yukidump commited on
Commit
67efb10
β€’
1 Parent(s): 3080479

ppllava_vicuna_final

Browse files
Files changed (29) hide show
  1. checkpoint-160/trainer_state.json +0 -0
  2. ppllava_vicuna7b_image_video/config.json +98 -0
  3. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/generation_config.json +0 -0
  4. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00001-of-00003.safetensors +1 -1
  5. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00002-of-00003.safetensors +1 -1
  6. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00003-of-00003.safetensors +1 -1
  7. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model.safetensors.index.json +0 -0
  8. ppllava_vicuna7b_image_video/trainer_state.json +485 -0
  9. {checkpoint-160 β†’ ppllava_vicuna7b_image_video}/training_args.bin +2 -2
  10. {checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/added_tokens.json +0 -0
  11. ppllava_vicuna7b_image_video_dpo/config.json +97 -0
  12. ppllava_vicuna7b_image_video_dpo/generation_config.json +7 -0
  13. ppllava_vicuna7b_image_video_dpo/model-00001-of-00003.safetensors +3 -0
  14. ppllava_vicuna7b_image_video_dpo/model-00002-of-00003.safetensors +3 -0
  15. ppllava_vicuna7b_image_video_dpo/model-00003-of-00003.safetensors +3 -0
  16. ppllava_vicuna7b_image_video_dpo/model.safetensors.index.json +0 -0
  17. {checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/special_tokens_map.json +0 -0
  18. {checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/tokenizer.model +0 -0
  19. {checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/tokenizer_config.json +0 -0
  20. ppllava_vicuna7b_image_video_dpo/trainer_state.json +2181 -0
  21. ppllava_vicuna7b_image_video_dpo/training_args.bin +3 -0
  22. {checkpoint-160 β†’ ppllava_vicuna7b_image_video_multiimage}/config.json +1 -1
  23. ppllava_vicuna7b_image_video_multiimage/generation_config.json +7 -0
  24. ppllava_vicuna7b_image_video_multiimage/model-00001-of-00003.safetensors +3 -0
  25. ppllava_vicuna7b_image_video_multiimage/model-00002-of-00003.safetensors +3 -0
  26. ppllava_vicuna7b_image_video_multiimage/model-00003-of-00003.safetensors +3 -0
  27. ppllava_vicuna7b_image_video_multiimage/model.safetensors.index.json +0 -0
  28. ppllava_vicuna7b_image_video_multiimage/trainer_state.json +821 -0
  29. ppllava_vicuna7b_image_video_multiimage/training_args.bin +3 -0
checkpoint-160/trainer_state.json DELETED
The diff for this file is too large to render. See raw diff
 
ppllava_vicuna7b_image_video/config.json ADDED
@@ -0,0 +1,98 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlavaInterleaveForConditionalGeneration"
4
+ ],
5
+ "btadapter": true,
6
+ "btadapter_depth": 4,
7
+ "clip_post_pretrain": null,
8
+ "clip_weight": "openai/clip-vit-large-patch14-336",
9
+ "frame_shape": [
10
+ 24,
11
+ 24
12
+ ],
13
+ "hidden_size": 4096,
14
+ "ignore_index": -100,
15
+ "image_grid_pinpoints": [
16
+ [
17
+ 336,
18
+ 672
19
+ ],
20
+ [
21
+ 672,
22
+ 336
23
+ ],
24
+ [
25
+ 672,
26
+ 672
27
+ ],
28
+ [
29
+ 1008,
30
+ 336
31
+ ],
32
+ [
33
+ 336,
34
+ 1008
35
+ ]
36
+ ],
37
+ "image_pooling_kernel": [
38
+ 1,
39
+ 3,
40
+ 3
41
+ ],
42
+ "image_pooling_stride": [
43
+ 1,
44
+ 3,
45
+ 3
46
+ ],
47
+ "image_token_index": 32000,
48
+ "long_clip": true,
49
+ "max_T": 64,
50
+ "model_type": "llava_next",
51
+ "pad_token_id": 0,
52
+ "pllava_pooling_shape": null,
53
+ "pooling": "clipST_3d",
54
+ "pooling_kernel": [
55
+ 2,
56
+ 3,
57
+ 3
58
+ ],
59
+ "pooling_stride": [
60
+ 2,
61
+ 3,
62
+ 3
63
+ ],
64
+ "pooling_temp": 0.01,
65
+ "projector_hidden_act": "gelu",
66
+ "text_config": {
67
+ "_name_or_path": "lmsys/vicuna-7b-v1.5",
68
+ "architectures": [
69
+ "LlamaForCausalLM"
70
+ ],
71
+ "max_position_embeddings": 4096,
72
+ "model_type": "llama",
73
+ "pad_token_id": 0,
74
+ "rms_norm_eps": 1e-05,
75
+ "torch_dtype": "float16",
76
+ "vocab_size": 32064
77
+ },
78
+ "torch_dtype": "bfloat16",
79
+ "transformers_version": "4.40.0",
80
+ "use_image_newline_parameter": true,
81
+ "video_input": "mean",
82
+ "vision_config": {
83
+ "depth": 4,
84
+ "hidden_size": 1024,
85
+ "image_size": 336,
86
+ "intermediate_size": 4096,
87
+ "max_T": 64,
88
+ "model_type": "clip_vision_model",
89
+ "num_attention_heads": 16,
90
+ "num_hidden_layers": 24,
91
+ "patch_size": 14,
92
+ "projection_dim": 768,
93
+ "vocab_size": 32000
94
+ },
95
+ "vision_feature_layer": -2,
96
+ "vision_feature_select_strategy": "default",
97
+ "vocab_size": 32064
98
+ }
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/generation_config.json RENAMED
File without changes
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00001-of-00003.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:74c7f5d993a912aa5407d99fe5b88b36c727a99eab963941297cac08432831c3
3
  size 4980641554
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2468064fc5ed3bc395743532b2d6565ee20a678efb5a292527aa25d80a547574
3
  size 4980641554
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00002-of-00003.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0c2bdfeb77728fd9619f3bb3f4f7f5f4352ff82f3f730a314d4bd9d9dd641651
3
  size 4957878536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6093c7766f8f45b2f51be4ebb77ed2df5c49b410970e0620a6d5d4ff7e8b0c13
3
  size 4957878536
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model-00003-of-00003.safetensors RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b57c8234e6548525e69218c26509caed76ebf9911b1190db753fa142814aaa63
3
  size 4580905400
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e74b0d8a25e3c65d3965fcfdaf83dd54ef2f8087c14d12f202a80af25e0ad167
3
  size 4580905400
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/model.safetensors.index.json RENAMED
File without changes
ppllava_vicuna7b_image_video/trainer_state.json ADDED
@@ -0,0 +1,485 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.9998461775111521,
5
+ "eval_steps": 500,
6
+ "global_step": 3250,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.015382248884786957,
13
+ "grad_norm": 1.2864877128592522,
14
+ "learning_rate": 1.0204081632653063e-05,
15
+ "loss": 1.211,
16
+ "step": 50
17
+ },
18
+ {
19
+ "epoch": 0.030764497769573913,
20
+ "grad_norm": 1.227813083166002,
21
+ "learning_rate": 1.999998013186146e-05,
22
+ "loss": 1.0177,
23
+ "step": 100
24
+ },
25
+ {
26
+ "epoch": 0.046146746654360866,
27
+ "grad_norm": 1.16006238153043,
28
+ "learning_rate": 1.9986572140101312e-05,
29
+ "loss": 0.9909,
30
+ "step": 150
31
+ },
32
+ {
33
+ "epoch": 0.06152899553914783,
34
+ "grad_norm": 1.2120743368126807,
35
+ "learning_rate": 1.9948367447837503e-05,
36
+ "loss": 0.9853,
37
+ "step": 200
38
+ },
39
+ {
40
+ "epoch": 0.07691124442393478,
41
+ "grad_norm": 1.0434928463049529,
42
+ "learning_rate": 1.9885460917481137e-05,
43
+ "loss": 0.9766,
44
+ "step": 250
45
+ },
46
+ {
47
+ "epoch": 0.09229349330872173,
48
+ "grad_norm": 1.0799628915416184,
49
+ "learning_rate": 1.9798008746210894e-05,
50
+ "loss": 0.9819,
51
+ "step": 300
52
+ },
53
+ {
54
+ "epoch": 0.10767574219350869,
55
+ "grad_norm": 1.0390775292248948,
56
+ "learning_rate": 1.968622807813479e-05,
57
+ "loss": 0.9678,
58
+ "step": 350
59
+ },
60
+ {
61
+ "epoch": 0.12305799107829565,
62
+ "grad_norm": 1.3974223368156584,
63
+ "learning_rate": 1.955039646512041e-05,
64
+ "loss": 0.9782,
65
+ "step": 400
66
+ },
67
+ {
68
+ "epoch": 0.1384402399630826,
69
+ "grad_norm": 1.1107339192286112,
70
+ "learning_rate": 1.9390851177632496e-05,
71
+ "loss": 0.9602,
72
+ "step": 450
73
+ },
74
+ {
75
+ "epoch": 0.15382248884786956,
76
+ "grad_norm": 1.0579993349228267,
77
+ "learning_rate": 1.9207988367288956e-05,
78
+ "loss": 0.9571,
