 1.1063067349318946e-07,
1668
+ "loss": 0.36453771591186523,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 0.6081545948982239,
1674
+ "learning_rate": 6.776836137615262e-08,
1675
+ "loss": 0.35844419002532957,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.4227214455604553,
1681
+ "learning_rate": 3.535213313355934e-08,
1682
+ "loss": 0.399729585647583,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.9676080942153931,
1688
+ "learning_rate": 1.3388770409805685e-08,
1689
+ "loss": 0.43811402320861814,
1690
+ "step": 1200
1691
+ },
1692
+ {
1693
+ "epoch": 0.9983429991714996,
1694
+ "grad_norm": 0.9380176067352295,
1695
+ "learning_rate": 1.8828680551918888e-09,
1696
+ "loss": 0.36945352554321287,
1697
+ "step": 1205
1698
+ }
1699
+ ],
1700
+ "logging_steps": 5,
1701
+ "max_steps": 1207,
1702
+ "num_input_tokens_seen": 0,
1703
+ "num_train_epochs": 1,
1704
+ "save_steps": 200,
1705
+ "stateful_callbacks": {
1706
+ "TrainerControl": {
1707
+ "args": {
1708
+ "should_epoch_stop": false,
1709
+ "should_evaluate": false,
1710
+ "should_log": false,
1711
+ "should_save": true,
1712
+ "should_training_stop": true
1713
+ },
1714
+ "attributes": {}
1715
+ }
1716
+ },
1717
+ "total_flos": 2.5323923576872166e+17,
1718
+ "train_batch_size": 1,
1719
+ "trial_name": null,
1720
+ "trial_params": null
1721
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/config.json ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_plane_indices": [
171
+ 3,
172
+ 11,
173
+ 19,
174
+ 27,
175
+ 35,
176
+ 43,
177
+ 51,
178
+ 59
179
+ ],
180
+ "speaker_plane_order": "interleaved",
181
+ "speaker_rotary_dim": 16,
182
+ "speaker_rotary_inference_seed": 42,
183
+ "speaker_rotary_std": 0.02,
184
+ "speaker_sample_rate": 16000,
185
+ "speaker_tfmel_args": {
186
+ "do_preemph": true,
187
+ "do_spec_aug": false,
188
+ "eps": 1e-08,
189
+ "f_max": 7600,
190
+ "f_min": 20,
191
+ "hop_length": 160,
192
+ "n_fft": 512,
193
+ "n_mels": 72,
194
+ "norm_signal": true,
195
+ "win_length": 400
196
+ },
197
+ "speaker_window_seconds": 2.0,
198
+ "text_config": {
199
+ "attention_bias": false,
200
+ "attention_dropout": 0.0,
201
+ "bos_token_id": null,
202
+ "dtype": "bfloat16",
203
+ "eos_token_id": null,
204
+ "head_dim": 128,
205
+ "hidden_act": "silu",
206
+ "hidden_size": 1024,
207
+ "initializer_range": 0.02,
208
+ "intermediate_size": 3072,
209
+ "layer_types": [
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention",
227
+ "full_attention",
228
+ "full_attention",
229
+ "full_attention",
230
+ "full_attention",
231
+ "full_attention",
232
+ "full_attention",
233
+ "full_attention",
234
+ "full_attention",
235
+ "full_attention",
236
+ "full_attention",
237
+ "full_attention"
238
+ ],
239
+ "max_position_embeddings": 131072,
240
+ "max_window_layers": 28,
241
+ "model_type": "qwen3",
242
+ "num_attention_heads": 16,
243
+ "num_hidden_layers": 28,
244
+ "num_key_value_heads": 8,
245
+ "pad_token_id": 151643,
246
+ "rms_norm_eps": 1e-06,
247
+ "rope_parameters": {
248
+ "rope_theta": 1000000,
249
+ "rope_type": "default"
250
+ },
251
+ "sliding_window": null,
252
+ "tie_word_embeddings": true,
253
+ "use_cache": false,
254
+ "use_sliding_window": false,
255
+ "vocab_size": 151936
256
+ },
257
+ "tie_word_embeddings": true,
258
+ "transformers_version": "5.14.0",
259
+ "use_cache": false
260
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe6a5534156882e11783efeedfe33fb4b18b4b17b6b6ebdc5ab64c2f19c8fc4a
3
+ size 1868024544
MOSS_Speaker-RoPE/output_spk_odd8_s3/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:eb21895752d2380b32ae2e2c40905c7aa6c8e55c5e6ba33cd065bf77d4d53f74
3
+ size 11423344
MOSS_Speaker-RoPE/output_spk_odd8_s3/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 1.0,
3
+ "total_flos": 2.5323923576872166e+17,
4
+ "train_loss": 0.43766265026083045,
5
+ "train_runtime": 20328.9815,
6
+ "train_samples_per_second": 0.059,
7
+ "train_steps_per_second": 0.059
8
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/trainer_state.json ADDED
@@ -0,0 +1,1730 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 1.0,
6
+ "eval_steps": 500,
7
+ "global_step": 1207,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 20.629674911499023,
15
+ "learning_rate": 3.3057851239669424e-06,
16
+ "loss": 0.8916114807128906,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 12.812925338745117,
22
+ "learning_rate": 7.43801652892562e-06,
23
+ "loss": 0.558007287979126,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 7.200536251068115,
29
+ "learning_rate": 1.1570247933884299e-05,
30
+ "loss": 0.6662204742431641,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 1.363626480102539,
36
+ "learning_rate": 1.5702479338842978e-05,
37
+ "loss": 0.45300092697143557,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 3.3842880725860596,
43
+ "learning_rate": 1.9834710743801653e-05,
44
+ "loss": 0.494948148727417,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 1.388489842414856,
50
+ "learning_rate": 2.396694214876033e-05,
51
+ "loss": 0.4781633377075195,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 0.5891152024269104,
57
+ "learning_rate": 2.809917355371901e-05,
58
+ "loss": 0.48348569869995117,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 0.9562617540359497,
64
+ "learning_rate": 3.2231404958677685e-05,
65
+ "loss": 0.4969147205352783,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 0.6433483958244324,
71
+ "learning_rate": 3.6363636363636364e-05,
72
+ "loss": 0.4836277008056641,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 0.898301362991333,
78
+ "learning_rate": 4.049586776859504e-05,
79
+ "loss": 0.4123377799987793,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 0.8125119805335999,
85
+ "learning_rate": 4.462809917355372e-05,
86
+ "loss": 0.45714097023010253,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 0.5890400409698486,
92
+ "learning_rate": 4.87603305785124e-05,
93
+ "loss": 0.4525331497192383,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 0.6006086468696594,
99
+ "learning_rate": 5.289256198347108e-05,
100
+ "loss": 0.3964966297149658,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 1.017815113067627,
106
+ "learning_rate": 5.702479338842975e-05,
107
+ "loss": 0.37221901416778563,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 0.9333603978157043,
113
+ "learning_rate": 6.115702479338842e-05,
114
+ "loss": 0.4029701232910156,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 1.0475515127182007,
120
+ "learning_rate": 6.528925619834711e-05,
121
+ "loss": 0.45854949951171875,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 1.0967167615890503,
127
+ "learning_rate": 6.94214876033058e-05,
128
+ "loss": 0.455504035949707,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 0.7258841395378113,
134
+ "learning_rate": 7.355371900826447e-05,
135
+ "loss": 0.7798455238342286,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 1.0904923677444458,
141
+ "learning_rate": 7.768595041322314e-05,
142
+ "loss": 0.46855831146240234,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 0.924197256565094,
148
+ "learning_rate": 8.181818181818183e-05,
149
+ "loss": 0.4880524635314941,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 1.9587433338165283,
155
+ "learning_rate": 8.595041322314051e-05,
156
+ "loss": 0.463060998916626,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 0.6440144181251526,
162
+ "learning_rate": 9.008264462809917e-05,
163
+ "loss": 0.48012351989746094,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 0.8251569867134094,
169
+ "learning_rate": 9.421487603305785e-05,
170
+ "loss": 0.4483522415161133,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 0.8789721727371216,
176
+ "learning_rate": 9.834710743801654e-05,
177
+ "loss": 0.4580700397491455,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 0.8313543796539307,
183
+ "learning_rate": 9.999811713194481e-05,
184
+ "loss": 0.46596560478210447,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 0.6774608492851257,
190
+ "learning_rate": 9.99866112295902e-05,
191
+ "loss": 0.4806799411773682,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 0.6783014535903931,
197
+ "learning_rate": 9.996464786686644e-05,
198
+ "loss": 0.4560887336730957,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 0.7160524129867554,
204
+ "learning_rate": 9.993223163862385e-05,
205
+ "loss": 0.4927361011505127,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 0.5270054340362549,
211
+ "learning_rate": 9.988936932650681e-05,
212
+ "loss": 0.42853679656982424,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 1.0818812847137451,
218
+ "learning_rate": 9.983606989753509e-05,
219
+ "loss": 0.5378652572631836,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 0.968137800693512,
225
+ "learning_rate": 9.977234450222783e-05,
226
+ "loss": 0.48653225898742675,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 0.8479903936386108,
232
+ "learning_rate": 9.969820647227086e-05,
233
+ "loss": 0.5200669765472412,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 0.7051109075546265,
239
+ "learning_rate": 9.961367131772759e-05,
240
+ "loss": 0.4588489532470703,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 2.205345630645752,
246
+ "learning_rate": 9.951875672379424e-05,
247
+ "loss": 0.5904168128967285,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 0.8751126527786255,
253
+ "learning_rate": 9.941348254710003e-05,
254