79
+ "step": 500
80
+ },
81
+ {
82
+ "epoch": 0.1692047377326565,
83
+ "grad_norm": 1.1144151233888806,
84
+ "learning_rate": 1.9002262083214728e-05,
85
+ "loss": 0.9669,
86
+ "step": 550
87
+ },
88
+ {
89
+ "epoch": 0.18458698661744347,
90
+ "grad_norm": 1.187591950721483,
91
+ "learning_rate": 1.8774183144635894e-05,
92
+ "loss": 0.958,
93
+ "step": 600
94
+ },
95
+ {
96
+ "epoch": 0.19996923550223042,
97
+ "grad_norm": 1.3767826488829,
98
+ "learning_rate": 1.8524317872513434e-05,
99
+ "loss": 0.9481,
100
+ "step": 650
101
+ },
102
+ {
103
+ "epoch": 0.21535148438701737,
104
+ "grad_norm": 1.0853939205331895,
105
+ "learning_rate": 1.8253286683365855e-05,
106
+ "loss": 0.9434,
107
+ "step": 700
108
+ },
109
+ {
110
+ "epoch": 0.23073373327180433,
111
+ "grad_norm": 1.0358359930096885,
112
+ "learning_rate": 1.7961762548772413e-05,
113
+ "loss": 0.9524,
114
+ "step": 750
115
+ },
116
+ {
117
+ "epoch": 0.2461159821565913,
118
+ "grad_norm": 1.0885430781645802,
119
+ "learning_rate": 1.7650469324381843e-05,
120
+ "loss": 0.9443,
121
+ "step": 800
122
+ },
123
+ {
124
+ "epoch": 0.26149823104137826,
125
+ "grad_norm": 1.139432989706956,
126
+ "learning_rate": 1.732017995257575e-05,
127
+ "loss": 0.9398,
128
+ "step": 850
129
+ },
130
+ {
131
+ "epoch": 0.2768804799261652,
132
+ "grad_norm": 1.1616765089635157,
133
+ "learning_rate": 1.697171454324943e-05,
134
+ "loss": 0.9318,
135
+ "step": 900
136
+ },
137
+ {
138
+ "epoch": 0.29226272881095217,
139
+ "grad_norm": 1.102418985915195,
140
+ "learning_rate": 1.660593833747557e-05,
141
+ "loss": 0.9348,
142
+ "step": 950
143
+ },
144
+ {
145
+ "epoch": 0.3076449776957391,
146
+ "grad_norm": 1.0377834162761999,
147
+ "learning_rate": 1.6223759559107028e-05,
148
+ "loss": 0.9206,
149
+ "step": 1000
150
+ },
151
+ {
152
+ "epoch": 0.3230272265805261,
153
+ "grad_norm": 1.103549274843009,
154
+ "learning_rate": 1.5826127159653203e-05,
155
+ "loss": 0.9353,
156
+ "step": 1050
157
+ },
158
+ {
159
+ "epoch": 0.338409475465313,
160
+ "grad_norm": 1.0090406165137933,
161
+ "learning_rate": 1.5414028462029454e-05,
162
+ "loss": 0.9363,
163
+ "step": 1100
164
+ },
165
+ {
166
+ "epoch": 0.3537917243501,
167
+ "grad_norm": 1.1329344585144494,
168
+ "learning_rate": 1.4988486709030166e-05,
169
+ "loss": 0.9347,
170
+ "step": 1150
171
+ },
172
+ {
173
+ "epoch": 0.36917397323488693,
174
+ "grad_norm": 1.0307615452725163,
175
+ "learning_rate": 1.4550558522612584e-05,
176
+ "loss": 0.9231,
177
+ "step": 1200
178
+ },
179
+ {
180
+ "epoch": 0.3845562221196739,
181
+ "grad_norm": 1.1133951229282608,
182
+ "learning_rate": 1.410133128030009e-05,
183
+ "loss": 0.9281,
184
+ "step": 1250
185
+ },
186
+ {
187
+ "epoch": 0.39993847100446084,
188
+ "grad_norm": 1.0490189419573543,
189
+ "learning_rate": 1.3641920415219224e-05,
190
+ "loss": 0.9229,
191
+ "step": 1300
192
+ },
193
+ {
194
+ "epoch": 0.4153207198892478,
195
+ "grad_norm": 1.0225887710538208,
196
+ "learning_rate": 1.3173466646474575e-05,
197
+ "loss": 0.9115,
198
+ "step": 1350
199
+ },
200
+ {
201
+ "epoch": 0.43070296877403474,
202
+ "grad_norm": 1.0818630597333305,
203
+ "learning_rate": 1.2697133146738452e-05,
204
+ "loss": 0.9124,
205
+ "step": 1400
206
+ },
207
+ {
208
+ "epoch": 0.4460852176588217,
209
+ "grad_norm": 1.1222485167556653,
210
+ "learning_rate": 1.2214102654088248e-05,
211
+ "loss": 0.9166,
212
+ "step": 1450
213
+ },
214
+ {
215
+ "epoch": 0.46146746654360865,
216
+ "grad_norm": 0.9980409625749838,
217
+ "learning_rate": 1.1725574535262863e-05,
218
+ "loss": 0.9047,
219
+ "step": 1500
220
+ },
221
+ {
222
+ "epoch": 0.4768497154283956,
223
+ "grad_norm": 1.0397039742249545,
224
+ "learning_rate": 1.1232761807630032e-05,
225
+ "loss": 0.9079,
226
+ "step": 1550
227
+ },
228
+ {
229
+ "epoch": 0.4922319643131826,
230
+ "grad_norm": 1.0715968646982938,
231
+ "learning_rate": 1.0736888127259132e-05,
232
+ "loss": 0.9061,
233
+ "step": 1600
234
+ },
235
+ {
236
+ "epoch": 0.5076142131979695,
237
+ "grad_norm": 1.0724763367894545,
238
+ "learning_rate": 1.023918475057803e-05,
239
+ "loss": 0.8947,
240
+ "step": 1650
241
+ },
242
+ {
243
+ "epoch": 0.5229964620827565,
244
+ "grad_norm": 1.03882946183293,
245
+ "learning_rate": 9.740887477158194e-06,
246
+ "loss": 0.9053,
247
+ "step": 1700
248
+ },
249
+ {
250
+ "epoch": 0.5383787109675434,
251
+ "grad_norm": 1.0204487030578802,
252
+ "learning_rate": 9.24323358121919e-06,
253
+ "loss": 0.9016,
254
+ "step": 1750
255
+ },
256
+ {
257
+ "epoch": 0.5537609598523304,
258
+ "grad_norm": 1.0085542891249055,
259
+ "learning_rate": 8.747458739471603e-06,
260
+ "loss": 0.8988,
261
+ "step": 1800
262
+ },
263
+ {
264
+ "epoch": 0.5691432087371173,
265
+ "grad_norm": 1.1293770623412562,
266
+ "learning_rate": 8.254793962926579e-06,
267
+ "loss": 0.8967,
268
+ "step": 1850
269
+ },
270
+ {
271
+ "epoch": 0.5845254576219043,
272
+ "grad_norm": 1.1860769177336596,
273
+ "learning_rate": 7.76646254029032e-06,
274
+ "loss": 0.8967,
275
+ "step": 1900
276
+ },
277
+ {
278
+ "epoch": 0.5999077065066912,
279
+ "grad_norm": 1.062132771926321,
280
+ "learning_rate": 7.283677000533106e-06,
281
+ "loss": 0.8936,
282
+ "step": 1950
283
+ },
284
+ {
285
+ "epoch": 0.6152899553914782,
286
+ "grad_norm": 1.1679358844534626,
287
+ "learning_rate": 6.807636102174764e-06,
288
+ "loss": 0.8928,
289
+ "step": 2000
290
+ },
291
+ {
292
+ "epoch": 0.6306722042762651,
293
+ "grad_norm": 1.0610336394985995,
294
+ "learning_rate": 6.339521856762254e-06,
295
+ "loss": 0.8845,
296
+ "step": 2050
297
+ },
298
+ {
299
+ "epoch": 0.6460544531610521,
300
+ "grad_norm": 1.084745782348564,
301
+ "learning_rate": 5.880496593930021e-06,
302
+ "loss": 0.8869,
303
+ "step": 2100
304
+ },
305
+ {
306
+ "epoch": 0.6614367020458392,
307
+ "grad_norm": 0.9948936563241075,
308
+ "learning_rate": 5.431700075330654e-06,
309
+ "loss": 0.8817,
310
+ "step": 2150
311
+ },
312
+ {
313
+ "epoch": 0.676818950930626,
314
+ "grad_norm": 1.0955833872820746,
315
+ "learning_rate": 4.9942466646019e-06,
316
+ "loss": 0.8857,
317
+ "step": 2200
318
+ },
319
+ {
320
+ "epoch": 0.6922011998154131,
321
+ "grad_norm": 1.0328397346172227,
322
+ "learning_rate": 4.569222560397126e-06,
323
+ "loss": 0.8842,
324
+ "step": 2250
325
+ },
326
+ {
327
+ "epoch": 0.7075834487002,
328
+ "grad_norm": 1.0439140807601641,
329
+ "learning_rate": 4.157683099349444e-06,
330
+ "loss": 0.8797,
331
+ "step": 2300
332
+ },
333
+ {
334
+ "epoch": 0.722965697584987,
335
+ "grad_norm": 1.0659188075287298,
336
+ "learning_rate": 3.760650135666476e-06,
337
+ "loss": 0.8765,
338
+ "step": 2350
339
+ },
340
+ {
341
+ "epoch": 0.7383479464697739,
342
+ "grad_norm": 1.0691950735737183,
343
+ "learning_rate": 3.3791095038620515e-06,
344
+ "loss": 0.8715,
345
+ "step": 2400
346
+ },
347
+ {
348
+ "epoch": 0.7537301953545609,
349
+ "grad_norm": 1.0882575411106976,
350
+ "learning_rate": 3.0140085709249666e-06,
351
+ "loss": 0.8754,
352
+ "step": 2450
353
+ },
354
+ {
355
+ "epoch": 0.7691124442393478,
356
+ "grad_norm": 1.1112353348166593,
357
+ "learning_rate": 2.6662538840027984e-06,
358
+ "loss": 0.873,
359
+ "step": 2500
360
+ },
361
+ {
362
+ "epoch": 0.7844946931241348,
363
+ "grad_norm": 1.1312531360561675,
364
+ "learning_rate": 2.3367089194415226e-06,
365
+ "loss": 0.8776,
366
+ "step": 2550
367
+ },
368
+ {
369
+ "epoch": 0.7998769420089217,
370
+ "grad_norm": 1.118822941453432,
371
+ "learning_rate": 2.026191938770128e-06,
372
+ "loss": 0.8769,
373
+ "step": 2600
374
+ },
375
+ {
376
+ "epoch": 0.8152591908937087,
377
+ "grad_norm": 1.0617562102838363,
378
+ "learning_rate": 1.7354739569538081e-06,
379
+ "loss": 0.8611,
380
+ "step": 2650
381
+ },
382
+ {
383
+ "epoch": 0.8306414397784956,
384
+ "grad_norm": 1.1190891753348724,
385
+ "learning_rate": 1.4652768279605734e-06,
386
+ "loss": 0.8778,
387
+ "step": 2700
388
+ },
389
+ {
390
+ "epoch": 0.8460236886632826,
391
+ "grad_norm": 1.1025665252752193,
392
+ "learning_rate": 1.2162714523948104e-06,
393
+ "loss": 0.8582,
394
+ "step": 2750
395
+ },
396
+ {
397
+ "epoch": 0.8614059375480695,
398
+ "grad_norm": 1.0621744666965152,
399
+ "learning_rate": 9.89076111648284e-07,
400
+ "loss": 0.8643,
401
+ "step": 2800
402
+ },
403
+ {
404
+ "epoch": 0.8767881864328565,
405
+ "grad_norm": 0.9763313630276178,
406
+ "learning_rate": 7.842549327048366e-07,
407
+ "loss": 0.8667,
408
+ "step": 2850
409
+ },
410
+ {
411
+ "epoch": 0.8921704353176434,
412
+ "grad_norm": 1.0393762329975094,
413
+ "learning_rate": 6.023164874107202e-07,
414
+ "loss": 0.8677,
415
+ "step": 2900
416
+ },
417
+ {
418
+ "epoch": 0.9075526842024304,
419
+ "grad_norm": 1.1026189621389317,
420
+ "learning_rate": 4.437125296885847e-07,
421
+ "loss": 0.8649,
422
+ "step": 2950
423
+ },
424
+ {
425
+ "epoch": 0.9229349330872173,
426
+ "grad_norm": 1.073850367410434,
427
+ "learning_rate": 3.088368738305747e-07,
428
+ "loss": 0.8771,
429
+ "step": 3000
430
+ },
431
+ {
432
+ "epoch": 0.9383171819720043,
433
+ "grad_norm": 1.0327578416174346,
434
+ "learning_rate": 1.9802441665580208e-07,
435
+ "loss": 0.8678,
436
+ "step": 3050
437
+ },
438
+ {
439
+ "epoch": 0.9536994308567912,
440
+ "grad_norm": 1.035131432577654,
441
+ "learning_rate": 1.1155030596013638e-07,
442
+ "loss": 0.8662,
443
+ "step": 3100
444
+ },
445
+ {
446
+ "epoch": 0.9690816797415782,
447
+ "grad_norm": 1.0361099744000193,
448
+ "learning_rate": 4.962925732307278e-08,
449
+ "loss": 0.8588,
450
+ "step": 3150
451
+ },
452
+ {
453
+ "epoch": 0.9844639286263652,
454
+ "grad_norm": 1.087919608094763,
455
+ "learning_rate": 1.2415020968049585e-08,
456
+ "loss": 0.8733,
457
+ "step": 3200
458
+ },
459
+ {
460
+ "epoch": 0.9998461775111521,
461
+ "grad_norm": 1.0473814351391442,
462
+ "learning_rate": 0.0,
463
+ "loss": 0.8751,
464
+ "step": 3250
465
+ },
466
+ {
467
+ "epoch": 0.9998461775111521,
468
+ "step": 3250,
469
+ "total_flos": 1.7053270040117248e+16,
470
+ "train_loss": 0.9156783118614784,
471
+ "train_runtime": 46310.2792,
472
+ "train_samples_per_second": 0.562,
473
+ "train_steps_per_second": 0.07
474
+ }
475
+ ],
476
+ "logging_steps": 50,
477
+ "max_steps": 3250,
478
+ "num_input_tokens_seen": 0,
479
+ "num_train_epochs": 1,
480
+ "save_steps": 500,
481
+ "total_flos": 1.7053270040117248e+16,
482
+ "train_batch_size": 4,
483
+ "trial_name": null,
484
+ "trial_params": null
485
+ }
{checkpoint-160 β†’ ppllava_vicuna7b_image_video}/training_args.bin RENAMED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6d53e83224f2769f2c964ded43fbc4dbe376aaf199dca61db75f92ac145c4ee2
3
- size 6712
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3db21088a03dab3e9042367b9f7e3bb67dfc608a5563e28fb3fb76fa7ab52d6c
3
+ size 7032
{checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/added_tokens.json RENAMED
File without changes
ppllava_vicuna7b_image_video_dpo/config.json ADDED
@@ -0,0 +1,97 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlavaInterleaveForConditionalGeneration"
4
+ ],
5
+ "btadapter": true,
6
+ "btadapter_depth": 4,
7
+ "clip_post_pretrain": null,
8
+ "clip_weight": "openai/clip-vit-large-patch14-336",
9
+ "frame_shape": [
10
+ 24,
11
+ 24
12
+ ],
13
+ "hidden_size": 4096,
14
+ "ignore_index": -100,
15
+ "image_grid_pinpoints": [
16
+ [
17
+ 336,
18
+ 672
19
+ ],
20
+ [
21
+ 672,
22
+ 336
23
+ ],
24
+ [
25
+ 672,
26
+ 672
27
+ ],
28
+ [
29
+ 1008,
30
+ 336
31
+ ],
32
+ [
33
+ 336,
34
+ 1008
35
+ ]
36
+ ],
37
+ "image_pooling_kernel": [
38
+ 1,
39
+ 3,
40
+ 3
41
+ ],
42
+ "image_pooling_stride": [
43
+ 1,
44
+ 3,
45
+ 3
46
+ ],
47
+ "image_token_index": 32000,
48
+ "long_clip": true,
49
+ "max_T": 64,
50
+ "model_type": "llava_next",
51
+ "pad_token_id": 0,
52
+ "pllava_pooling_shape": null,
53
+ "pooling": "clipST_3d",
54
+ "pooling_kernel": [
55
+ 1,
56
+ 3,
57
+ 3
58
+ ],
59
+ "pooling_stride": [
60
+ 1,
61
+ 3,
62
+ 3
63
+ ],
64
+ "pooling_temp": 0.01,
65
+ "projector_hidden_act": "gelu",
66
+ "text_config": {
67
+ "_name_or_path": "lmsys/vicuna-7b-v1.5",
68
+ "architectures": [
69
+ "LlamaForCausalLM"
70
+ ],
71
+ "max_position_embeddings": 4096,
72
+ "model_type": "llama",
73
+ "pad_token_id": 0,
74
+ "rms_norm_eps": 1e-05,
75
+ "torch_dtype": "float16",
76
+ "vocab_size": 32064
77
+ },
78
+ "torch_dtype": "bfloat16",
79
+ "transformers_version": "4.40.0",
80
+ "use_image_newline_parameter": true,
81
+ "vision_config": {
82
+ "depth": 4,
83
+ "hidden_size": 1024,
84
+ "image_size": 336,
85
+ "intermediate_size": 4096,
86
+ "max_T": 64,
87
+ "model_type": "clip_vision_model",
88
+ "num_attention_heads": 16,
89
+ "num_hidden_layers": 24,
90
+ "patch_size": 14,
91
+ "projection_dim": 768,
92
+ "vocab_size": 32000
93
+ },
94
+ "vision_feature_layer": -2,
95
+ "vision_feature_select_strategy": "default",
96
+ "vocab_size": 32064
97
+ }
ppllava_vicuna7b_image_video_dpo/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.40.0"
7
+ }
ppllava_vicuna7b_image_video_dpo/model-00001-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:033595b39aba4b33e3ebf887c82eb7d3a5d4f3e7e3dae0d4b29064ffe4b8765a
3
+ size 4980641554
ppllava_vicuna7b_image_video_dpo/model-00002-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0788e4bcf2a4a5d588e40b8cee732df58e090aabf887b9090a53d09627b1fead
3
+ size 4957878536
ppllava_vicuna7b_image_video_dpo/model-00003-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1764bff43c23749dabbafa5f3e135224ac2cc07442e552f5e8b1ef904ca7cdee
3
+ size 4580905400
ppllava_vicuna7b_image_video_dpo/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
{checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/special_tokens_map.json RENAMED
File without changes
{checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/tokenizer.model RENAMED
File without changes
{checkpoint-160 β†’ ppllava_vicuna7b_image_video_dpo}/tokenizer_config.json RENAMED
File without changes
ppllava_vicuna7b_image_video_dpo/trainer_state.json ADDED
@@ -0,0 +1,2181 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.9056603773584906,
5
+ "eval_steps": 500,
6
+ "global_step": 120,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.007547169811320755,
13
+ "grad_norm": 4.023828029632568,
14
+ "learning_rate": 1.8518518518518518e-07,
15
+ "logps/chosen": -29.54261589050293,
16
+ "logps/rejected": -34.2619514465332,
17
+ "loss": 0.6928,
18
+ "losses/dpo": 0.6891345977783203,
19
+ "losses/sft": 1.0977673530578613,
20
+ "losses/total": 0.6891345977783203,
21
+ "ref_logps/chosen": -29.5439395904541,
22
+ "ref_logps/rejected": -34.251976013183594,
23
+ "rewards/accuracies": 0.515625,
24
+ "rewards/chosen": 0.00013234722428023815,
25
+ "rewards/margins": 0.0011300418991595507,
26
+ "rewards/rejected": -0.0009976944420486689,
27
+ "step": 1
28
+ },
29
+ {
30
+ "epoch": 0.01509433962264151,
31
+ "grad_norm": 3.774498701095581,
32
+ "learning_rate": 3.7037037037037036e-07,
33
+ "logps/chosen": -28.00726890563965,
34
+ "logps/rejected": -34.31509017944336,
35
+ "loss": 0.6933,
36
+ "losses/dpo": 0.6932892799377441,
37
+ "losses/sft": 1.1102490425109863,
38
+ "losses/total": 0.6932892799377441,
39
+ "ref_logps/chosen": -27.994796752929688,
40
+ "ref_logps/rejected": -34.30337905883789,
41
+ "rewards/accuracies": 0.4921875,
42
+ "rewards/chosen": -0.0012474276591092348,
43
+ "rewards/margins": -7.60925468057394e-05,
44
+ "rewards/rejected": -0.0011713354615494609,