+ "loss": 0.43442707061767577,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 0.671466052532196,
260
+ "learning_rate": 9.929787081155304e-05,
261
+ "loss": 0.3809401512145996,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 1.235303282737732,
267
+ "learning_rate": 9.91719457037327e-05,
268
+ "loss": 0.4351661205291748,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 0.5449215769767761,
274
+ "learning_rate": 9.903573356782977e-05,
275
+ "loss": 0.41109681129455566,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 0.6575606465339661,
281
+ "learning_rate": 9.888926290013521e-05,
282
+ "loss": 0.46851186752319335,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 0.528086245059967,
288
+ "learning_rate": 9.873256434307828e-05,
289
+ "loss": 0.40171051025390625,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 0.44805943965911865,
295
+ "learning_rate": 9.856567067881628e-05,
296
+ "loss": 0.3575094223022461,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 0.6913977861404419,
302
+ "learning_rate": 9.838861682237618e-05,
303
+ "loss": 0.4208680152893066,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 0.6070751547813416,
309
+ "learning_rate": 9.820143981435031e-05,
310
+ "loss": 0.4100965976715088,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 0.7453505396842957,
316
+ "learning_rate": 9.800417881314716e-05,
317
+ "loss": 0.5092792510986328,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 0.718826949596405,
323
+ "learning_rate": 9.779687508679935e-05,
324
+ "loss": 0.5120072841644288,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 0.8064695596694946,
330
+ "learning_rate": 9.75795720043301e-05,
331
+ "loss": 0.43071684837341306,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 0.6657348275184631,
337
+ "learning_rate": 9.73523150266802e-05,
338
+ "loss": 0.43650312423706056,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 0.6227285861968994,
344
+ "learning_rate": 9.711515169719732e-05,
345
+ "loss": 0.4965176582336426,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 0.6354798078536987,
351
+ "learning_rate": 9.686813163168973e-05,
352
+ "loss": 0.4135477542877197,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 0.8802199959754944,
358
+ "learning_rate": 9.661130650804643e-05,
359
+ "loss": 0.4657131671905518,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 0.6120206713676453,
365
+ "learning_rate": 9.634473005542579e-05,
366
+ "loss": 0.5105055332183838,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 0.6608315706253052,
372
+ "learning_rate": 9.606845804301524e-05,
373
+ "loss": 0.48976764678955076,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 0.6136434078216553,
379
+ "learning_rate": 9.578254826836391e-05,
380
+ "loss": 0.39600679874420164,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 0.7732479572296143,
386
+ "learning_rate": 9.548706054529124e-05,
387
+ "loss": 0.43799166679382323,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 0.8353448510169983,
393
+ "learning_rate": 9.518205669137352e-05,
394
+ "loss": 0.39786520004272463,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 0.5929117798805237,
400
+ "learning_rate": 9.486760051501139e-05,
401
+ "loss": 0.4511585235595703,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 0.7609494924545288,
407
+ "learning_rate": 9.454375780208071e-05,
408
+ "loss": 0.460142993927002,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 0.4482679069042206,
414
+ "learning_rate": 9.421059630216991e-05,
415
+ "loss": 0.42333049774169923,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 0.5880385637283325,
421
+ "learning_rate": 9.38681857144063e-05,
422
+ "loss": 0.39257287979125977,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 0.4800879955291748,
428
+ "learning_rate": 9.351659767287478e-05,
429
+ "loss": 0.46072850227355955,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 0.5711155533790588,
435
+ "learning_rate": 9.315590573163153e-05,
436
+ "loss": 0.4340324401855469,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 1.6473472118377686,
442
+ "learning_rate": 9.278618534931611e-05,
443
+ "loss": 0.4806147575378418,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 0.9981988668441772,
449
+ "learning_rate": 9.240751387336522e-05,
450
+ "loss": 0.45314898490905764,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 0.9140734076499939,
456
+ "learning_rate": 9.201997052383106e-05,
457
+ "loss": 0.48151440620422364,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 0.7532357573509216,
463
+ "learning_rate": 9.162363637680822e-05,
464
+ "loss": 0.4129056453704834,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 0.9811023473739624,
470
+ "learning_rate": 9.121859434747204e-05,
471
+ "loss": 0.484523344039917,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.4147282838821411,
477
+ "learning_rate": 9.080492917273237e-05,
478
+ "loss": 0.5427152633666992,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 1.0163606405258179,
484
+ "learning_rate": 9.038272739350618e-05,
485
+ "loss": 0.4134209156036377,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 0.6049103140830994,
491
+ "learning_rate": 8.995207733661267e-05,
492
+ "loss": 0.40175676345825195,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 0.7045316696166992,
498
+ "learning_rate": 8.951306909629492e-05,
499
+ "loss": 0.40282230377197265,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 0.6261049509048462,
505
+ "learning_rate": 8.906579451537171e-05,
506
+ "loss": 0.3981405973434448,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 1.287496566772461,
512
+ "learning_rate": 8.861034716602343e-05,
513
+ "loss": 0.4517494201660156,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 0.6869074702262878,
519
+ "learning_rate": 8.814682233021636e-05,
520
+ "loss": 0.44563779830932615,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 0.6493797302246094,
526
+ "learning_rate": 8.76753169797692e-05,
527
+ "loss": 0.4556128978729248,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 0.5839359760284424,
533
+ "learning_rate": 8.719592975606601e-05,
534
+ "loss": 0.43351993560791013,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 0.3983668088912964,
540
+ "learning_rate": 8.670876094941991e-05,
541
+ "loss": 0.4440763473510742,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 0.6596688032150269,
547
+ "learning_rate": 8.62139124780919e-05,
548
+ "loss": 0.5530457973480225,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 0.6001258492469788,
554
+ "learning_rate": 8.571148786696894e-05,
555
+ "loss": 0.4752801418304443,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 0.405813604593277,
561
+ "learning_rate": 8.520159222590605e-05,
562
+ "loss": 0.38340303897857664,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 0.8015041351318359,
568
+ "learning_rate": 8.468433222773677e-05,
569
+ "loss": 0.4637146949768066,
570
+ "step": 400
571
+ },
572
+ {
573
+ "epoch": 0.3355426677713339,
574
+ "grad_norm": 0.6381627321243286,
575
+ "learning_rate": 8.415981608595667e-05,
576
+ "loss": 0.42714495658874513,
577
+ "step": 405
578
+ },
579
+ {
580
+ "epoch": 0.3396851698425849,
581
+ "grad_norm": 0.5829190015792847,
582
+ "learning_rate": 8.36281535320844e-05,
583
+ "loss": 0.3956311702728271,
584
+ "step": 410
585
+ },
586
+ {
587
+ "epoch": 0.34382767191383595,
588
+ "grad_norm": 0.8698112964630127,
589
+ "learning_rate": 8.30894557927055e-05,
590
+ "loss": 0.4846199989318848,
591
+ "step": 415
592
+ },
593
+ {
594
+ "epoch": 0.347970173985087,
595
+ "grad_norm": 0.6768237352371216,
596
+ "learning_rate": 8.254383556620299e-05,
597
+ "loss": 0.47880539894104,
598
+ "step": 420
599
+ },
600
+ {
601
+ "epoch": 0.352112676056338,
602
+ "grad_norm": 0.5710030198097229,
603
+ "learning_rate": 8.199140699918049e-05,
604
+ "loss": 0.4479952335357666,
605
+ "step": 425
606
+ },
607
+ {
608
+ "epoch": 0.35625517812758906,
609
+ "grad_norm": 0.6229125261306763,
610
+ "learning_rate": 8.143228566258192e-05,
611
+ "loss": 0.433013916015625,
612
+ "step": 430
613
+ },
614
+ {
615
+ "epoch": 0.3603976801988401,
616
+ "grad_norm": 0.6174870133399963,
617
+ "learning_rate": 8.08665885275137e-05,
618
+ "loss": 0.4820831298828125,
619
+ "step": 435
620
+ },
621
+ {
622
+ "epoch": 0.36454018227009116,
623
+ "grad_norm": 1.9886339902877808,
624
+ "learning_rate": 8.029443394077356e-05,
625
+ "loss": 0.4312156677246094,
626
+ "step": 440
627
+ },
628
+ {
629
+ "epoch": 0.36868268434134216,
630
+ "grad_norm": 0.7721105217933655,
631
+ "learning_rate": 7.971594160009194e-05,
632
+ "loss": 0.4745286464691162,
633
+ "step": 445
634
+ },
635
+ {
636
+ "epoch": 0.3728251864125932,
637
+ "grad_norm": 0.8215830326080322,
638
+ "learning_rate": 7.913123252909056e-05,
639
+ "loss": 0.43494420051574706,
640
+ "step": 450
641
+ },
642
+ {
643
+ "epoch": 0.37696768848384427,
644
+ "grad_norm": 0.8584961295127869,
645
+ "learning_rate": 7.854042905196354e-05,
646
+ "loss": 0.46215004920959474,
647
+ "step": 455
648
+ },
649
+ {
650
+ "epoch": 0.38111019055509526,
651
+ "grad_norm": 0.483641117811203,
652
+ "learning_rate": 7.794365476788667e-05,
653
+ "loss": 0.37451136112213135,
654
+ "step": 460
655
+ },
656
+ {
657
+ "epoch": 0.3852526926263463,
658
+ "grad_norm": 0.567133903503418,
659
+ "learning_rate": 7.734103452515975e-05,
660
+ "loss": 0.4840177059173584,
661
+ "step": 465
662
+ },
663
+ {
664
+ "epoch": 0.38939519469759737,
665
+ "grad_norm": 0.6834204196929932,
666
+ "learning_rate": 7.67326943950877e-05,
667
+ "loss": 0.46247282028198244,
668
+ "step": 470
669
+ },
670
+ {
671
+ "epoch": 0.39353769676884837,
672
+ "grad_norm": 0.8395087122917175,
673
+ "learning_rate": 7.611876164560578e-05,
674
+ "loss": 0.36043124198913573,
675
+ "step": 475
676
+ },
677
+ {
678
+ "epoch": 0.3976801988400994,
679
+ "grad_norm": 0.3639385998249054,
680
+ "learning_rate": 7.549936471465444e-05,
681
+ "loss": 0.42089052200317384,
682
+ "step": 480
683
+ },
684
+ {
685
+ "epoch": 0.40182270091135047,
686
+ "grad_norm": 0.4404556453227997,
687
+ "learning_rate": 7.487463318330946e-05,
688
+ "loss": 0.3542492866516113,
689
+ "step": 485
690
+ },
691
+ {
692
+ "epoch": 0.40596520298260147,
693
+ "grad_norm": 0.6046218276023865,
694
+ "learning_rate": 7.42446977486729e-05,
695
+ "loss": 0.38021628856658934,
696
+ "step": 490
697
+ },
698
+ {
699
+ "epoch": 0.4101077050538525,
700
+ "grad_norm": 1.6285887956619263,
701
+ "learning_rate": 7.360969019653067e-05,
702
+ "loss": 0.6484775066375732,
703
+ "step": 495
704
+ },
705
+ {
706
+ "epoch": 0.4142502071251036,
707
+ "grad_norm": 0.9363073110580444,
708
+ "learning_rate": 7.296974337378209e-05,
709
+ "loss": 0.41137142181396485,
710
+ "step": 500
711
+ },
712
+ {
713
+ "epoch": 0.4183927091963546,
714
+ "grad_norm": 0.4143483340740204,
715
+ "learning_rate": 7.23249911606477e-05,
716
+ "loss": 0.3924715518951416,
717
+ "step": 505
718
+ },
719
+ {
720
+ "epoch": 0.4225352112676056,
721
+ "grad_norm": 1.2242801189422607,
722
+ "learning_rate": 7.167556844266091e-05,
723
+ "loss": 0.4372579574584961,
724
+ "step": 510
725
+ },
726
+ {
727
+ "epoch": 0.4266777133388567,
728
+ "grad_norm": 0.49445945024490356,
729
+ "learning_rate": 7.102161108244907e-05,
730
+ "loss": 0.4041281700134277,
731
+ "step": 515
732
+ },
733
+ {
734
+ "epoch": 0.43082021541010773,
735
+ "grad_norm": 0.5578067898750305,
736
+ "learning_rate": 7.036325589131048e-05,
737
+ "loss": 0.44759469032287597,
738
+ "step": 520
739
+ },
740
+ {
741
+ "epoch": 0.43496271748135873,
742
+ "grad_norm": 0.6338609457015991,
743
+ "learning_rate": 6.970064060059258e-05,
744
+ "loss": 0.3942832946777344,
745
+ "step": 525
746
+ },
747
+ {
748
+ "epoch": 0.4391052195526098,
749
+ "grad_norm": 0.6305315494537354,
750
+ "learning_rate": 6.903390383287794e-05,
751
+ "loss": 0.3744609832763672,
752
+ "step": 530
753
+ },
754
+ {
755
+ "epoch": 0.44324772162386084,
756
+ "grad_norm": 0.6225669980049133,
757
+ "learning_rate": 6.83631850729837e-05,
758
+ "loss": 0.4644033432006836,
759
+ "step": 535
760
+ },
761
+ {
762
+ "epoch": 0.44739022369511183,
763
+ "grad_norm": 0.4533715844154358,
764
+ "learning_rate": 6.768862463878047e-05,
765
+ "loss": 0.43013477325439453,
766
+ "step": 540
767
+ },
768
+ {
769
+ "epoch": 0.4515327257663629,
770
+ "grad_norm": 0.668293297290802,
771
+ "learning_rate": 6.701036365183737e-05,
772
+ "loss": 0.5000359058380127,
773
+ "step": 545
774
+ },
775
+ {
776
+ "epoch": 0.45567522783761394,
777
+ "grad_norm": 0.6384305357933044,
778
+ "learning_rate": 6.632854400789835e-05,
779
+ "loss": 0.4234946250915527,
780
+ "step": 550
781
+ },
782
+ {
783
+ "epoch": 0.45981772990886494,
784
+ "grad_norm": 0.35919633507728577,
785
+ "learning_rate": 6.564330834719708e-05,
786
+ "loss": 0.4122346878051758,
787
+ "step": 555
788
+ },
789
+ {
790
+ "epoch": 0.463960231980116,
791
+ "grad_norm": 0.5707134008407593,
792
+ "learning_rate": 6.495480002461577e-05,
793
+ "loss": 0.4068869113922119,
794
+ "step": 560
795
+ },
796
+ {
797
+ "epoch": 0.46810273405136704,
798
+ "grad_norm": 1.7894991636276245,
799
+ "learning_rate": 6.426316307969465e-05,
800
+ "loss": 0.4764272689819336,
801
+ "step": 565
802
+ },
803
+ {
804
+ "epoch": 0.47224523612261804,
805
+ "grad_norm": 0.7096588611602783,
806
+ "learning_rate": 6.3568542206498e-05,
807
+ "loss": 0.40894289016723634,
808
+ "step": 570
809
+ },
810
+ {
811
+ "epoch": 0.4763877381938691,
812
+ "grad_norm": 0.6279922723770142,
813
+ "learning_rate": 6.28710827233436e-05,
814
+ "loss": 0.4551478385925293,
815
+ "step": 575
816
+ },
817
+ {
818
+ "epoch": 0.48053024026512015,
819
+ "grad_norm": 0.9392326474189758,
820
+ "learning_rate": 6.21709305424012e-05,
821
+ "loss": 0.42044363021850584,
822
+ "step": 580
823
+ },
824
+ {
825
+ "epoch": 0.48467274233637114,
826
+ "grad_norm": 0.8099265098571777,
827
+ "learning_rate": 6.146823213916702e-05,
828
+ "loss": 0.421197509765625,
829
+ "step": 585
830
+ },
831
+ {
832
+ "epoch": 0.4888152444076222,
833
+ "grad_norm": 0.407980740070343,
834
+ "learning_rate": 6.076313452182033e-05,
835
+ "loss": 0.45879392623901366,
836
+ "step": 590
837
+ },
838
+ {
839
+ "epoch": 0.49295774647887325,
840
+ "grad_norm": 0.6435403823852539,
841
+ "learning_rate": 6.005578520046848e-05,
842
+ "loss": 0.7919346332550049,
843
+ "step": 595
844
+ },
845
+ {
846
+ "epoch": 0.4971002485501243,
847
+ "grad_norm": 0.5578521490097046,
848
+ "learning_rate": 5.934633215628711e-05,
849
+ "loss": 0.4143537044525146,
850
+ "step": 600
851
+ },
852
+ {
853
+ "epoch": 0.5012427506213754,
854
+ "grad_norm": 0.5052011013031006,
855
+ "learning_rate": 5.863492381056164e-05,
856
+ "loss": 0.38072867393493653,
857
+ "step": 605
858
+ },
859
+ {
860
+ "epoch": 0.5053852526926264,
861
+ "grad_norm": 0.6519150733947754,
862
+ "learning_rate": 5.792170899363683e-05,
863
+ "loss": 0.4297952175140381,
864
+ "step": 610
865
+ },
866
+ {
867
+ "epoch": 0.5095277547638773,
868
+ "grad_norm": 0.9843361377716064,
869
+ "learning_rate": 5.72068369137806e-05,
870
+ "loss": 0.3838667869567871,
871
+ "step": 615
872
+ },
873
+ {
874
+ "epoch": 0.5136702568351285,
875
+ "grad_norm": 0.7498537302017212,
876
+ "learning_rate": 5.649045712596903e-05,
877
+ "loss": 0.40416569709777833,
878
+ "step": 620
879
+ },
880
+ {
881
+ "epoch": 0.5178127589063795,
882
+ "grad_norm": 0.6527976393699646,
883
+ "learning_rate": 5.577271950059849e-05,
884
+ "loss": 0.4134161949157715,
885
+ "step": 625
886
+ },
887
+ {
888
+ "epoch": 0.5219552609776305,
889
+ "grad_norm": 0.8321830630302429,
890
+ "learning_rate": 5.505377419213214e-05,
891
+ "loss": 0.43797922134399414,
892
+ "step": 630
893
+ },
894
+ {
895
+ "epoch": 0.5260977630488816,
896
+ "grad_norm": 0.6492611169815063,
897
+ "learning_rate": 5.433377160768669e-05,
898
+ "loss": 0.4532886028289795,
899
+ "step": 635
900
+ },
901
+ {
902
+ "epoch": 0.5302402651201326,
903
+ "grad_norm": 0.6094037294387817,
904
+ "learning_rate": 5.3612862375566574e-05,
905
+ "loss": 0.43567543029785155,
906
+ "step": 640
907
+ },
908
+ {
909
+ "epoch": 0.5343827671913836,
910
+ "grad_norm": 0.551053524017334,
911
+ "learning_rate": 5.2891197313751606e-05,
912
+ "loss": 0.4400018692016602,
913
+ "step": 645
914
+ },
915
+ {
916
+ "epoch": 0.5385252692626347,
917
+ "grad_norm": 0.6376934051513672,
918
+ "learning_rate": 5.216892739834519e-05,
919
+ "loss": 0.42303104400634767,
920
+ "step": 650
921
+ },
922
+ {
923
+ "epoch": 0.5426677713338857,
924
+ "grad_norm": 0.9007910490036011,
925
+ "learning_rate": 5.144620373198916e-05,
926
+ "loss": 0.45666775703430174,
927
+ "step": 655
928
+ },
929
+ {
930
+ "epoch": 0.5468102734051367,
931
+ "grad_norm": 0.650941014289856,
932
+ "learning_rate": 5.072317751225254e-05,
933
+ "loss": 0.4667072296142578,
934
+ "step": 660
935
+ },
936
+ {
937
+ "epoch": 0.5509527754763878,
938
+ "grad_norm": 0.48627233505249023,
939
+ "learning_rate": 5e-05,
940
+ "loss": 0.3643670558929443,
941
+ "step": 665
942
+ },
943
+ {
944
+ "epoch": 0.5550952775476388,
945
+ "grad_norm": 0.7920517325401306,
946
+ "learning_rate": 4.927682248774748e-05,
947
+ "loss": 0.44395933151245115,
948
+ "step": 670
949
+ },
950
+ {
951
+ "epoch": 0.5592377796188898,
952
+ "grad_norm": 0.5550063252449036,
953
+ "learning_rate": 4.855379626801085e-05,
954
+ "loss": 0.36930572986602783,
955
+ "step": 675
956
+ },
957
+ {
958
+ "epoch": 0.5633802816901409,
959
+ "grad_norm": 0.6230601668357849,
960
+ "learning_rate": 4.783107260165483e-05,
961
+ "loss": 0.4727597236633301,
962
+ "step": 680
963
+ },
964
+ {
965
+ "epoch": 0.5675227837613919,
966
+ "grad_norm": 0.6426054835319519,
967
+ "learning_rate": 4.710880268624841e-05,
968
+ "loss": 0.460523796081543,
969
+ "step": 685
970
+ },
971
+ {
972
+ "epoch": 0.5716652858326429,
973
+ "grad_norm": 0.5865359902381897,
974
+ "learning_rate": 4.6387137624433444e-05,
975
+ "loss": 0.4252904415130615,
976
+ "step": 690
977
+ },
978
+ {
979
+ "epoch": 0.575807787903894,
980
+ "grad_norm": 0.5453794598579407,
981
+ "learning_rate": 4.5666228392313315e-05,
982
+ "loss": 0.4039918422698975,
983
+ "step": 695
984
+ },
985
+ {
986
+ "epoch": 0.579950289975145,
987
+ "grad_norm": 1.0684459209442139,
988
+ "learning_rate": 4.494622580786788e-05,
989
+ "loss": 0.42879571914672854,
990
+ "step": 700
991
+ },
992
+ {
993
+ "epoch": 0.584092792046396,
994
+ "grad_norm": 0.8819708824157715,
995
+ "learning_rate": 4.422728049940151e-05,
996
+ "loss": 0.35405232906341555,
997
+ "step": 705
998
+ },
999
+ {
1000
+ "epoch": 0.5882352941176471,
1001
+ "grad_norm": 0.5523223280906677,
1002
+ "learning_rate": 4.3509542874030986e-05,
1003
+ "loss": 0.36108903884887694,
1004
+ "step": 710
1005
+ },
1006
+ {
1007
+ "epoch": 0.5923777961888981,
1008
+ "grad_norm": 0.49861323833465576,
1009
+ "learning_rate": 4.27931630862194e-05,
1010
+ "loss": 0.4754959583282471,
1011
+ "step": 715
1012
+ },
1013
+ {
1014
+ "epoch": 0.5965202982601492,
1015