45
+ "step": 2
46
+ },
47
+ {
48
+ "epoch": 0.022641509433962263,
49
+ "grad_norm": 4.361927032470703,
50
+ "learning_rate": 5.555555555555555e-07,
51
+ "logps/chosen": -32.22264099121094,
52
+ "logps/rejected": -39.77468490600586,
53
+ "loss": 0.6923,
54
+ "losses/dpo": 0.6875869035720825,
55
+ "losses/sft": 1.1771142482757568,
56
+ "losses/total": 0.6875869035720825,
57
+ "ref_logps/chosen": -32.23395919799805,
58
+ "ref_logps/rejected": -39.76538848876953,
59
+ "rewards/accuracies": 0.546875,
60
+ "rewards/chosen": 0.0011316344607621431,
61
+ "rewards/margins": 0.002061467617750168,
62
+ "rewards/rejected": -0.0009298332734033465,
63
+ "step": 3
64
+ },
65
+ {
66
+ "epoch": 0.03018867924528302,
67
+ "grad_norm": 4.030606746673584,
68
+ "learning_rate": 7.407407407407407e-07,
69
+ "logps/chosen": -29.95352554321289,
70
+ "logps/rejected": -35.96381378173828,
71
+ "loss": 0.6931,
72
+ "losses/dpo": 0.6984821557998657,
73
+ "losses/sft": 1.119036078453064,
74
+ "losses/total": 0.6984821557998657,
75
+ "ref_logps/chosen": -29.93365478515625,
76
+ "ref_logps/rejected": -35.9401969909668,
77
+ "rewards/accuracies": 0.53125,
78
+ "rewards/chosen": -0.0019873534329235554,
79
+ "rewards/margins": 0.00037475768476724625,
80
+ "rewards/rejected": -0.002362110884860158,
81
+ "step": 4
82
+ },
83
+ {
84
+ "epoch": 0.03773584905660377,
85
+ "grad_norm": 4.286405563354492,
86
+ "learning_rate": 9.259259259259259e-07,
87
+ "logps/chosen": -33.54985046386719,
88
+ "logps/rejected": -38.25576400756836,
89
+ "loss": 0.6932,
90
+ "losses/dpo": 0.6964197158813477,
91
+ "losses/sft": 0.9851586818695068,
92
+ "losses/total": 0.6964197158813477,
93
+ "ref_logps/chosen": -33.534912109375,
94
+ "ref_logps/rejected": -38.2386474609375,
95
+ "rewards/accuracies": 0.484375,
96
+ "rewards/chosen": -0.0014938730746507645,
97
+ "rewards/margins": 0.00021800631657242775,
98
+ "rewards/rejected": -0.0017118793912231922,
99
+ "step": 5
100
+ },
101
+ {
102
+ "epoch": 0.045283018867924525,
103
+ "grad_norm": 3.715064764022827,
104
+ "learning_rate": 1.111111111111111e-06,
105
+ "logps/chosen": -28.753326416015625,
106
+ "logps/rejected": -33.94843292236328,
107
+ "loss": 0.6956,
108
+ "losses/dpo": 0.696850061416626,
109
+ "losses/sft": 0.9852187633514404,
110
+ "losses/total": 0.696850061416626,
111
+ "ref_logps/chosen": -28.712608337402344,
112
+ "ref_logps/rejected": -33.953330993652344,
113
+ "rewards/accuracies": 0.4375,
114
+ "rewards/chosen": -0.004071743693202734,
115
+ "rewards/margins": -0.0045617930591106415,
116
+ "rewards/rejected": 0.0004900495987385511,
117
+ "step": 6
118
+ },
119
+ {
120
+ "epoch": 0.052830188679245285,
121
+ "grad_norm": 4.02261209487915,
122
+ "learning_rate": 1.2962962962962962e-06,
123
+ "logps/chosen": -29.45867347717285,
124
+ "logps/rejected": -34.641387939453125,
125
+ "loss": 0.6942,
126
+ "losses/dpo": 0.6925374269485474,
127
+ "losses/sft": 1.0857254266738892,
128
+ "losses/total": 0.6925374269485474,
129
+ "ref_logps/chosen": -29.440855026245117,
130
+ "ref_logps/rejected": -34.63925552368164,
131
+ "rewards/accuracies": 0.4765625,
132
+ "rewards/chosen": -0.001781845698133111,
133
+ "rewards/margins": -0.0015682985540479422,
134
+ "rewards/rejected": -0.00021354644559323788,
135
+ "step": 7
136
+ },
137
+ {
138
+ "epoch": 0.06037735849056604,
139
+ "grad_norm": 4.2059502601623535,
140
+ "learning_rate": 1.4814814814814815e-06,
141
+ "logps/chosen": -34.3134651184082,
142
+ "logps/rejected": -35.359073638916016,
143
+ "loss": 0.693,
144
+ "losses/dpo": 0.6923079490661621,
145
+ "losses/sft": 1.0715954303741455,
146
+ "losses/total": 0.6923079490661621,
147
+ "ref_logps/chosen": -34.33416748046875,
148
+ "ref_logps/rejected": -35.37400817871094,
149
+ "rewards/accuracies": 0.5234375,
150
+ "rewards/chosen": 0.0020700739696621895,
151
+ "rewards/margins": 0.000576640130020678,
152
+ "rewards/rejected": 0.001493434072472155,
153
+ "step": 8
154
+ },
155
+ {
156
+ "epoch": 0.06792452830188679,
157
+ "grad_norm": 5.0255889892578125,
158
+ "learning_rate": 1.6666666666666667e-06,
159
+ "logps/chosen": -30.589462280273438,
160
+ "logps/rejected": -39.17620849609375,
161
+ "loss": 0.6948,
162
+ "losses/dpo": 0.6894321441650391,
163
+ "losses/sft": 0.8859687447547913,
164
+ "losses/total": 0.6894321441650391,
165
+ "ref_logps/chosen": -30.609333038330078,
166
+ "ref_logps/rejected": -39.22538757324219,
167
+ "rewards/accuracies": 0.5390625,
168
+ "rewards/chosen": 0.0019869431853294373,
169
+ "rewards/margins": -0.002930472604930401,
170
+ "rewards/rejected": 0.004917416721582413,
171
+ "step": 9
172
+ },
173
+ {
174
+ "epoch": 0.07547169811320754,
175
+ "grad_norm": 4.278906345367432,
176
+ "learning_rate": 1.8518518518518519e-06,
177
+ "logps/chosen": -26.85662841796875,
178
+ "logps/rejected": -34.258155822753906,
179
+ "loss": 0.6886,
180
+ "losses/dpo": 0.6904473304748535,
181
+ "losses/sft": 1.016200304031372,
182
+ "losses/total": 0.6904473304748535,
183
+ "ref_logps/chosen": -27.009971618652344,
184
+ "ref_logps/rejected": -34.316612243652344,
185
+ "rewards/accuracies": 0.578125,
186
+ "rewards/chosen": 0.015334523282945156,
187
+ "rewards/margins": 0.009488685056567192,
188
+ "rewards/rejected": 0.005845838226377964,
189
+ "step": 10
190
+ },
191
+ {
192
+ "epoch": 0.0830188679245283,
193
+ "grad_norm": 3.700289487838745,
194
+ "learning_rate": 2.037037037037037e-06,
195
+ "logps/chosen": -29.495962142944336,
196
+ "logps/rejected": -33.246185302734375,
197
+ "loss": 0.6915,
198
+ "losses/dpo": 0.6893143653869629,
199
+ "losses/sft": 1.2568368911743164,
200
+ "losses/total": 0.6893143653869629,
201
+ "ref_logps/chosen": -29.658611297607422,
202
+ "ref_logps/rejected": -33.370635986328125,
203
+ "rewards/accuracies": 0.5390625,
204
+ "rewards/chosen": 0.01626480370759964,
205
+ "rewards/margins": 0.003819521516561508,
206
+ "rewards/rejected": 0.012445282191038132,
207
+ "step": 11
208
+ },
209
+ {
210
+ "epoch": 0.09056603773584905,
211
+ "grad_norm": 3.853537082672119,
212
+ "learning_rate": 2.222222222222222e-06,
213
+ "logps/chosen": -30.57427215576172,
214
+ "logps/rejected": -34.356842041015625,
215
+ "loss": 0.6943,
216
+ "losses/dpo": 0.702874481678009,
217
+ "losses/sft": 1.008819580078125,
218
+ "losses/total": 0.702874481678009,
219
+ "ref_logps/chosen": -30.768930435180664,
220
+ "ref_logps/rejected": -34.566131591796875,
221
+ "rewards/accuracies": 0.5234375,
222
+ "rewards/chosen": 0.01946595311164856,
223
+ "rewards/margins": -0.0014625652693212032,
224
+ "rewards/rejected": 0.0209285207092762,
225
+ "step": 12
226
+ },
227
+ {
228
+ "epoch": 0.09811320754716982,
229
+ "grad_norm": 4.208248138427734,
230
+ "learning_rate": 2.4074074074074075e-06,
231
+ "logps/chosen": -30.429744720458984,
232
+ "logps/rejected": -39.09281539916992,
233
+ "loss": 0.6895,
234
+ "losses/dpo": 0.7104582786560059,
235
+ "losses/sft": 1.0792236328125,
236
+ "losses/total": 0.7104582786560059,
237
+ "ref_logps/chosen": -30.72281265258789,
238
+ "ref_logps/rejected": -39.29808044433594,
239
+ "rewards/accuracies": 0.578125,
240
+ "rewards/chosen": 0.029306560754776,
241
+ "rewards/margins": 0.00877977255731821,
242
+ "rewards/rejected": 0.020526789128780365,
243
+ "step": 13
244
+ },
245
+ {
246
+ "epoch": 0.10566037735849057,
247
+ "grad_norm": 3.7299113273620605,
248
+ "learning_rate": 2.5925925925925925e-06,
249
+ "logps/chosen": -27.8311710357666,
250
+ "logps/rejected": -33.181427001953125,
251
+ "loss": 0.6862,
252
+ "losses/dpo": 0.6882991790771484,
253
+ "losses/sft": 1.0185338258743286,
254
+ "losses/total": 0.6882991790771484,
255
+ "ref_logps/chosen": -28.355030059814453,
256
+ "ref_logps/rejected": -33.544822692871094,
257
+ "rewards/accuracies": 0.5625,
258
+ "rewards/chosen": 0.052385710179805756,
259
+ "rewards/margins": 0.016046730801463127,
260
+ "rewards/rejected": 0.03633897751569748,
261
+ "step": 14
262
+ },
263
+ {
264
+ "epoch": 0.11320754716981132,
265
+ "grad_norm": 3.8657593727111816,
266
+ "learning_rate": 2.7777777777777783e-06,
267
+ "logps/chosen": -28.884387969970703,
268
+ "logps/rejected": -34.36138916015625,
269
+ "loss": 0.6913,
270
+ "losses/dpo": 0.6939971446990967,
271
+ "losses/sft": 1.0502238273620605,
272
+ "losses/total": 0.6939971446990967,
273
+ "ref_logps/chosen": -29.45929718017578,
274
+ "ref_logps/rejected": -34.86326599121094,
275
+ "rewards/accuracies": 0.5703125,
276
+ "rewards/chosen": 0.05749073624610901,
277
+ "rewards/margins": 0.0073026856407523155,
278
+ "rewards/rejected": 0.05018804967403412,
279
+ "step": 15
280
+ },
281
+ {
282
+ "epoch": 0.12075471698113208,
283
+ "grad_norm": 3.9700570106506348,
284
+ "learning_rate": 2.962962962962963e-06,
285
+ "logps/chosen": -30.124591827392578,
286
+ "logps/rejected": -31.684345245361328,
287
+ "loss": 0.6928,
288
+ "losses/dpo": 0.6692728996276855,
289
+ "losses/sft": 1.005508542060852,
290
+ "losses/total": 0.6692728996276855,
291
+ "ref_logps/chosen": -30.774919509887695,
292
+ "ref_logps/rejected": -32.27117919921875,
293
+ "rewards/accuracies": 0.515625,
294
+ "rewards/chosen": 0.06503286957740784,
295
+ "rewards/margins": 0.00634903647005558,
296
+ "rewards/rejected": 0.058683834969997406,
297
+ "step": 16
298
+ },
299
+ {
300
+ "epoch": 0.12830188679245283,
301
+ "grad_norm": 3.9416866302490234,
302
+ "learning_rate": 3.1481481481481483e-06,
303
+ "logps/chosen": -30.775650024414062,
304
+ "logps/rejected": -34.58787155151367,
305
+ "loss": 0.6834,
306
+ "losses/dpo": 0.7104189395904541,
307
+ "losses/sft": 0.9935987591743469,
308
+ "losses/total": 0.7104189395904541,
309
+ "ref_logps/chosen": -31.45960235595703,
310
+ "ref_logps/rejected": -35.01028060913086,
311
+ "rewards/accuracies": 0.5625,
312
+ "rewards/chosen": 0.0683954730629921,
313
+ "rewards/margins": 0.0261550135910511,
314
+ "rewards/rejected": 0.042240455746650696,
315
+ "step": 17
316
+ },
317
+ {
318
+ "epoch": 0.13584905660377358,
319
+ "grad_norm": 3.8401975631713867,
320
+ "learning_rate": 3.3333333333333333e-06,
321
+ "logps/chosen": -29.183441162109375,
322
+ "logps/rejected": -33.91481018066406,
323
+ "loss": 0.6689,
324
+ "losses/dpo": 0.6731781363487244,
325
+ "losses/sft": 1.0918514728546143,
326
+ "losses/total": 0.6731781363487244,
327
+ "ref_logps/chosen": -30.14259147644043,
328
+ "ref_logps/rejected": -34.299861907958984,
329
+ "rewards/accuracies": 0.6328125,
330
+ "rewards/chosen": 0.09591513872146606,
331
+ "rewards/margins": 0.057410385459661484,
332
+ "rewards/rejected": 0.03850475698709488,
333
+ "step": 18
334
+ },
335
+ {
336
+ "epoch": 0.14339622641509434,
337
+ "grad_norm": 3.793499231338501,
338
+ "learning_rate": 3.5185185185185187e-06,
339
+ "logps/chosen": -25.331146240234375,
340
+ "logps/rejected": -34.978057861328125,
341
+ "loss": 0.6692,
342
+ "losses/dpo": 0.7569133639335632,
343
+ "losses/sft": 0.9651777744293213,
344
+ "losses/total": 0.7569133639335632,
345
+ "ref_logps/chosen": -26.447914123535156,
346
+ "ref_logps/rejected": -35.43275833129883,
347
+ "rewards/accuracies": 0.65625,
348
+ "rewards/chosen": 0.1116766482591629,
349
+ "rewards/margins": 0.06620671600103378,
350
+ "rewards/rejected": 0.04546992480754852,
351
+ "step": 19
352
+ },
353
+ {
354
+ "epoch": 0.1509433962264151,
355
+ "grad_norm": 3.7824015617370605,
356
+ "learning_rate": 3.7037037037037037e-06,
357
+ "logps/chosen": -29.415006637573242,
358
+ "logps/rejected": -34.36804962158203,
359
+ "loss": 0.6666,
360
+ "losses/dpo": 0.6621171832084656,
361
+ "losses/sft": 0.7287149429321289,
362
+ "losses/total": 0.6621171832084656,
363
+ "ref_logps/chosen": -30.477230072021484,
364
+ "ref_logps/rejected": -34.688987731933594,
365
+ "rewards/accuracies": 0.6015625,
366
+ "rewards/chosen": 0.10622246563434601,
367
+ "rewards/margins": 0.0741288959980011,
368
+ "rewards/rejected": 0.03209357336163521,
369
+ "step": 20
370
+ },
371
+ {
372
+ "epoch": 0.15849056603773584,
373
+ "grad_norm": 4.047840595245361,
374
+ "learning_rate": 3.88888888888889e-06,
375
+ "logps/chosen": -29.250553131103516,
376
+ "logps/rejected": -39.266815185546875,
377
+ "loss": 0.6475,
378
+ "losses/dpo": 0.647861123085022,
379
+ "losses/sft": 0.9502246379852295,
380
+ "losses/total": 0.647861123085022,
381
+ "ref_logps/chosen": -30.495431900024414,
382
+ "ref_logps/rejected": -39.307708740234375,
383
+ "rewards/accuracies": 0.609375,
384
+ "rewards/chosen": 0.12448810040950775,
385
+ "rewards/margins": 0.12039919942617416,
386
+ "rewards/rejected": 0.0040889037773013115,
387
+ "step": 21
388
+ },
389
+ {
390
+ "epoch": 0.1660377358490566,
391
+ "grad_norm": 4.19422721862793,
392
+ "learning_rate": 4.074074074074074e-06,
393
+ "logps/chosen": -27.82355308532715,
394
+ "logps/rejected": -35.92847442626953,
395
+ "loss": 0.6735,
396
+ "losses/dpo": 0.5619722604751587,
397
+ "losses/sft": 0.8959212899208069,
398
+ "losses/total": 0.5619722604751587,
399
+ "ref_logps/chosen": -29.235668182373047,
400
+ "ref_logps/rejected": -36.55792999267578,
401
+ "rewards/accuracies": 0.5859375,
402
+ "rewards/chosen": 0.14121143519878387,
403
+ "rewards/margins": 0.07826600223779678,
404
+ "rewards/rejected": 0.06294544041156769,
405
+ "step": 22
406
+ },
407
+ {
408
+ "epoch": 0.17358490566037735,
409
+ "grad_norm": 3.807786464691162,
410
+ "learning_rate": 4.2592592592592596e-06,
411
+ "logps/chosen": -27.979816436767578,
412
+ "logps/rejected": -38.823341369628906,
413
+ "loss": 0.6627,
414
+ "losses/dpo": 0.6460346579551697,
415
+ "losses/sft": 1.1445313692092896,
416
+ "losses/total": 0.6460346579551697,
417
+ "ref_logps/chosen": -28.973108291625977,
418
+ "ref_logps/rejected": -38.76581573486328,
419
+ "rewards/accuracies": 0.5859375,
420
+ "rewards/chosen": 0.09932918101549149,
421
+ "rewards/margins": 0.10508126765489578,
422
+ "rewards/rejected": -0.005752084776759148,
423
+ "step": 23
424
+ },
425
+ {
426
+ "epoch": 0.1811320754716981,
427
+ "grad_norm": 4.215601921081543,
428
+ "learning_rate": 4.444444444444444e-06,
429
+ "logps/chosen": -30.01553726196289,
430
+ "logps/rejected": -36.99456787109375,
431
+ "loss": 0.6726,
432
+ "losses/dpo": 0.7518600225448608,
433
+ "losses/sft": 1.1231310367584229,
434
+ "losses/total": 0.7518600225448608,
435
+ "ref_logps/chosen": -30.876399993896484,
436
+ "ref_logps/rejected": -37.021759033203125,
437
+ "rewards/accuracies": 0.625,
438
+ "rewards/chosen": 0.08608602732419968,
439
+ "rewards/margins": 0.08336685597896576,
440
+ "rewards/rejected": 0.0027191713452339172,
441
+ "step": 24
442
+ },
443
+ {
444
+ "epoch": 0.18867924528301888,
445
+ "grad_norm": 3.991433620452881,
446
+ "learning_rate": 4.62962962962963e-06,
447
+ "logps/chosen": -29.20758056640625,
448
+ "logps/rejected": -33.56412124633789,
449
+ "loss": 0.6631,
450
+ "losses/dpo": 0.5801557302474976,
451
+ "losses/sft": 1.1197171211242676,
452
+ "losses/total": 0.5801557302474976,
453
+ "ref_logps/chosen": -30.49410629272461,
454
+ "ref_logps/rejected": -33.72259521484375,
455
+ "rewards/accuracies": 0.6171875,
456
+ "rewards/chosen": 0.1286524534225464,
457
+ "rewards/margins": 0.11280516535043716,
458
+ "rewards/rejected": 0.01584728993475437,
459
+ "step": 25
460
+ },
461
+ {
462
+ "epoch": 0.19622641509433963,
463
+ "grad_norm": 4.136443614959717,
464
+ "learning_rate": 4.814814814814815e-06,
465
+ "logps/chosen": -27.908275604248047,
466