+ "grad_norm": 1.2883014678955078,
1016
+ "learning_rate": 4.207829100636318e-05,
1017
+ "loss": 0.38763861656188964,
1018
+ "step": 720
1019
+ },
1020
+ {
1021
+ "epoch": 0.6006628003314002,
1022
+ "grad_norm": 0.5406184792518616,
1023
+ "learning_rate": 4.1365076189438365e-05,
1024
+ "loss": 0.39906909465789797,
1025
+ "step": 725
1026
+ },
1027
+ {
1028
+ "epoch": 0.6048053024026512,
1029
+ "grad_norm": 0.8991988301277161,
1030
+ "learning_rate": 4.065366784371289e-05,
1031
+ "loss": 0.418521785736084,
1032
+ "step": 730
1033
+ },
1034
+ {
1035
+ "epoch": 0.6089478044739023,
1036
+ "grad_norm": 0.5573598742485046,
1037
+ "learning_rate": 3.994421479953153e-05,
1038
+ "loss": 0.43770723342895507,
1039
+ "step": 735
1040
+ },
1041
+ {
1042
+ "epoch": 0.6130903065451533,
1043
+ "grad_norm": 0.7751765847206116,
1044
+ "learning_rate": 3.9236865478179685e-05,
1045
+ "loss": 0.36180825233459474,
1046
+ "step": 740
1047
+ },
1048
+ {
1049
+ "epoch": 0.6172328086164043,
1050
+ "grad_norm": 0.5557352304458618,
1051
+ "learning_rate": 3.853176786083298e-05,
1052
+ "loss": 0.4191638946533203,
1053
+ "step": 745
1054
+ },
1055
+ {
1056
+ "epoch": 0.6213753106876554,
1057
+ "grad_norm": 1.012902021408081,
1058
+ "learning_rate": 3.7829069457598825e-05,
1059
+ "loss": 0.43454732894897463,
1060
+ "step": 750
1061
+ },
1062
+ {
1063
+ "epoch": 0.6255178127589064,
1064
+ "grad_norm": 0.8307779431343079,
1065
+ "learning_rate": 3.7128917276656406e-05,
1066
+ "loss": 0.3497183799743652,
1067
+ "step": 755
1068
+ },
1069
+ {
1070
+ "epoch": 0.6296603148301574,
1071
+ "grad_norm": 0.6312240958213806,
1072
+ "learning_rate": 3.6431457793502e-05,
1073
+ "loss": 0.43721756935119627,
1074
+ "step": 760
1075
+ },
1076
+ {
1077
+ "epoch": 0.6338028169014085,
1078
+ "grad_norm": 0.9830090403556824,
1079
+ "learning_rate": 3.573683692030537e-05,
1080
+ "loss": 0.48459711074829104,
1081
+ "step": 765
1082
+ },
1083
+ {
1084
+ "epoch": 0.6379453189726595,
1085
+ "grad_norm": 0.4116923213005066,
1086
+ "learning_rate": 3.504519997538422e-05,
1087
+ "loss": 0.31733293533325196,
1088
+ "step": 770
1089
+ },
1090
+ {
1091
+ "epoch": 0.6420878210439105,
1092
+ "grad_norm": 1.0353782176971436,
1093
+ "learning_rate": 3.435669165280294e-05,
1094
+ "loss": 0.43854188919067383,
1095
+ "step": 775
1096
+ },
1097
+ {
1098
+ "epoch": 0.6462303231151616,
1099
+ "grad_norm": 0.7071158289909363,
1100
+ "learning_rate": 3.3671455992101666e-05,
1101
+ "loss": 0.4189001560211182,
1102
+ "step": 780
1103
+ },
1104
+ {
1105
+ "epoch": 0.6503728251864126,
1106
+ "grad_norm": 1.6019755601882935,
1107
+ "learning_rate": 3.298963634816263e-05,
1108
+ "loss": 0.7713140487670899,
1109
+ "step": 785
1110
+ },
1111
+ {
1112
+ "epoch": 0.6545153272576636,
1113
+ "grad_norm": 0.5350252389907837,
1114
+ "learning_rate": 3.231137536121954e-05,
1115
+ "loss": 0.3926787137985229,
1116
+ "step": 790
1117
+ },
1118
+ {
1119
+ "epoch": 0.6586578293289147,
1120
+ "grad_norm": 0.37985295057296753,
1121
+ "learning_rate": 3.1636814927016324e-05,
1122
+ "loss": 0.4014743328094482,
1123
+ "step": 795
1124
+ },
1125
+ {
1126
+ "epoch": 0.6628003314001657,
1127
+ "grad_norm": 0.6357853412628174,
1128
+ "learning_rate": 3.096609616712207e-05,
1129
+ "loss": 0.4581602096557617,
1130
+ "step": 800
1131
+ },
1132
+ {
1133
+ "epoch": 0.6669428334714167,
1134
+ "grad_norm": 0.5001391768455505,
1135
+ "learning_rate": 3.0299359399407434e-05,
1136
+ "loss": 0.3605159044265747,
1137
+ "step": 805
1138
+ },
1139
+ {
1140
+ "epoch": 0.6710853355426678,
1141
+ "grad_norm": 0.9979525804519653,
1142
+ "learning_rate": 2.963674410868953e-05,
1143
+ "loss": 0.5704981803894043,
1144
+ "step": 810
1145
+ },
1146
+ {
1147
+ "epoch": 0.6752278376139188,
1148
+ "grad_norm": 0.6091923713684082,
1149
+ "learning_rate": 2.8978388917550935e-05,
1150
+ "loss": 0.3738471269607544,
1151
+ "step": 815
1152
+ },
1153
+ {
1154
+ "epoch": 0.6793703396851698,
1155
+ "grad_norm": 0.4976308345794678,
1156
+ "learning_rate": 2.8324431557339105e-05,
1157
+ "loss": 0.43538880348205566,
1158
+ "step": 820
1159
+ },
1160
+ {
1161
+ "epoch": 0.6835128417564209,
1162
+ "grad_norm": 0.9942322969436646,
1163
+ "learning_rate": 2.7675008839352314e-05,
1164
+ "loss": 0.3554206609725952,
1165
+ "step": 825
1166
+ },
1167
+ {
1168
+ "epoch": 0.6876553438276719,
1169
+ "grad_norm": 0.6866982579231262,
1170
+ "learning_rate": 2.703025662621793e-05,
1171
+ "loss": 0.4553103446960449,
1172
+ "step": 830
1173
+ },
1174
+ {
1175
+ "epoch": 0.6917978458989229,
1176
+ "grad_norm": 0.89406818151474,
1177
+ "learning_rate": 2.6390309803469326e-05,
1178
+ "loss": 0.43181838989257815,
1179
+ "step": 835
1180
+ },
1181
+ {
1182
+ "epoch": 0.695940347970174,
1183
+ "grad_norm": 0.5093903541564941,
1184
+ "learning_rate": 2.5755302251327096e-05,
1185
+ "loss": 0.4385825157165527,
1186
+ "step": 840
1187
+ },
1188
+ {
1189
+ "epoch": 0.700082850041425,
1190
+ "grad_norm": 0.5549860596656799,
1191
+ "learning_rate": 2.5125366816690555e-05,
1192
+ "loss": 0.329516339302063,
1193
+ "step": 845
1194
+ },
1195
+ {
1196
+ "epoch": 0.704225352112676,
1197
+ "grad_norm": 1.4509713649749756,
1198
+ "learning_rate": 2.450063528534558e-05,
1199
+ "loss": 0.3463990926742554,
1200
+ "step": 850
1201
+ },
1202
+ {
1203
+ "epoch": 0.7083678541839271,
1204
+ "grad_norm": 0.5200580358505249,
1205
+ "learning_rate": 2.3881238354394236e-05,
1206
+ "loss": 0.40448341369628904,
1207
+ "step": 855
1208
+ },
1209
+ {
1210
+ "epoch": 0.7125103562551781,
1211
+ "grad_norm": 1.1131764650344849,
1212
+ "learning_rate": 2.32673056049123e-05,
1213
+ "loss": 0.426401424407959,
1214
+ "step": 860
1215
+ },
1216
+ {
1217
+ "epoch": 0.7166528583264291,
1218
+ "grad_norm": 0.7889172434806824,
1219
+ "learning_rate": 2.2658965474840262e-05,
1220
+ "loss": 0.38459455966949463,
1221
+ "step": 865
1222
+ },
1223
+ {
1224
+ "epoch": 0.7207953603976802,
1225
+ "grad_norm": 0.517167866230011,
1226
+ "learning_rate": 2.205634523211335e-05,
1227
+ "loss": 0.36908698081970215,
1228
+ "step": 870
1229
+ },
1230
+ {
1231
+ "epoch": 0.7249378624689312,
1232
+ "grad_norm": 0.5234260559082031,
1233
+ "learning_rate": 2.1459570948036485e-05,
1234
+ "loss": 0.3863369941711426,
1235
+ "step": 875
1236
+ },
1237
+ {
1238
+ "epoch": 0.7290803645401823,
1239
+ "grad_norm": 0.7236451506614685,
1240
+ "learning_rate": 2.086876747090945e-05,
1241
+ "loss": 0.40941739082336426,
1242
+ "step": 880
1243
+ },
1244
+ {
1245
+ "epoch": 0.7332228666114333,
1246
+ "grad_norm": 0.4438631236553192,
1247
+ "learning_rate": 2.0284058399908045e-05,
1248
+ "loss": 0.34200191497802734,
1249
+ "step": 885
1250
+ },
1251
+ {
1252
+ "epoch": 0.7373653686826843,
1253
+ "grad_norm": 0.39247673749923706,
1254
+ "learning_rate": 1.9705566059226448e-05,
1255
+ "loss": 0.4146144390106201,
1256
+ "step": 890
1257
+ },
1258
+ {
1259
+ "epoch": 0.7415078707539354,
1260
+ "grad_norm": 1.1701165437698364,
1261
+ "learning_rate": 1.9133411472486324e-05,
1262
+ "loss": 0.4748171329498291,
1263
+ "step": 895
1264
+ },
1265
+ {
1266
+ "epoch": 0.7456503728251864,
1267
+ "grad_norm": 0.5402036309242249,
1268
+ "learning_rate": 1.8567714337418092e-05,
1269
+ "loss": 0.3901249170303345,
1270
+ "step": 900
1271
+ },
1272
+ {
1273
+ "epoch": 0.7497928748964374,
1274
+ "grad_norm": 0.6763427257537842,
1275
+ "learning_rate": 1.8008593000819518e-05,
1276
+ "loss": 0.4359886646270752,
1277
+ "step": 905
1278
+ },
1279
+ {
1280
+ "epoch": 0.7539353769676885,
1281
+ "grad_norm": 0.5417786240577698,
1282
+ "learning_rate": 1.745616443379699e-05,
1283
+ "loss": 0.39662876129150393,
1284
+ "step": 910
1285
+ },
1286
+ {
1287
+ "epoch": 0.7580778790389395,
1288
+ "grad_norm": 0.7444621920585632,
1289
+ "learning_rate": 1.691054420729452e-05,
1290
+ "loss": 0.3708503723144531,
1291
+ "step": 915
1292
+ },
1293
+ {
1294
+ "epoch": 0.7622203811101905,
1295
+ "grad_norm": 0.7745838165283203,
1296
+ "learning_rate": 1.6371846467915603e-05,
1297
+ "loss": 0.3678457736968994,
1298
+ "step": 920
1299
+ },
1300
+ {
1301
+ "epoch": 0.7663628831814416,
1302
+ "grad_norm": 0.4944072961807251,
1303
+ "learning_rate": 1.5840183914043338e-05,
1304
+ "loss": 0.3787250280380249,
1305
+ "step": 925
1306
+ },
1307
+ {
1308
+ "epoch": 0.7705053852526926,
1309
+ "grad_norm": 0.7562234997749329,
1310
+ "learning_rate": 1.5315667772263224e-05,
1311
+ "loss": 0.4543753623962402,
1312
+ "step": 930
1313
+ },
1314
+ {
1315
+ "epoch": 0.7746478873239436,
1316
+ "grad_norm": 0.5324202179908752,
1317
+ "learning_rate": 1.4798407774093954e-05,
1318
+ "loss": 0.34689984321594236,
1319
+ "step": 935
1320
+ },
1321
+ {
1322
+ "epoch": 0.7787903893951947,
1323
+ "grad_norm": 0.6181265711784363,
1324
+ "learning_rate": 1.4288512133031073e-05,
1325
+ "loss": 0.396760082244873,
1326
+ "step": 940
1327
+ },
1328
+ {
1329
+ "epoch": 0.7829328914664457,
1330
+ "grad_norm": 0.5689008831977844,
1331
+ "learning_rate": 1.3786087521908103e-05,
1332
+ "loss": 0.3692843198776245,
1333
+ "step": 945
1334
+ },
1335
+ {
1336
+ "epoch": 0.7870753935376967,
1337
+ "grad_norm": 0.631606936454773,
1338
+ "learning_rate": 1.3291239050580084e-05,
1339
+ "loss": 0.683953332901001,
1340
+ "step": 950
1341
+ },
1342
+ {
1343
+ "epoch": 0.7912178956089478,
1344
+ "grad_norm": 0.7201916575431824,
1345
+ "learning_rate": 1.2804070243933996e-05,
1346
+ "loss": 0.37404820919036863,
1347
+ "step": 955
1348
+ },
1349
+ {
1350
+ "epoch": 0.7953603976801988,
1351
+ "grad_norm": 0.5721992254257202,
1352
+ "learning_rate": 1.2324683020230798e-05,
1353
+ "loss": 0.7489796638488769,
1354
+ "step": 960
1355
+ },
1356
+ {
1357
+ "epoch": 0.7995028997514498,
1358
+ "grad_norm": 0.7852861881256104,
1359
+ "learning_rate": 1.1853177669783643e-05,
1360
+ "loss": 0.4024354457855225,
1361
+ "step": 965
1362
+ },
1363
+ {
1364
+ "epoch": 0.8036454018227009,
1365
+ "grad_norm": 0.5761847496032715,
1366
+ "learning_rate": 1.1389652833976577e-05,
1367
+ "loss": 0.4159713268280029,
1368
+ "step": 970
1369
+ },
1370
+ {
1371
+ "epoch": 0.8077879038939519,
1372
+ "grad_norm": 0.4608430564403534,
1373
+ "learning_rate": 1.0934205484628284e-05,
1374
+ "loss": 0.38759047985076905,
1375
+ "step": 975
1376
+ },
1377
+ {
1378
+ "epoch": 0.8119304059652029,
1379
+ "grad_norm": 0.4075900614261627,
1380
+ "learning_rate": 1.0486930903705094e-05,
1381
+ "loss": 0.38979687690734866,
1382
+ "step": 980
1383
+ },
1384
+ {
1385
+ "epoch": 0.816072908036454,
1386
+ "grad_norm": 0.6070588231086731,
1387
+ "learning_rate": 1.0047922663387349e-05,
1388
+ "loss": 0.43787498474121095,
1389
+ "step": 985
1390
+ },
1391
+ {
1392
+ "epoch": 0.820215410107705,
1393
+ "grad_norm": 0.5205054879188538,
1394
+ "learning_rate": 9.617272606493844e-06,
1395
+ "loss": 0.39637553691864014,
1396
+ "step": 990
1397
+ },
1398
+ {
1399
+ "epoch": 0.824357912178956,
1400
+ "grad_norm": 0.8867539763450623,
1401
+ "learning_rate": 9.195070827267632e-06,
1402
+ "loss": 0.4684460639953613,
1403
+ "step": 995
1404
+ },
1405
+ {
1406
+ "epoch": 0.8285004142502072,
1407
+ "grad_norm": 0.7465023398399353,
1408
+ "learning_rate": 8.781405652527974e-06,
1409
+ "loss": 0.3949731826782227,
1410
+ "step": 1000
1411
+ },
1412
+ {
1413
+ "epoch": 0.8326429163214581,
1414
+ "grad_norm": 0.6426427960395813,
1415
+ "learning_rate": 8.376363623191791e-06,
1416
+ "loss": 0.4073126792907715,
1417
+ "step": 1005
1418
+ },
1419
+ {
1420
+ "epoch": 0.8367854183927091,
1421
+ "grad_norm": 0.6218783259391785,
1422
+ "learning_rate": 7.980029476168944e-06,
1423
+ "loss": 0.43186631202697756,
1424
+ "step": 1010
1425
+ },
1426
+ {
1427
+ "epoch": 0.8409279204639603,
1428
+ "grad_norm": 0.6543543934822083,
1429
+ "learning_rate": 7.59248612663479e-06,
1430
+ "loss": 0.35069923400878905,
1431
+ "step": 1015
1432
+ },
1433
+ {
1434
+ "epoch": 0.8450704225352113,
1435
+ "grad_norm": 0.8917542099952698,
1436
+ "learning_rate": 7.213814650683887e-06,
1437
+ "loss": 0.47125835418701173,
1438
+ "step": 1020
1439
+ },
1440
+ {
1441
+ "epoch": 0.8492129246064622,
1442
+ "grad_norm": 0.4148687422275543,
1443
+ "learning_rate": 6.844094268368484e-06,
1444
+ "loss": 0.38721065521240233,
1445
+ "step": 1025
1446
+ },
1447
+ {
1448
+ "epoch": 0.8533554266777134,
1449
+ "grad_norm": 0.6250177621841431,
1450
+ "learning_rate": 6.483402327125232e-06,
1451
+ "loss": 0.44888858795166015,
1452
+ "step": 1030
1453
+ },
1454
+ {
1455
+ "epoch": 0.8574979287489644,
1456
+ "grad_norm": 0.4487530291080475,
1457
+ "learning_rate": 6.131814285593707e-06,
1458
+ "loss": 0.37155201435089114,
1459
+ "step": 1035
1460
+ },
1461
+ {
1462
+ "epoch": 0.8616404308202155,
1463
+ "grad_norm": 0.450059711933136,
1464
+ "learning_rate": 5.7894036978301035e-06,
1465
+ "loss": 0.44730415344238283,
1466
+ "step": 1040
1467
+ },
1468
+ {
1469
+ "epoch": 0.8657829328914665,
1470
+ "grad_norm": 0.858016312122345,
1471
+ "learning_rate": 5.4562421979192925e-06,
1472
+ "loss": 0.4445347309112549,
1473
+ "step": 1045
1474
+ },
1475
+ {
1476
+ "epoch": 0.8699254349627175,
1477
+ "grad_norm": 0.5695084929466248,
1478
+ "learning_rate": 5.132399484988615e-06,
1479
+ "loss": 0.4245354175567627,
1480
+ "step": 1050
1481
+ },
1482
+ {
1483
+ "epoch": 0.8740679370339686,
1484
+ "grad_norm": 0.4118499159812927,
1485
+ "learning_rate": 4.817943308626488e-06,
1486
+ "loss": 0.39943842887878417,
1487
+ "step": 1055
1488
+ },
1489
+ {
1490
+ "epoch": 0.8782104391052196,
1491
+ "grad_norm": 0.7165275812149048,
1492
+ "learning_rate": 4.512939454708776e-06,
1493
+ "loss": 0.43923196792602537,
1494
+ "step": 1060
1495
+ },
1496
+ {
1497
+ "epoch": 0.8823529411764706,
1498
+ "grad_norm": 0.5231788754463196,
1499
+ "learning_rate": 4.2174517316361075e-06,
1500
+ "loss": 0.41583733558654784,
1501
+ "step": 1065
1502
+ },
1503
+ {
1504
+ "epoch": 0.8864954432477217,
1505
+ "grad_norm": 0.4489861726760864,
1506
+ "learning_rate": 3.93154195698478e-06,
1507
+ "loss": 0.4420301914215088,
1508
+ "step": 1070
1509
+ },
1510
+ {
1511
+ "epoch": 0.8906379453189727,
1512
+ "grad_norm": 0.7916709184646606,
1513
+ "learning_rate": 3.655269944574202e-06,
1514
+ "loss": 0.415742826461792,
1515
+ "step": 1075
1516
+ },
1517
+ {
1518
+ "epoch": 0.8947804473902237,
1519
+ "grad_norm": 1.1566158533096313,
1520
+ "learning_rate": 3.388693491953576e-06,
1521
+ "loss": 0.4660985469818115,
1522
+ "step": 1080
1523
+ },
1524
+ {
1525
+ "epoch": 0.8989229494614748,
1526
+ "grad_norm": 0.5106304287910461,
1527
+ "learning_rate": 3.1318683683102754e-06,
1528
+ "loss": 0.36864919662475587,
1529
+ "step": 1085
1530
+ },
1531
+ {
1532
+ "epoch": 0.9030654515327258,
1533
+ "grad_norm": 0.6080639362335205,
1534
+ "learning_rate": 2.8848483028026895e-06,
1535
+ "loss": 0.44466662406921387,
1536
+ "step": 1090
1537
+ },
1538
+ {
1539
+ "epoch": 0.9072079536039768,
1540
+ "grad_norm": 0.5853272080421448,
1541
+ "learning_rate": 2.647684973319814e-06,
1542
+ "loss": 0.3511647701263428,
1543
+ "step": 1095
1544
+ },
1545
+ {
1546
+ "epoch": 0.9113504556752279,
1547
+ "grad_norm": 0.41179347038269043,
1548
+ "learning_rate": 2.4204279956698995e-06,
1549
+ "loss": 0.4063854217529297,
1550
+ "step": 1100
1551
+ },
1552
+ {
1553
+ "epoch": 0.9154929577464789,
1554
+ "grad_norm": 0.5927942991256714,
1555
+ "learning_rate": 2.2031249132006547e-06,
1556
+ "loss": 0.42436866760253905,
1557
+ "step": 1105
1558
+ },
1559
+ {
1560
+ "epoch": 0.9196354598177299,
1561
+ "grad_norm": 0.7885688543319702,
1562
+ "learning_rate": 1.9958211868528474e-06,
1563
+ "loss": 0.3807806968688965,
1564
+ "step": 1110
1565
+ },
1566
+ {
1567
+ "epoch": 0.923777961888981,
1568
+ "grad_norm": 0.5567767024040222,
1569
+ "learning_rate": 1.7985601856496947e-06,
1570
+ "loss": 0.4070634365081787,
1571
+ "step": 1115
1572
+ },
1573
+ {
1574
+ "epoch": 0.927920463960232,
1575
+ "grad_norm": 0.7330124378204346,
1576
+ "learning_rate": 1.6113831776238097e-06,
1577
+ "loss": 0.41397967338562014,
1578
+ "step": 1120
1579
+ },
1580
+ {
1581
+ "epoch": 0.932062966031483,
1582
+ "grad_norm": 0.45662936568260193,
1583
+ "learning_rate": 1.4343293211837195e-06,
1584
+ "loss": 0.3888988494873047,
1585
+ "step": 1125
1586
+ },
1587
+ {
1588
+ "epoch": 0.9362054681027341,
1589
+ "grad_norm": 0.48038941621780396,
1590
+ "learning_rate": 1.2674356569217282e-06,
1591
+ "loss": 0.3652131795883179,
1592
+ "step": 1130
1593
+ },
1594
+ {
1595
+ "epoch": 0.9403479701739851,
1596
+ "grad_norm": 0.6213480830192566,
1597
+ "learning_rate": 1.1107370998648058e-06,
1598
+ "loss": 0.3998314142227173,
1599
+ "step": 1135
1600
+ },
1601
+ {
1602
+ "epoch": 0.9444904722452361,
1603
+ "grad_norm": 0.7687107920646667,
1604
+ "learning_rate": 9.64266432170219e-07,
1605
+ "loss": 0.45170745849609373,
1606
+ "step": 1140
1607
+ },
1608
+ {
1609
+ "epoch": 0.9486329743164872,
1610
+ "grad_norm": 0.6509118676185608,
1611
+ "learning_rate": 8.280542962673166e-07,
1612
+ "loss": 0.42131795883178713,
1613
+ "step": 1145
1614
+ },
1615
+ {
1616
+ "epoch": 0.9527754763877382,
1617
+ "grad_norm": 0.6040301322937012,
1618
+ "learning_rate": 7.021291884469649e-07,
1619
+ "loss": 0.3819763422012329,
1620
+ "step": 1150
1621
+ },
1622
+ {
1623
+ "epoch": 0.9569179784589892,
1624
+ "grad_norm": 0.8355879783630371,
1625
+ "learning_rate": 5.865174528999706e-07,
1626
+ "loss": 0.42635231018066405,
1627
+ "step": 1155
1628
+ },
1629
+ {
1630
+ "epoch": 0.9610604805302403,
1631
+ "grad_norm": 0.548719048500061,
1632
+ "learning_rate": 4.812432762057673e-07,
1633
+ "loss": 0.35939807891845704,
1634
+ "step": 1160
1635
+ },
1636
+ {
1637
+ "epoch": 0.9652029826014913,
1638
+ "grad_norm": 0.5332063436508179,
1639
+ "learning_rate": 3.8632868227242435e-07,
1640
+ "loss": 0.4174611568450928,
1641
+ "step": 1165
1642
+ },
1643
+ {
1644
+ "epoch": 0.9693454846727423,
1645
+ "grad_norm": 0.5104030966758728,
1646
+ "learning_rate": 3.017935277291495e-07,
1647
+ "loss": 0.35938918590545654,
1648
+ "step": 1170
1649
+ },
1650
+ {
1651
+ "epoch": 0.9734879867439934,
1652
+ "grad_norm": 1.0479686260223389,
1653
+ "learning_rate": 2.2765549777217788e-07,
1654
+ "loss": 0.353275465965271,
1655
+ "step": 1175
1656
+ },
1657
+ {
1658
+ "epoch": 0.9776304888152444,
1659
+ "grad_norm": 0.6005430817604065,
1660
+ "learning_rate": 1.639301024649209e-07,
1661
+ "loss": 0.44223575592041015,
1662
+ "step": 1180
1663
+ },
1664
+ {
1665
+ "epoch": 0.9817729908864954,
1666
+ "grad_norm": 0.6831442713737488,
1667
+ "learning_rate": 1.1063067349318946e-07,
1668
+ "loss": 0.36453771591186523,
1669
+ "step": 1185
1670
+ },
1671
+ {
1672
+ "epoch": 0.9859154929577465,
1673
+ "grad_norm": 0.6081545948982239,
1674
+ "learning_rate": 6.776836137615262e-08,
1675
+ "loss": 0.35844419002532957,
1676
+ "step": 1190
1677
+ },
1678
+ {
1679
+ "epoch": 0.9900579950289975,
1680
+ "grad_norm": 0.4227214455604553,
1681
+ "learning_rate": 3.535213313355934e-08,
1682
+ "loss": 0.399729585647583,
1683
+ "step": 1195
1684
+ },
1685
+ {
1686
+ "epoch": 0.9942004971002486,
1687