+ "logps/rejected": -36.784366607666016,
467
+ "loss": 0.6475,
468
+ "losses/dpo": 0.7332102060317993,
469
+ "losses/sft": 0.7210355401039124,
470
+ "losses/total": 0.7332102060317993,
471
+ "ref_logps/chosen": -29.110748291015625,
472
+ "ref_logps/rejected": -36.522212982177734,
473
+ "rewards/accuracies": 0.6796875,
474
+ "rewards/chosen": 0.12024737149477005,
475
+ "rewards/margins": 0.14646272361278534,
476
+ "rewards/rejected": -0.026215344667434692,
477
+ "step": 26
478
+ },
479
+ {
480
+ "epoch": 0.2037735849056604,
481
+ "grad_norm": 3.6287527084350586,
482
+ "learning_rate": 5e-06,
483
+ "logps/chosen": -25.643430709838867,
484
+ "logps/rejected": -33.46253204345703,
485
+ "loss": 0.603,
486
+ "losses/dpo": 0.6435042023658752,
487
+ "losses/sft": 0.7656551599502563,
488
+ "losses/total": 0.6435042023658752,
489
+ "ref_logps/chosen": -27.53249740600586,
490
+ "ref_logps/rejected": -32.8140869140625,
491
+ "rewards/accuracies": 0.671875,
492
+ "rewards/chosen": 0.18890666961669922,
493
+ "rewards/margins": 0.25375133752822876,
494
+ "rewards/rejected": -0.06484466791152954,
495
+ "step": 27
496
+ },
497
+ {
498
+ "epoch": 0.21132075471698114,
499
+ "grad_norm": 5.5130720138549805,
500
+ "learning_rate": 4.978902953586498e-06,
501
+ "logps/chosen": -32.41269302368164,
502
+ "logps/rejected": -36.27760314941406,
503
+ "loss": 0.6718,
504
+ "losses/dpo": 0.7667650580406189,
505
+ "losses/sft": 1.180605411529541,
506
+ "losses/total": 0.7667650580406189,
507
+ "ref_logps/chosen": -32.947940826416016,
508
+ "ref_logps/rejected": -35.710548400878906,
509
+ "rewards/accuracies": 0.625,
510
+ "rewards/chosen": 0.05352487415075302,
511
+ "rewards/margins": 0.11023038625717163,
512
+ "rewards/rejected": -0.05670551210641861,
513
+ "step": 28
514
+ },
515
+ {
516
+ "epoch": 0.2188679245283019,
517
+ "grad_norm": 4.2082695960998535,
518
+ "learning_rate": 4.957805907172996e-06,
519
+ "logps/chosen": -30.38204574584961,
520
+ "logps/rejected": -38.79253387451172,
521
+ "loss": 0.6259,
522
+ "losses/dpo": 0.6825762987136841,
523
+ "losses/sft": 1.3010542392730713,
524
+ "losses/total": 0.6825762987136841,
525
+ "ref_logps/chosen": -31.19179916381836,
526
+ "ref_logps/rejected": -37.440677642822266,
527
+ "rewards/accuracies": 0.6640625,
528
+ "rewards/chosen": 0.08097504824399948,
529
+ "rewards/margins": 0.21616099774837494,
530
+ "rewards/rejected": -0.13518595695495605,
531
+ "step": 29
532
+ },
533
+ {
534
+ "epoch": 0.22641509433962265,
535
+ "grad_norm": 4.6551737785339355,
536
+ "learning_rate": 4.936708860759495e-06,
537
+ "logps/chosen": -30.135501861572266,
538
+ "logps/rejected": -37.48721694946289,
539
+ "loss": 0.6665,
540
+ "losses/dpo": 0.5913557410240173,
541
+ "losses/sft": 1.0625046491622925,
542
+ "losses/total": 0.5913557410240173,
543
+ "ref_logps/chosen": -30.550817489624023,
544
+ "ref_logps/rejected": -36.66627502441406,
545
+ "rewards/accuracies": 0.5859375,
546
+ "rewards/chosen": 0.04153158515691757,
547
+ "rewards/margins": 0.12362557649612427,
548
+ "rewards/rejected": -0.08209399878978729,
549
+ "step": 30
550
+ },
551
+ {
552
+ "epoch": 0.2339622641509434,
553
+ "grad_norm": 4.481552600860596,
554
+ "learning_rate": 4.915611814345992e-06,
555
+ "logps/chosen": -29.738265991210938,
556
+ "logps/rejected": -40.146636962890625,
557
+ "loss": 0.6341,
558
+ "losses/dpo": 0.6154491305351257,
559
+ "losses/sft": 0.7811817526817322,
560
+ "losses/total": 0.6154491305351257,
561
+ "ref_logps/chosen": -29.985429763793945,
562
+ "ref_logps/rejected": -38.49150085449219,
563
+ "rewards/accuracies": 0.625,
564
+ "rewards/chosen": 0.02471642568707466,
565
+ "rewards/margins": 0.19022998213768005,
566
+ "rewards/rejected": -0.1655135601758957,
567
+ "step": 31
568
+ },
569
+ {
570
+ "epoch": 0.24150943396226415,
571
+ "grad_norm": 4.287616729736328,
572
+ "learning_rate": 4.89451476793249e-06,
573
+ "logps/chosen": -28.966516494750977,
574
+ "logps/rejected": -38.64032745361328,
575
+ "loss": 0.6213,
576
+ "losses/dpo": 0.5834950804710388,
577
+ "losses/sft": 1.19804847240448,
578
+ "losses/total": 0.5834950804710388,
579
+ "ref_logps/chosen": -29.167491912841797,
580
+ "ref_logps/rejected": -36.74009704589844,
581
+ "rewards/accuracies": 0.671875,
582
+ "rewards/chosen": 0.0200975239276886,
583
+ "rewards/margins": 0.21012060344219208,
584
+ "rewards/rejected": -0.19002306461334229,
585
+ "step": 32
586
+ },
587
+ {
588
+ "epoch": 0.2490566037735849,
589
+ "grad_norm": 4.030296325683594,
590
+ "learning_rate": 4.873417721518987e-06,
591
+ "logps/chosen": -26.15782928466797,
592
+ "logps/rejected": -33.3205451965332,
593
+ "loss": 0.6113,
594
+ "losses/dpo": 0.5751947164535522,
595
+ "losses/sft": 0.7719302177429199,
596
+ "losses/total": 0.5751947164535522,
597
+ "ref_logps/chosen": -26.509279251098633,
598
+ "ref_logps/rejected": -31.373973846435547,
599
+ "rewards/accuracies": 0.71875,
600
+ "rewards/chosen": 0.03514501452445984,
601
+ "rewards/margins": 0.22980214655399323,
602
+ "rewards/rejected": -0.1946571320295334,
603
+ "step": 33
604
+ },
605
+ {
606
+ "epoch": 0.25660377358490566,
607
+ "grad_norm": 4.367461204528809,
608
+ "learning_rate": 4.852320675105486e-06,
609
+ "logps/chosen": -32.38875961303711,
610
+ "logps/rejected": -40.95237350463867,
611
+ "loss": 0.6236,
612
+ "losses/dpo": 0.6270190477371216,
613
+ "losses/sft": 1.1532888412475586,
614
+ "losses/total": 0.6270190477371216,
615
+ "ref_logps/chosen": -31.350940704345703,
616
+ "ref_logps/rejected": -37.58573913574219,
617
+ "rewards/accuracies": 0.71875,
618
+ "rewards/chosen": -0.10378223657608032,
619
+ "rewards/margins": 0.2328808605670929,
620
+ "rewards/rejected": -0.3366630971431732,
621
+ "step": 34
622
+ },
623
+ {
624
+ "epoch": 0.2641509433962264,
625
+ "grad_norm": 4.614798545837402,
626
+ "learning_rate": 4.831223628691984e-06,
627
+ "logps/chosen": -32.56526184082031,
628
+ "logps/rejected": -40.50109100341797,
629
+ "loss": 0.6059,
630
+ "losses/dpo": 0.7241290211677551,
631
+ "losses/sft": 1.3340615034103394,
632
+ "losses/total": 0.7241290211677551,
633
+ "ref_logps/chosen": -30.577272415161133,
634
+ "ref_logps/rejected": -35.84385681152344,
635
+ "rewards/accuracies": 0.6796875,
636
+ "rewards/chosen": -0.19879914820194244,
637
+ "rewards/margins": 0.26692429184913635,
638
+ "rewards/rejected": -0.46572345495224,
639
+ "step": 35
640
+ },
641
+ {
642
+ "epoch": 0.27169811320754716,
643
+ "grad_norm": 4.4597649574279785,
644
+ "learning_rate": 4.8101265822784815e-06,
645
+ "logps/chosen": -34.825111389160156,
646
+ "logps/rejected": -44.15177917480469,
647
+ "loss": 0.5946,
648
+ "losses/dpo": 0.5129742622375488,
649
+ "losses/sft": 0.856516420841217,
650
+ "losses/total": 0.5129742622375488,
651
+ "ref_logps/chosen": -32.00858688354492,
652
+ "ref_logps/rejected": -38.221412658691406,
653
+ "rewards/accuracies": 0.6953125,
654
+ "rewards/chosen": -0.2816521227359772,
655
+ "rewards/margins": 0.31138482689857483,
656
+ "rewards/rejected": -0.593036949634552,
657
+ "step": 36
658
+ },
659
+ {
660
+ "epoch": 0.2792452830188679,
661
+ "grad_norm": 4.562221527099609,
662
+ "learning_rate": 4.789029535864979e-06,
663
+ "logps/chosen": -33.27934646606445,
664
+ "logps/rejected": -39.6660041809082,
665
+ "loss": 0.598,
666
+ "losses/dpo": 0.6043655276298523,
667
+ "losses/sft": 1.0378785133361816,
668
+ "losses/total": 0.6043655276298523,
669
+ "ref_logps/chosen": -29.917388916015625,
670
+ "ref_logps/rejected": -33.3465690612793,
671
+ "rewards/accuracies": 0.6484375,
672
+ "rewards/chosen": -0.33619576692581177,
673
+ "rewards/margins": 0.29574763774871826,
674
+ "rewards/rejected": -0.6319433450698853,
675