+ "grad_norm": 0.9676080942153931,
1688
+ "learning_rate": 1.3388770409805685e-08,
1689
+ "loss": 0.43811402320861814,
1690
+ "step": 1200
1691
+ },
1692
+ {
1693
+ "epoch": 0.9983429991714996,
1694
+ "grad_norm": 0.9380176067352295,
1695
+ "learning_rate": 1.8828680551918888e-09,
1696
+ "loss": 0.36945352554321287,
1697
+ "step": 1205
1698
+ },
1699
+ {
1700
+ "epoch": 1.0,
1701
+ "step": 1207,
1702
+ "total_flos": 2.5323923576872166e+17,
1703
+ "train_loss": 0.43766265026083045,
1704
+ "train_runtime": 20328.9815,
1705
+ "train_samples_per_second": 0.059,
1706
+ "train_steps_per_second": 0.059
1707
+ }
1708
+ ],
1709
+ "logging_steps": 5,
1710
+ "max_steps": 1207,
1711
+ "num_input_tokens_seen": 0,
1712
+ "num_train_epochs": 1,
1713
+ "save_steps": 200,
1714
+ "stateful_callbacks": {
1715
+ "TrainerControl": {
1716
+ "args": {
1717
+ "should_epoch_stop": false,
1718
+ "should_evaluate": false,
1719
+ "should_log": false,
1720
+ "should_save": true,
1721
+ "should_training_stop": true
1722
+ },
1723
+ "attributes": {}
1724
+ }
1725
+ },
1726
+ "total_flos": 2.5323923576872166e+17,
1727
+ "train_batch_size": 1,
1728
+ "trial_name": null,
1729
+ "trial_params": null
1730
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ff6c93d4e24f3dbcb6d51b3baf817d83d0ead3706ab19f7f34fac257d122e9e7
3
+ size 5201
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/chat_template.jinja ADDED
@@ -0,0 +1,103 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {%- macro render_content(content) -%}
2
+ {%- if content is string -%}
3
+ {{- content -}}
4
+ {%- else -%}
5
+ {%- for item in content -%}
6
+ {%- if item.type == 'audio' or 'audio' in item or 'audio_url' in item -%}
7
+ {{- '<|audio_start|><|audio_pad|><|audio_end|>\n' -}}
8
+ {%- elif item.type == 'text' -%}
9
+ {{- item.text -}}
10
+ {%- endif -%}
11
+ {%- endfor -%}
12
+ {%- endif -%}
13
+ {%- endmacro -%}
14
+ {%- if tools %}
15
+ {{- '<|im_start|>system\n' }}
16
+ {%- if messages[0].role == 'system' %}
17
+ {{- render_content(messages[0].content) + '\n\n' }}
18
+ {%- else %}
19
+ {{- 'You are a helpful assistant.\n\n' }}
20
+ {%- endif %}
21
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
22
+ {%- for tool in tools %}
23
+ {{- "\n" }}
24
+ {{- tool | tojson }}
25
+ {%- endfor %}
26
+ {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
27
+ {%- else %}
28
+ {%- if messages[0].role == 'system' %}
29
+ {{- '<|im_start|>system\n' + render_content(messages[0].content) + '<|im_end|>\n' }}
30
+ {%- else %}
31
+ {{- '<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n' }}
32
+ {%- endif %}
33
+ {%- endif %}
34
+ {%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
35
+ {%- for message in messages[::-1] %}
36
+ {%- set index = (messages|length - 1) - loop.index0 %}
37
+ {%- set content = render_content(message.content) %}
38
+ {%- if ns.multi_step_tool and message.role == "user" and content is string and not(content.startswith('<tool_response>') and content.endswith('</tool_response>')) %}
39
+ {%- set ns.multi_step_tool = false %}
40
+ {%- set ns.last_query_index = index %}
41
+ {%- endif %}
42
+ {%- endfor %}
43
+ {%- for message in messages %}
44
+ {%- set content = render_content(message.content) %}
45
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
46
+ {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>\n' }}
47
+ {%- elif message.role == "assistant" %}
48
+ {%- set reasoning_content = '' %}
49
+ {%- if message.reasoning_content is string %}
50
+ {%- set reasoning_content = message.reasoning_content %}
51
+ {%- else %}
52
+ {%- if '</think>' in content %}
53
+ {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
54
+ {%- set content = content.split('</think>')[-1].lstrip('\n') %}
55
+ {%- endif %}
56
+ {%- endif %}
57
+ {%- if loop.index0 > ns.last_query_index %}
58
+ {%- if loop.last or (not loop.last and reasoning_content) %}
59
+ {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
60
+ {%- else %}
61
+ {{- '<|im_start|>' + message.role + '\n' + content }}
62
+ {%- endif %}
63
+ {%- else %}
64
+ {{- '<|im_start|>' + message.role + '\n' + content }}
65
+ {%- endif %}
66
+ {%- if message.tool_calls %}
67
+ {%- for tool_call in message.tool_calls %}
68
+ {%- if (loop.first and content) or (not loop.first) %}
69
+ {{- '\n' }}
70
+ {%- endif %}
71
+ {%- if tool_call.function %}
72
+ {%- set tool_call = tool_call.function %}
73
+ {%- endif %}
74
+ {{- '<tool_call>\n{"name": "' }}
75
+ {{- tool_call.name }}
76
+ {{- '", "arguments": ' }}
77
+ {%- if tool_call.arguments is string %}
78
+ {{- tool_call.arguments }}
79
+ {%- else %}
80
+ {{- tool_call.arguments | tojson }}
81
+ {%- endif %}
82
+ {{- '}\n</tool_call>' }}
83
+ {%- endfor %}
84
+ {%- endif %}
85
+ {{- '<|im_end|>\n' }}
86
+ {%- elif message.role == "tool" %}
87
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
88
+ {{- '<|im_start|>user' }}
89
+ {%- endif %}
90
+ {{- '\n<tool_response>\n' }}
91
+ {{- content }}
92
+ {{- '\n</tool_response>' }}
93
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
94
+ {{- '<|im_end|>\n' }}
95
+ {%- endif %}
96
+ {%- endif %}
97
+ {%- endfor %}
98
+ {%- if add_generation_prompt %}
99
+ {{- '<|im_start|>assistant\n' }}
100
+ {%- if enable_thinking is defined and enable_thinking is false %}
101
+ {{- '<think>\n\n</think>\n\n' }}
102
+ {%- endif %}
103
+ {%- endif %}
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/config.json ADDED
@@ -0,0 +1,260 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "adaptor_input_dim": 4096,
3
+ "architectures": [
4
+ "MossSpeakerRopeForConditionalGeneration"
5
+ ],
6
+ "audio_config": {
7
+ "activation_dropout": 0.0,
8
+ "activation_function": "gelu",
9
+ "apply_spec_augment": false,
10
+ "attention_dropout": 0.0,
11
+ "begin_suppress_tokens": [
12
+ 220,
13
+ 50256
14
+ ],
15
+ "bos_token_id": 50256,
16
+ "classifier_proj_size": 256,
17
+ "d_model": 1024,
18
+ "decoder_attention_heads": 6,
19
+ "decoder_ffn_dim": 1536,
20
+ "decoder_layerdrop": 0.0,
21
+ "decoder_layers": 4,
22
+ "decoder_start_token_id": 50257,
23
+ "dropout": 0.0,
24
+ "dtype": "bfloat16",
25
+ "encoder_attention_heads": 16,
26
+ "encoder_ffn_dim": 4096,
27
+ "encoder_layerdrop": 0.0,
28
+ "encoder_layers": 24,
29
+ "eos_token_id": 50256,
30
+ "init_std": 0.02,
31
+ "mask_feature_length": 10,
32
+ "mask_feature_min_masks": 0,
33
+ "mask_feature_prob": 0.0,
34
+ "mask_time_length": 10,
35
+ "mask_time_min_masks": 2,
36
+ "mask_time_prob": 0.05,
37
+ "max_source_positions": 1500,
38
+ "max_target_positions": 448,
39
+ "median_filter_width": 7,
40
+ "model_type": "whisper",
41
+ "num_mel_bins": 80,
42
+ "pad_token_id": 50256,
43
+ "scale_embedding": false,
44
+ "suppress_tokens": null,
45
+ "tie_word_embeddings": true,
46
+ "use_cache": true,
47
+ "use_weighted_layer_sum": false,
48
+ "vocab_size": 51865
49
+ },
50
+ "audio_merge_size": 4,
51
+ "audio_token_id": 151671,
52
+ "bos_token_id": null,
53
+ "dtype": "bfloat16",
54
+ "eos_token_id": 151645,
55
+ "model_type": "moss_speaker_rope",
56
+ "pad_token_id": 151643,
57
+ "speaker_embed_dim": 192,
58
+ "speaker_encoder_args": {
59
+ "C": 64,
60
+ "F": 72,
61
+ "block_1d_type": "conv+att",
62
+ "block_2d_type": "basic_resnet",
63
+ "causal": "none",
64
+ "compress_tconvs": true,
65
+ "emb_bn": false,
66
+ "embed_dim": 192,
67
+ "feat_dim": 72,
68
+ "fm_weigthing_type": "NC",
69
+ "global_context_att": true,
70
+ "group_divisor": 1,
71
+ "hop_length": 160,
72
+ "out_channels": 224,
73
+ "pooling_func": "ASTP",
74
+ "return_2d_output": true,
75
+ "spec": null,
76
+ "spec_in_channels": 1,
77
+ "stages_setup": [
78
+ [
79
+ [
80
+ 1,
81
+ 1
82
+ ],
83
+ 3,
84
+ 3,
85
+ [
86
+ [
87
+ 3,
88
+ 3
89
+ ]
90
+ ],
91
+ 64
92
+ ],
93
+ [
94
+ [
95
+ 2,
96
+ 1
97
+ ],
98
+ 4,
99
+ 2,
100
+ [
101
+ [
102
+ 3,
103
+ 3
104
+ ]
105
+ ],
106
+ 64
107
+ ],
108
+ [
109
+ [
110
+ 1,
111
+ 2
112
+ ],
113
+ 5,
114
+ 2,
115
+ [
116
+ [
117
+ 3,
118
+ 3
119
+ ]
120
+ ],
121
+ 48
122
+ ],
123
+ [
124
+ [
125
+ 2,
126
+ 1
127
+ ],
128
+ 5,
129
+ 1,
130
+ [
131
+ [
132
+ 3,
133
+ 3
134
+ ]
135
+ ],
136
+ 48
137
+ ],
138
+ [
139
+ [
140
+ 1,
141
+ 2
142
+ ],
143
+ 4,
144
+ 0.75,
145
+ [
146
+ [
147
+ 3,
148
+ 3
149
+ ]
150
+ ],
151
+ 32
152
+ ],
153
+ [
154
+ [
155
+ 2,
156
+ 1
157
+ ],
158
+ 3,
159
+ 0.5,
160
+ [
161
+ [
162
+ 3,
163
+ 3
164
+ ]
165
+ ],
166
+ 24
167
+ ]
168
+ ]
169
+ },
170
+ "speaker_plane_indices": [
171
+ 3,
172
+ 11,
173
+ 19,
174
+ 27,
175
+ 35,
176
+ 43,
177
+ 51,
178
+ 59
179
+ ],
180
+ "speaker_plane_order": "interleaved",
181
+ "speaker_rotary_dim": 16,
182
+ "speaker_rotary_inference_seed": 42,
183
+ "speaker_rotary_std": 0.02,
184
+ "speaker_sample_rate": 16000,
185
+ "speaker_tfmel_args": {
186
+ "do_preemph": true,
187
+ "do_spec_aug": false,
188
+ "eps": 1e-08,
189
+ "f_max": 7600,
190
+ "f_min": 20,
191
+ "hop_length": 160,
192
+ "n_fft": 512,
193
+ "n_mels": 72,
194
+ "norm_signal": true,
195
+ "win_length": 400
196
+ },
197
+ "speaker_window_seconds": 2.0,
198
+ "text_config": {
199
+ "attention_bias": false,
200
+ "attention_dropout": 0.0,
201
+ "bos_token_id": null,
202
+ "dtype": "bfloat16",
203
+ "eos_token_id": null,
204
+ "head_dim": 128,
205
+ "hidden_act": "silu",
206
+ "hidden_size": 1024,
207
+ "initializer_range": 0.02,
208
+ "intermediate_size": 3072,
209
+ "layer_types": [
210
+ "full_attention",
211
+ "full_attention",
212
+ "full_attention",
213
+ "full_attention",
214
+ "full_attention",
215
+ "full_attention",
216
+ "full_attention",
217
+ "full_attention",
218
+ "full_attention",
219
+ "full_attention",
220
+ "full_attention",
221
+ "full_attention",
222
+ "full_attention",
223
+ "full_attention",
224
+ "full_attention",
225
+ "full_attention",
226
+ "full_attention",
227
+ "full_attention",
228
+ "full_attention",
229
+ "full_attention",
230
+ "full_attention",
231
+ "full_attention",
232
+ "full_attention",
233
+ "full_attention",
234
+ "full_attention",
235
+ "full_attention",
236
+ "full_attention",
237
+ "full_attention"
238
+ ],
239
+ "max_position_embeddings": 131072,
240
+ "max_window_layers": 28,
241
+ "model_type": "qwen3",
242
+ "num_attention_heads": 16,
243
+ "num_hidden_layers": 28,
244
+ "num_key_value_heads": 8,
245
+ "pad_token_id": 151643,
246
+ "rms_norm_eps": 1e-06,
247
+ "rope_parameters": {
248
+ "rope_theta": 1000000,
249
+ "rope_type": "default"
250
+ },
251
+ "sliding_window": null,
252
+ "tie_word_embeddings": true,
253
+ "use_cache": false,
254
+ "use_sliding_window": false,
255
+ "vocab_size": 151936
256
+ },
257
+ "tie_word_embeddings": true,
258
+ "transformers_version": "5.14.0",
259
+ "use_cache": false
260
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/generation_config.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "eos_token_id": [
3
+ 151645,
4
+ 151645
5
+ ],
6
+ "max_new_tokens": 5120,
7
+ "pad_token_id": 151643,
8
+ "transformers_version": "5.14.0"
9
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/processor_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "audio_merge_size": 4,
3
+ "audio_tokens_per_second": 12.5,
4
+ "enable_time_marker": true,
5
+ "feature_extractor": {
6
+ "chunk_length": 30,
7
+ "dither": 0.0,
8
+ "feature_extractor_type": "WhisperFeatureExtractor",
9
+ "feature_size": 80,
10
+ "hop_length": 160,
11
+ "n_fft": 400,
12
+ "n_samples": 480000,
13
+ "nb_max_frames": 3000,
14
+ "padding_side": "right",
15
+ "padding_value": 0.0,
16
+ "return_attention_mask": false,
17
+ "sampling_rate": 16000
18
+ },
19
+ "processor_class": "MossSpeakerRopeProcessor",
20
+ "speaker_window_seconds": 2.0,
21
+ "time_marker_every_seconds": 5
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/tokenizer_config.json ADDED
@@ -0,0 +1,22 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": false,
3
+ "backend": "tokenizers",
4
+ "bos_token": null,
5
+ "clean_up_tokenization_spaces": false,
6
+ "eos_token": "<|im_end|>",
7
+ "errors": "replace",
8
+ "extra_special_tokens": [
9
+ "<|audio_start|>",
10
+ "<|audio_end|>",
11
+ "<|audio_pad|>"
12
+ ],
13
+ "fix_mistral_regex": true,
14
+ "is_local": true,
15
+ "local_files_only": false,
16
+ "model_max_length": 131072,
17
+ "pad_token": "<|endoftext|>",
18
+ "processor_class": "MossSpeakerRopeProcessor",
19
+ "split_special_tokens": false,
20
+ "tokenizer_class": "Qwen2Tokenizer",
21
+ "unk_token": null
22
+ }
MOSS_Speaker-RoPE/output_spk_odd8_s3_postnorm_old/checkpoint-400/trainer_state.json ADDED
@@ -0,0 +1,594 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_global_step": null,
3
+ "best_metric": null,
4
+ "best_model_checkpoint": null,
5
+ "epoch": 0.33140016570008285,
6
+ "eval_steps": 500,
7
+ "global_step": 400,
8
+ "is_hyper_param_search": false,
9
+ "is_local_process_zero": true,
10
+ "is_world_process_zero": true,
11
+ "log_history": [
12
+ {
13
+ "epoch": 0.004142502071251036,
14
+ "grad_norm": 31.104061126708984,
15
+ "learning_rate": 3.3057851239669424e-06,
16
+ "loss": 1.1910821914672851,
17
+ "step": 5
18
+ },
19
+ {
20
+ "epoch": 0.008285004142502071,
21
+ "grad_norm": 35.740692138671875,
22
+ "learning_rate": 7.43801652892562e-06,
23
+ "loss": 0.7337178230285645,
24
+ "step": 10
25
+ },
26
+ {
27
+ "epoch": 0.012427506213753107,
28
+ "grad_norm": 9.301889419555664,
29
+ "learning_rate": 1.1570247933884299e-05,
30
+ "loss": 0.7512782096862793,
31
+ "step": 15
32
+ },
33
+ {
34
+ "epoch": 0.016570008285004142,
35
+ "grad_norm": 2.6755707263946533,
36
+ "learning_rate": 1.5702479338842978e-05,
37
+ "loss": 0.4849919319152832,
38
+ "step": 20
39
+ },
40
+ {
41
+ "epoch": 0.020712510356255178,
42
+ "grad_norm": 1.123673677444458,
43
+ "learning_rate": 1.9834710743801653e-05,
44
+ "loss": 0.529819917678833,
45
+ "step": 25
46
+ },
47
+ {
48
+ "epoch": 0.024855012427506214,
49
+ "grad_norm": 2.8139865398406982,
50
+ "learning_rate": 2.396694214876033e-05,
51
+ "loss": 0.5005036354064941,
52
+ "step": 30
53
+ },
54
+ {
55
+ "epoch": 0.02899751449875725,
56
+ "grad_norm": 1.766931414604187,
57
+ "learning_rate": 2.809917355371901e-05,
58
+ "loss": 0.4980137825012207,
59
+ "step": 35
60
+ },
61
+ {
62
+ "epoch": 0.033140016570008285,
63
+ "grad_norm": 1.1081942319869995,
64
+ "learning_rate": 3.2231404958677685e-05,
65
+ "loss": 0.5065455913543702,
66
+ "step": 40
67
+ },
68
+ {
69
+ "epoch": 0.037282518641259324,
70
+ "grad_norm": 1.07539963722229,
71
+ "learning_rate": 3.6363636363636364e-05,
72
+ "loss": 0.4957615852355957,
73
+ "step": 45
74
+ },
75
+ {
76
+ "epoch": 0.041425020712510356,
77
+ "grad_norm": 1.0425145626068115,
78
+ "learning_rate": 4.049586776859504e-05,
79
+ "loss": 0.42017226219177245,
80
+ "step": 50
81
+ },
82
+ {
83
+ "epoch": 0.045567522783761395,
84
+ "grad_norm": 0.8854573965072632,
85
+ "learning_rate": 4.462809917355372e-05,
86
+ "loss": 0.46346406936645507,
87
+ "step": 55
88
+ },
89
+ {
90
+ "epoch": 0.04971002485501243,
91
+ "grad_norm": 0.678348183631897,
92
+ "learning_rate": 4.87603305785124e-05,
93
+ "loss": 0.4590586185455322,
94
+ "step": 60
95
+ },
96
+ {
97
+ "epoch": 0.053852526926263466,
98
+ "grad_norm": 0.6650195121765137,
99
+ "learning_rate": 5.289256198347108e-05,
100
+ "loss": 0.3999401807785034,
101
+ "step": 65
102
+ },
103
+ {
104
+ "epoch": 0.0579950289975145,
105
+ "grad_norm": 0.9779027700424194,
106
+ "learning_rate": 5.702479338842975e-05,
107
+ "loss": 0.37582452297210694,
108
+ "step": 70
109
+ },
110
+ {
111
+ "epoch": 0.06213753106876554,
112
+ "grad_norm": 1.0380560159683228,
113
+ "learning_rate": 6.115702479338842e-05,
114
+ "loss": 0.4077299118041992,
115
+ "step": 75
116
+ },
117
+ {
118
+ "epoch": 0.06628003314001657,
119
+ "grad_norm": 1.2877566814422607,
120
+ "learning_rate": 6.528925619834711e-05,
121
+ "loss": 0.46419506072998046,
122
+ "step": 80
123
+ },
124
+ {
125
+ "epoch": 0.07042253521126761,
126
+ "grad_norm": 1.029144287109375,
127
+ "learning_rate": 6.94214876033058e-05,
128
+ "loss": 0.4615780830383301,
129
+ "step": 85
130
+ },
131
+ {
132
+ "epoch": 0.07456503728251865,
133
+ "grad_norm": 0.5973961353302002,
134
+ "learning_rate": 7.355371900826447e-05,
135
+ "loss": 0.7924308776855469,
136
+ "step": 90
137
+ },
138
+ {
139
+ "epoch": 0.07870753935376967,
140
+ "grad_norm": 1.3301033973693848,
141
+ "learning_rate": 7.768595041322314e-05,
142
+ "loss": 0.47766985893249514,
143
+ "step": 95
144
+ },
145
+ {
146
+ "epoch": 0.08285004142502071,
147
+ "grad_norm": 1.078078269958496,
148
+ "learning_rate": 8.181818181818183e-05,
149
+ "loss": 0.500474214553833,
150
+ "step": 100
151
+ },
152
+ {
153
+ "epoch": 0.08699254349627175,
154
+ "grad_norm": 1.0514358282089233,
155
+ "learning_rate": 8.595041322314051e-05,
156
+ "loss": 0.47164087295532225,
157
+ "step": 105
158
+ },
159
+ {
160
+ "epoch": 0.09113504556752279,
161
+ "grad_norm": 0.6962097883224487,
162
+ "learning_rate": 9.008264462809917e-05,
163
+ "loss": 0.4897346019744873,
164
+ "step": 110
165
+ },
166
+ {
167
+ "epoch": 0.09527754763877382,
168
+ "grad_norm": 0.9387894868850708,
169
+ "learning_rate": 9.421487603305785e-05,
170
+ "loss": 0.45862503051757814,
171
+ "step": 115
172
+ },
173
+ {
174
+ "epoch": 0.09942004971002485,
175
+ "grad_norm": 1.0993168354034424,
176
+ "learning_rate": 9.834710743801654e-05,
177
+ "loss": 0.463796329498291,
178
+ "step": 120
179
+ },
180
+ {
181
+ "epoch": 0.1035625517812759,
182
+ "grad_norm": 0.8082759380340576,
183
+ "learning_rate": 9.999811713194481e-05,
184
+ "loss": 0.47555227279663087,
185
+ "step": 125
186
+ },
187
+ {
188
+ "epoch": 0.10770505385252693,
189
+ "grad_norm": 0.7035478353500366,
190
+ "learning_rate": 9.99866112295902e-05,
191
+ "loss": 0.49724931716918946,
192
+ "step": 130
193
+ },
194
+ {
195
+ "epoch": 0.11184755592377796,
196
+ "grad_norm": 0.6950880885124207,
197
+ "learning_rate": 9.996464786686644e-05,
198
+ "loss": 0.4550027370452881,
199
+ "step": 135
200
+ },
201
+ {
202
+ "epoch": 0.115990057995029,
203
+ "grad_norm": 1.0088214874267578,
204
+ "learning_rate": 9.993223163862385e-05,
205
+ "loss": 0.5083928108215332,
206
+ "step": 140
207
+ },
208
+ {
209
+ "epoch": 0.12013256006628004,
210
+ "grad_norm": 0.747430145740509,
211
+ "learning_rate": 9.988936932650681e-05,
212
+ "loss": 0.43213858604431155,
213
+ "step": 145
214
+ },
215
+ {
216
+ "epoch": 0.12427506213753108,
217
+ "grad_norm": 0.9024979472160339,
218
+ "learning_rate": 9.983606989753509e-05,
219
+ "loss": 0.5470420360565186,
220
+ "step": 150
221
+ },
222
+ {
223
+ "epoch": 0.12841756420878211,
224
+ "grad_norm": 1.1015739440917969,
225
+ "learning_rate": 9.977234450222783e-05,
226
+ "loss": 0.4933629989624023,
227
+ "step": 155
228
+ },
229
+ {
230
+ "epoch": 0.13256006628003314,
231
+ "grad_norm": 0.8720569610595703,
232
+ "learning_rate": 9.969820647227086e-05,
233
+ "loss": 0.5295196533203125,
234
+ "step": 160
235
+ },
236
+ {
237
+ "epoch": 0.13670256835128416,
238
+ "grad_norm": 0.7969361543655396,
239
+ "learning_rate": 9.961367131772759e-05,
240
+ "loss": 0.4577418327331543,
241
+ "step": 165
242
+ },
243
+ {