+ "step": 37
676
+ },
677
+ {
678
+ "epoch": 0.28679245283018867,
679
+ "grad_norm": 5.42985725402832,
680
+ "learning_rate": 4.767932489451477e-06,
681
+ "logps/chosen": -36.16193389892578,
682
+ "logps/rejected": -43.71050262451172,
683
+ "loss": 0.6555,
684
+ "losses/dpo": 0.6992093324661255,
685
+ "losses/sft": 1.3419675827026367,
686
+ "losses/total": 0.6992093324661255,
687
+ "ref_logps/chosen": -31.39295196533203,
688
+ "ref_logps/rejected": -36.14948654174805,
689
+ "rewards/accuracies": 0.65625,
690
+ "rewards/chosen": -0.47689831256866455,
691
+ "rewards/margins": 0.2792032063007355,
692
+ "rewards/rejected": -0.7561015486717224,
693
+ "step": 38
694
+ },
695
+ {
696
+ "epoch": 0.2943396226415094,
697
+ "grad_norm": 4.9820051193237305,
698
+ "learning_rate": 4.746835443037975e-06,
699
+ "logps/chosen": -33.79109191894531,
700
+ "logps/rejected": -42.09747314453125,
701
+ "loss": 0.5883,
702
+ "losses/dpo": 0.6338505148887634,
703
+ "losses/sft": 1.2124994993209839,
704
+ "losses/total": 0.6338505148887634,
705
+ "ref_logps/chosen": -28.94110107421875,
706
+ "ref_logps/rejected": -33.52173614501953,
707
+ "rewards/accuracies": 0.7109375,
708
+ "rewards/chosen": -0.4849993586540222,
709
+ "rewards/margins": 0.3725742697715759,
710
+ "rewards/rejected": -0.8575735688209534,
711
+ "step": 39
712
+ },
713
+ {
714
+ "epoch": 0.3018867924528302,
715
+ "grad_norm": 5.098467826843262,
716
+ "learning_rate": 4.725738396624473e-06,
717
+ "logps/chosen": -34.21745681762695,
718
+ "logps/rejected": -43.00538635253906,
719
+ "loss": 0.6283,
720
+ "losses/dpo": 0.7655435800552368,
721
+ "losses/sft": 1.7303447723388672,
722
+ "losses/total": 0.7655435800552368,
723
+ "ref_logps/chosen": -27.329925537109375,
724
+ "ref_logps/rejected": -33.21820068359375,
725
+ "rewards/accuracies": 0.671875,
726
+ "rewards/chosen": -0.6887531876564026,
727
+ "rewards/margins": 0.2899653911590576,
728
+ "rewards/rejected": -0.9787185788154602,
729
+ "step": 40
730
+ },
731
+ {
732
+ "epoch": 0.30943396226415093,
733
+ "grad_norm": 4.568510055541992,
734
+ "learning_rate": 4.7046413502109714e-06,
735
+ "logps/chosen": -34.30303955078125,
736
+ "logps/rejected": -45.21646499633789,
737
+ "loss": 0.5376,
738
+ "losses/dpo": 0.466990202665329,
739
+ "losses/sft": 1.3411931991577148,
740
+ "losses/total": 0.466990202665329,
741
+ "ref_logps/chosen": -27.724037170410156,
742
+ "ref_logps/rejected": -33.146297454833984,
743
+ "rewards/accuracies": 0.7265625,
744
+ "rewards/chosen": -0.657900333404541,
745
+ "rewards/margins": 0.5491164326667786,
746
+ "rewards/rejected": -1.2070167064666748,
747
+ "step": 41
748
+ },
749
+ {
750
+ "epoch": 0.3169811320754717,
751
+ "grad_norm": 4.940774440765381,
752
+ "learning_rate": 4.683544303797468e-06,
753
+ "logps/chosen": -37.19551086425781,
754
+ "logps/rejected": -48.22083282470703,
755
+ "loss": 0.5488,
756
+ "losses/dpo": 0.7535874843597412,
757
+ "losses/sft": 1.5672276020050049,
758
+ "losses/total": 0.7535874843597412,
759
+ "ref_logps/chosen": -30.618778228759766,
760
+ "ref_logps/rejected": -36.40138244628906,
761
+ "rewards/accuracies": 0.7421875,
762
+ "rewards/chosen": -0.6576732993125916,
763
+ "rewards/margins": 0.5242711901664734,
764
+ "rewards/rejected": -1.181944489479065,
765
+ "step": 42
766
+ },
767
+ {
768
+ "epoch": 0.32452830188679244,
769
+ "grad_norm": 5.207009792327881,
770
+ "learning_rate": 4.662447257383967e-06,
771
+ "logps/chosen": -38.67348861694336,
772
+ "logps/rejected": -46.259944915771484,
773
+ "loss": 0.6086,
774
+ "losses/dpo": 0.580439567565918,
775
+ "losses/sft": 1.5248361825942993,
776
+ "losses/total": 0.580439567565918,
777
+ "ref_logps/chosen": -30.998300552368164,
778
+ "ref_logps/rejected": -34.962703704833984,
779
+ "rewards/accuracies": 0.671875,
780
+ "rewards/chosen": -0.767518937587738,
781
+ "rewards/margins": 0.3622052073478699,
782
+ "rewards/rejected": -1.129724144935608,
783
+ "step": 43
784
+ },
785
+ {
786
+ "epoch": 0.3320754716981132,
787
+ "grad_norm": 5.374455451965332,
788
+ "learning_rate": 4.641350210970465e-06,
789
+ "logps/chosen": -35.60981750488281,
790
+ "logps/rejected": -45.90471649169922,
791
+ "loss": 0.6258,
792
+ "losses/dpo": 0.7469555139541626,
793
+ "losses/sft": 1.5075256824493408,
794
+ "losses/total": 0.7469555139541626,
795
+ "ref_logps/chosen": -27.66849136352539,
796
+ "ref_logps/rejected": -34.36457824707031,
797
+ "rewards/accuracies": 0.6484375,
798
+ "rewards/chosen": -0.79413241147995,
799
+ "rewards/margins": 0.35988110303878784,
800
+ "rewards/rejected": -1.1540133953094482,
801
+ "step": 44
802
+ },
803
+ {
804
+ "epoch": 0.33962264150943394,
805
+ "grad_norm": 5.304266452789307,
806
+ "learning_rate": 4.620253164556963e-06,
807
+ "logps/chosen": -38.42974090576172,
808
+ "logps/rejected": -45.68645477294922,
809
+ "loss": 0.6133,
810
+ "losses/dpo": 0.6757426261901855,
811
+ "losses/sft": 1.243510127067566,
812
+ "losses/total": 0.6757426261901855,
813
+ "ref_logps/chosen": -29.328536987304688,
814
+ "ref_logps/rejected": -32.262840270996094,
815
+ "rewards/accuracies": 0.6796875,
816
+ "rewards/chosen": -0.9101204872131348,
817
+ "rewards/margins": 0.4322410821914673,
818
+ "rewards/rejected": -1.342361569404602,
819
+ "step": 45
820
+ },
821
+ {
822
+ "epoch": 0.3471698113207547,
823
+ "grad_norm": 5.428191184997559,
824
+ "learning_rate": 4.5991561181434605e-06,
825
+ "logps/chosen": -38.54790115356445,
826
+ "logps/rejected": -49.04346466064453,
827
+ "loss": 0.5944,
828
+ "losses/dpo": 0.7710250020027161,
829
+ "losses/sft": 1.5200599431991577,
830
+ "losses/total": 0.7710250020027161,
831
+ "ref_logps/chosen": -29.571321487426758,
832
+ "ref_logps/rejected": -35.5488166809082,
833
+ "rewards/accuracies": 0.671875,
834
+ "rewards/chosen": -0.897658109664917,
835
+ "rewards/margins": 0.45180660486221313,
836
+ "rewards/rejected": -1.3494646549224854,
837
+ "step": 46
838
+ },
839
+ {
840
+ "epoch": 0.35471698113207545,
841
+ "grad_norm": 5.276749610900879,
842
+ "learning_rate": 4.578059071729958e-06,
843
+ "logps/chosen": -42.278629302978516,
844
+ "logps/rejected": -49.082420349121094,
845
+ "loss": 0.5844,
846
+ "losses/dpo": 0.5217583179473877,
847
+ "losses/sft": 1.251055121421814,
848
+ "losses/total": 0.5217583179473877,
849
+ "ref_logps/chosen": -32.70330047607422,
850
+ "ref_logps/rejected": -34.48522186279297,
851
+ "rewards/accuracies": 0.7734375,
852
+ "rewards/chosen": -0.9575330018997192,
853
+ "rewards/margins": 0.5021871328353882,
854
+ "rewards/rejected": -1.4597201347351074,
855
+ "step": 47
856
+ },
857
+ {
858
+ "epoch": 0.3622641509433962,
859
+ "grad_norm": 4.85624361038208,
860
+ "learning_rate": 4.556962025316456e-06,
861
+ "logps/chosen": -37.618473052978516,
862
+ "logps/rejected": -49.08882141113281,
863
+ "loss": 0.5444,
864
+ "losses/dpo": 0.4470449984073639,
865
+ "losses/sft": 1.1268991231918335,
866
+ "losses/total": 0.4470449984073639,
867
+ "ref_logps/chosen": -29.47239112854004,
868
+ "ref_logps/rejected": -34.867340087890625,
869
+ "rewards/accuracies": 0.734375,
870
+ "rewards/chosen": -0.8146083354949951,
871
+ "rewards/margins": 0.6075396537780762,
872
+ "rewards/rejected": -1.4221479892730713,
873
+ "step": 48
874
+ },
875
+ {
876
+ "epoch": 0.36981132075471695,
877
+ "grad_norm": 5.045034408569336,
878
+ "learning_rate": 4.535864978902954e-06,
879
+ "logps/chosen": -33.628997802734375,
880
+ "logps/rejected": -46.788063049316406,
881
+ "loss": 0.5441,
882
+ "losses/dpo": 0.4152263402938843,
883
+ "losses/sft": 1.2738722562789917,
884
+ "losses/total": 0.4152263402938843,
885
+ "ref_logps/chosen": -25.57724952697754,
886
+ "ref_logps/rejected": -33.02429962158203,