244
+ "epoch": 0.14084507042253522,
245
+ "grad_norm": 3.101267099380493,
246
+ "learning_rate": 9.951875672379424e-05,
247
+ "loss": 0.6081419467926026,
248
+ "step": 170
249
+ },
250
+ {
251
+ "epoch": 0.14498757249378624,
252
+ "grad_norm": 0.7658870816230774,
253
+ "learning_rate": 9.941348254710003e-05,
254
+ "loss": 0.4367526531219482,
255
+ "step": 175
256
+ },
257
+ {
258
+ "epoch": 0.1491300745650373,
259
+ "grad_norm": 0.6250601410865784,
260
+ "learning_rate": 9.929787081155304e-05,
261
+ "loss": 0.3846379041671753,
262
+ "step": 180
263
+ },
264
+ {
265
+ "epoch": 0.15327257663628832,
266
+ "grad_norm": 0.8455803990364075,
267
+ "learning_rate": 9.91719457037327e-05,
268
+ "loss": 0.4439352035522461,
269
+ "step": 185
270
+ },
271
+ {
272
+ "epoch": 0.15741507870753935,
273
+ "grad_norm": 0.5648259520530701,
274
+ "learning_rate": 9.903573356782977e-05,
275
+ "loss": 0.41419034004211425,
276
+ "step": 190
277
+ },
278
+ {
279
+ "epoch": 0.1615575807787904,
280
+ "grad_norm": 1.176491618156433,
281
+ "learning_rate": 9.888926290013521e-05,
282
+ "loss": 0.4723679542541504,
283
+ "step": 195
284
+ },
285
+ {
286
+ "epoch": 0.16570008285004142,
287
+ "grad_norm": 0.5706663727760315,
288
+ "learning_rate": 9.873256434307828e-05,
289
+ "loss": 0.4028864860534668,
290
+ "step": 200
291
+ },
292
+ {
293
+ "epoch": 0.16984258492129245,
294
+ "grad_norm": 0.5537962317466736,
295
+ "learning_rate": 9.856567067881628e-05,
296
+ "loss": 0.3607978820800781,
297
+ "step": 205
298
+ },
299
+ {
300
+ "epoch": 0.1739850869925435,
301
+ "grad_norm": 0.6196736097335815,
302
+ "learning_rate": 9.838861682237618e-05,
303
+ "loss": 0.42556142807006836,
304
+ "step": 210
305
+ },
306
+ {
307
+ "epoch": 0.17812758906379453,
308
+ "grad_norm": 0.6695680022239685,
309
+ "learning_rate": 9.820143981435031e-05,
310
+ "loss": 0.41333789825439454,
311
+ "step": 215
312
+ },
313
+ {
314
+ "epoch": 0.18227009113504558,
315
+ "grad_norm": 0.655034065246582,
316
+ "learning_rate": 9.800417881314716e-05,
317
+ "loss": 0.5096014976501465,
318
+ "step": 220
319
+ },
320
+ {
321
+ "epoch": 0.1864125932062966,
322
+ "grad_norm": 0.5938177704811096,
323
+ "learning_rate": 9.779687508679935e-05,
324
+ "loss": 0.5216481685638428,
325
+ "step": 225
326
+ },
327
+ {
328
+ "epoch": 0.19055509527754763,
329
+ "grad_norm": 0.7804147601127625,
330
+ "learning_rate": 9.75795720043301e-05,
331
+ "loss": 0.43359966278076173,
332
+ "step": 230
333
+ },
334
+ {
335
+ "epoch": 0.19469759734879868,
336
+ "grad_norm": 0.7048928141593933,
337
+ "learning_rate": 9.73523150266802e-05,
338
+ "loss": 0.4395446300506592,
339
+ "step": 235
340
+ },
341
+ {
342
+ "epoch": 0.1988400994200497,
343
+ "grad_norm": 0.6047908067703247,
344
+ "learning_rate": 9.711515169719732e-05,
345
+ "loss": 0.4992673397064209,
346
+ "step": 240
347
+ },
348
+ {
349
+ "epoch": 0.20298260149130073,
350
+ "grad_norm": 0.6019694209098816,
351
+ "learning_rate": 9.686813163168973e-05,
352
+ "loss": 0.41515302658081055,
353
+ "step": 245
354
+ },
355
+ {
356
+ "epoch": 0.2071251035625518,
357
+ "grad_norm": 0.8708873391151428,
358
+ "learning_rate": 9.661130650804643e-05,
359
+ "loss": 0.4704007625579834,
360
+ "step": 250
361
+ },
362
+ {
363
+ "epoch": 0.2112676056338028,
364
+ "grad_norm": 0.6083199381828308,
365
+ "learning_rate": 9.634473005542579e-05,
366
+ "loss": 0.5111989974975586,
367
+ "step": 255
368
+ },
369
+ {
370
+ "epoch": 0.21541010770505387,
371
+ "grad_norm": 0.5877307653427124,
372
+ "learning_rate": 9.606845804301524e-05,
373
+ "loss": 0.48946352005004884,
374
+ "step": 260
375
+ },
376
+ {
377
+ "epoch": 0.2195526097763049,
378
+ "grad_norm": 0.6169144511222839,
379
+ "learning_rate": 9.578254826836391e-05,
380
+ "loss": 0.39699459075927734,
381
+ "step": 265
382
+ },
383
+ {
384
+ "epoch": 0.22369511184755592,
385
+ "grad_norm": 0.7712729573249817,
386
+ "learning_rate": 9.548706054529124e-05,
387
+ "loss": 0.43967700004577637,
388
+ "step": 270
389
+ },
390
+ {
391
+ "epoch": 0.22783761391880697,
392
+ "grad_norm": 0.7823882102966309,
393
+ "learning_rate": 9.518205669137352e-05,
394
+ "loss": 0.39926810264587403,
395
+ "step": 275
396
+ },
397
+ {
398
+ "epoch": 0.231980115990058,
399
+ "grad_norm": 0.6713236570358276,
400
+ "learning_rate": 9.486760051501139e-05,
401
+ "loss": 0.454238748550415,
402
+ "step": 280
403
+ },
404
+ {
405
+ "epoch": 0.23612261806130902,
406
+ "grad_norm": 0.8165543079376221,
407
+ "learning_rate": 9.454375780208071e-05,
408
+ "loss": 0.46351184844970705,
409
+ "step": 285
410
+ },
411
+ {
412
+ "epoch": 0.24026512013256007,
413
+ "grad_norm": 0.37962615489959717,
414
+ "learning_rate": 9.421059630216991e-05,
415
+ "loss": 0.4180287837982178,
416
+ "step": 290
417
+ },
418
+ {
419
+ "epoch": 0.2444076222038111,
420
+ "grad_norm": 0.6618972420692444,
421
+ "learning_rate": 9.38681857144063e-05,
422
+ "loss": 0.3922520637512207,
423
+ "step": 295
424
+ },
425
+ {
426
+ "epoch": 0.24855012427506215,
427
+ "grad_norm": 0.42513608932495117,
428
+ "learning_rate": 9.351659767287478e-05,
429
+ "loss": 0.46528077125549316,
430
+ "step": 300
431
+ },
432
+ {
433
+ "epoch": 0.2526926263463132,
434
+ "grad_norm": 0.5760522484779358,
435
+ "learning_rate": 9.315590573163153e-05,
436
+ "loss": 0.4353321075439453,
437
+ "step": 305
438
+ },
439
+ {
440
+ "epoch": 0.25683512841756423,
441
+ "grad_norm": 0.7091104984283447,
442
+ "learning_rate": 9.278618534931611e-05,
443
+ "loss": 0.4817354679107666,
444
+ "step": 310
445
+ },
446
+ {
447
+ "epoch": 0.2609776304888152,
448
+ "grad_norm": 0.7984012365341187,
449
+ "learning_rate": 9.240751387336522e-05,
450
+ "loss": 0.45444478988647463,
451
+ "step": 315
452
+ },
453
+ {
454
+ "epoch": 0.2651201325600663,
455
+ "grad_norm": 0.7693517804145813,
456
+ "learning_rate": 9.201997052383106e-05,
457
+ "loss": 0.4780622959136963,
458
+ "step": 320
459
+ },
460
+ {
461
+ "epoch": 0.26926263463131733,
462
+ "grad_norm": 0.5142762660980225,
463
+ "learning_rate": 9.162363637680822e-05,
464
+ "loss": 0.4113470554351807,
465
+ "step": 325
466
+ },
467
+ {
468
+ "epoch": 0.27340513670256833,
469
+ "grad_norm": 0.682898223400116,
470
+ "learning_rate": 9.121859434747204e-05,
471
+ "loss": 0.48437914848327634,
472
+ "step": 330
473
+ },
474
+ {
475
+ "epoch": 0.2775476387738194,
476
+ "grad_norm": 1.3431150913238525,
477
+ "learning_rate": 9.080492917273237e-05,
478
+ "loss": 0.5317914962768555,
479
+ "step": 335
480
+ },
481
+ {
482
+ "epoch": 0.28169014084507044,
483
+ "grad_norm": 0.7310293316841125,
484
+ "learning_rate": 9.038272739350618e-05,
485
+ "loss": 0.41095829010009766,
486
+ "step": 340
487
+ },
488
+ {
489
+ "epoch": 0.28583264291632143,
490
+ "grad_norm": 0.564328134059906,
491
+ "learning_rate": 8.995207733661267e-05,
492
+ "loss": 0.4005306243896484,
493
+ "step": 345
494
+ },
495
+ {
496
+ "epoch": 0.2899751449875725,
497
+ "grad_norm": 0.6681087017059326,
498
+ "learning_rate": 8.951306909629492e-05,
499
+ "loss": 0.399173641204834,
500
+ "step": 350
501
+ },
502
+ {
503
+ "epoch": 0.29411764705882354,
504
+ "grad_norm": 0.5092662572860718,
505
+ "learning_rate": 8.906579451537171e-05,
506
+ "loss": 0.3901823043823242,
507
+ "step": 355
508
+ },
509
+ {
510
+ "epoch": 0.2982601491300746,
511
+ "grad_norm": 1.0772058963775635,
512
+ "learning_rate": 8.861034716602343e-05,
513
+ "loss": 0.4496616840362549,
514
+ "step": 360
515
+ },
516
+ {
517
+ "epoch": 0.3024026512013256,
518
+ "grad_norm": 0.596607506275177,
519
+ "learning_rate": 8.814682233021636e-05,
520
+ "loss": 0.43782577514648435,
521
+ "step": 365
522
+ },
523
+ {
524
+ "epoch": 0.30654515327257664,
525
+ "grad_norm": 0.6527637243270874,
526
+ "learning_rate": 8.76753169797692e-05,
527
+ "loss": 0.4539346218109131,
528
+ "step": 370
529
+ },
530
+ {
531
+ "epoch": 0.3106876553438277,
532
+ "grad_norm": 0.6109674572944641,
533
+ "learning_rate": 8.719592975606601e-05,
534
+ "loss": 0.4309091091156006,
535
+ "step": 375
536
+ },
537
+ {
538
+ "epoch": 0.3148301574150787,
539
+ "grad_norm": 0.4367820918560028,
540
+ "learning_rate": 8.670876094941991e-05,
541
+ "loss": 0.4418576717376709,
542
+ "step": 380
543
+ },
544
+ {
545
+ "epoch": 0.31897265948632975,
546
+ "grad_norm": 0.6753432154655457,
547
+ "learning_rate": 8.62139124780919e-05,
548
+ "loss": 0.5508285522460937,
549
+ "step": 385
550
+ },
551
+ {
552
+ "epoch": 0.3231151615575808,
553
+ "grad_norm": 0.605097234249115,
554
+ "learning_rate": 8.571148786696894e-05,
555
+ "loss": 0.47382965087890627,
556
+ "step": 390
557
+ },
558
+ {
559
+ "epoch": 0.3272576636288318,
560
+ "grad_norm": 0.4321838617324829,
561
+ "learning_rate": 8.520159222590605e-05,
562
+ "loss": 0.3830132961273193,
563
+ "step": 395
564
+ },
565
+ {
566
+ "epoch": 0.33140016570008285,
567
+ "grad_norm": 0.6553069353103638,
568
+ "learning_rate": 8.468433222773677e-05,
569
+ "loss": 0.4628459930419922,
570
+ "step": 400
571
+ }
572
+ ],
573
+ "logging_steps": 5,
574
+ "max_steps": 1207,
575
+ "num_input_tokens_seen": 0,
576
+ "num_train_epochs": 1,
577
+ "save_steps": 200,
578
+ "stateful_callbacks": {
579
+ "TrainerControl": {
580
+ "args": {
581
+ "should_epoch_stop": false,
582
+ "should_evaluate": false,
583
+ "should_log": false,
584
+ "should_save": true,
585
+ "should_training_stop": false
586
+ },
587
+ "attributes": {}
588
+ }
589
+ },
590
+ "total_flos": 8.375348468992243e+16,
591
+ "train_batch_size": 1,
592
+ "trial_name": null,
593
+ "trial_params": null
594
+ }