887
+ "rewards/accuracies": 0.7265625,
888
+ "rewards/chosen": -0.8051748275756836,
889
+ "rewards/margins": 0.5712013840675354,
890
+ "rewards/rejected": -1.3763761520385742,
891
+ "step": 49
892
+ },
893
+ {
894
+ "epoch": 0.37735849056603776,
895
+ "grad_norm": 4.926025867462158,
896
+ "learning_rate": 4.514767932489452e-06,
897
+ "logps/chosen": -40.1754264831543,
898
+ "logps/rejected": -53.34943389892578,
899
+ "loss": 0.5357,
900
+ "losses/dpo": 0.42820218205451965,
901
+ "losses/sft": 1.4205973148345947,
902
+ "losses/total": 0.42820218205451965,
903
+ "ref_logps/chosen": -30.12688636779785,
904
+ "ref_logps/rejected": -36.76472091674805,
905
+ "rewards/accuracies": 0.7421875,
906
+ "rewards/chosen": -1.0048537254333496,
907
+ "rewards/margins": 0.6536170840263367,
908
+ "rewards/rejected": -1.658470869064331,
909
+ "step": 50
910
+ },
911
+ {
912
+ "epoch": 0.3849056603773585,
913
+ "grad_norm": 5.366219997406006,
914
+ "learning_rate": 4.4936708860759495e-06,
915
+ "logps/chosen": -41.36455535888672,
916
+ "logps/rejected": -51.32182312011719,
917
+ "loss": 0.5498,
918
+ "losses/dpo": 0.5175436735153198,
919
+ "losses/sft": 1.3962128162384033,
920
+ "losses/total": 0.5175436735153198,
921
+ "ref_logps/chosen": -31.545040130615234,
922
+ "ref_logps/rejected": -34.96824264526367,
923
+ "rewards/accuracies": 0.6875,
924
+ "rewards/chosen": -0.9819516539573669,
925
+ "rewards/margins": 0.6534063220024109,
926
+ "rewards/rejected": -1.6353578567504883,
927
+ "step": 51
928
+ },
929
+ {
930
+ "epoch": 0.39245283018867927,
931
+ "grad_norm": 5.386270523071289,
932
+ "learning_rate": 4.472573839662447e-06,
933
+ "logps/chosen": -37.03908920288086,
934
+ "logps/rejected": -50.11444854736328,
935
+ "loss": 0.5742,
936
+ "losses/dpo": 0.8617876768112183,
937
+ "losses/sft": 2.0467562675476074,
938
+ "losses/total": 0.8617876768112183,
939
+ "ref_logps/chosen": -27.370098114013672,
940
+ "ref_logps/rejected": -34.820045471191406,
941
+ "rewards/accuracies": 0.7109375,
942
+ "rewards/chosen": -0.9668989181518555,
943
+ "rewards/margins": 0.5625417232513428,
944
+ "rewards/rejected": -1.5294406414031982,
945
+ "step": 52
946
+ },
947
+ {
948
+ "epoch": 0.4,
949
+ "grad_norm": 4.991089820861816,
950
+ "learning_rate": 4.451476793248945e-06,
951
+ "logps/chosen": -36.202457427978516,
952
+ "logps/rejected": -46.09682083129883,
953
+ "loss": 0.5363,
954
+ "losses/dpo": 0.4159512221813202,
955
+ "losses/sft": 1.2818732261657715,
956
+ "losses/total": 0.4159512221813202,
957
+ "ref_logps/chosen": -27.4339542388916,
958
+ "ref_logps/rejected": -30.41155433654785,
959
+ "rewards/accuracies": 0.734375,
960
+ "rewards/chosen": -0.8768501281738281,
961
+ "rewards/margins": 0.6916766166687012,
962
+ "rewards/rejected": -1.5685267448425293,
963
+ "step": 53
964
+ },
965
+ {
966
+ "epoch": 0.4075471698113208,
967
+ "grad_norm": 5.318048000335693,
968
+ "learning_rate": 4.430379746835443e-06,
969
+ "logps/chosen": -40.78857421875,
970
+ "logps/rejected": -48.64263153076172,
971
+ "loss": 0.5552,
972
+ "losses/dpo": 0.5250239372253418,
973
+ "losses/sft": 1.3805458545684814,
974
+ "losses/total": 0.5250239372253418,
975
+ "ref_logps/chosen": -31.36919403076172,
976
+ "ref_logps/rejected": -32.82971954345703,
977
+ "rewards/accuracies": 0.6640625,
978
+ "rewards/chosen": -0.9419378638267517,
979
+ "rewards/margins": 0.6393535733222961,
980
+ "rewards/rejected": -1.5812914371490479,
981
+ "step": 54
982
+ },
983
+ {
984
+ "epoch": 0.41509433962264153,
985
+ "grad_norm": 5.705236911773682,
986
+ "learning_rate": 4.409282700421942e-06,
987
+ "logps/chosen": -39.51799011230469,
988
+ "logps/rejected": -50.53668975830078,
989
+ "loss": 0.5586,
990
+ "losses/dpo": 0.7799692153930664,
991
+ "losses/sft": 1.2691373825073242,
992
+ "losses/total": 0.7799692153930664,
993
+ "ref_logps/chosen": -29.568626403808594,
994
+ "ref_logps/rejected": -34.02891540527344,
995
+ "rewards/accuracies": 0.6953125,
996
+ "rewards/chosen": -0.9949361681938171,
997
+ "rewards/margins": 0.6558418273925781,
998
+ "rewards/rejected": -1.6507779359817505,
999
+ "step": 55
1000
+ },
1001
+ {
1002
+ "epoch": 0.4226415094339623,
1003
+ "grad_norm": 5.554488658905029,
1004
+ "learning_rate": 4.3881856540084394e-06,
1005
+ "logps/chosen": -38.21480941772461,
1006
+ "logps/rejected": -47.99286651611328,
1007
+ "loss": 0.5795,
1008
+ "losses/dpo": 0.5031176805496216,
1009
+ "losses/sft": 1.3220183849334717,
1010
+ "losses/total": 0.5031176805496216,
1011
+ "ref_logps/chosen": -29.2410888671875,
1012
+ "ref_logps/rejected": -33.15393829345703,
1013
+ "rewards/accuracies": 0.703125,
1014
+ "rewards/chosen": -0.8973721265792847,
1015
+ "rewards/margins": 0.5865209698677063,
1016
+ "rewards/rejected": -1.4838931560516357,
1017
+ "step": 56
1018
+ },
1019
+ {
1020
+ "epoch": 0.43018867924528303,
1021
+ "grad_norm": 5.692888259887695,
1022
+ "learning_rate": 4.367088607594937e-06,
1023
+ "logps/chosen": -41.21876907348633,
1024
+ "logps/rejected": -53.25712585449219,
1025
+ "loss": 0.5296,
1026
+ "losses/dpo": 0.4523683488368988,
1027
+ "losses/sft": 1.5277466773986816,
1028
+ "losses/total": 0.4523683488368988,
1029
+ "ref_logps/chosen": -31.35172462463379,
1030
+ "ref_logps/rejected": -35.32482147216797,
1031
+ "rewards/accuracies": 0.7421875,
1032
+ "rewards/chosen": -0.9867046475410461,
1033
+ "rewards/margins": 0.806525707244873,
1034
+ "rewards/rejected": -1.7932302951812744,
1035
+ "step": 57
1036
+ },
1037
+ {
1038
+ "epoch": 0.4377358490566038,
1039
+ "grad_norm": 5.991930961608887,
1040
+ "learning_rate": 4.345991561181435e-06,
1041
+ "logps/chosen": -39.15966033935547,
1042
+ "logps/rejected": -50.0914306640625,
1043
+ "loss": 0.5912,
1044
+ "losses/dpo": 0.8516952991485596,
1045
+ "losses/sft": 1.4551334381103516,
1046
+ "losses/total": 0.8516952991485596,
1047
+ "ref_logps/chosen": -28.90890884399414,
1048
+ "ref_logps/rejected": -32.83641052246094,
1049
+ "rewards/accuracies": 0.6953125,
1050
+ "rewards/chosen": -1.0250749588012695,
1051
+ "rewards/margins": 0.7004267573356628,
1052
+ "rewards/rejected": -1.7255017757415771,
1053
+ "step": 58
1054
+ },
1055
+ {
1056
+ "epoch": 0.44528301886792454,
1057
+ "grad_norm": 5.214711666107178,
1058
+ "learning_rate": 4.324894514767933e-06,
1059
+ "logps/chosen": -40.67414474487305,
1060
+ "logps/rejected": -57.66020584106445,
1061
+ "loss": 0.4895,
1062
+ "losses/dpo": 0.5767215490341187,
1063
+ "losses/sft": 1.3480836153030396,
1064
+ "losses/total": 0.5767215490341187,
1065
+ "ref_logps/chosen": -31.317989349365234,
1066
+ "ref_logps/rejected": -38.934444427490234,
1067
+ "rewards/accuracies": 0.7578125,
1068
+ "rewards/chosen": -0.9356154799461365,
1069
+ "rewards/margins": 0.93696129322052,
1070
+ "rewards/rejected": -1.8725767135620117,
1071
+ "step": 59
1072
+ },
1073
+ {
1074
+ "epoch": 0.4528301886792453,
1075
+ "grad_norm": 6.370638847351074,
1076
+ "learning_rate": 4.303797468354431e-06,
1077
+ "logps/chosen": -40.94059753417969,
1078
+ "logps/rejected": -51.065185546875,
1079
+ "loss": 0.6121,
1080
+ "losses/dpo": 0.4795827865600586,
1081
+ "losses/sft": 1.4322285652160645,
1082
+ "losses/total": 0.4795827865600586,
1083
+ "ref_logps/chosen": -29.871326446533203,
1084
+ "ref_logps/rejected": -34.67937469482422,
1085
+ "rewards/accuracies": 0.703125,
1086
+ "rewards/chosen": -1.1069271564483643,
1087
+ "rewards/margins": 0.5316535830497742,
1088
+ "rewards/rejected": -1.6385807991027832,
1089
+ "step": 60
1090
+ },
1091
+ {
1092
+ "epoch": 0.46037735849056605,
1093
+ "grad_norm": 6.262545585632324,
1094
+ "learning_rate": 4.2827004219409285e-06,
1095
+ "logps/chosen": -39.27662658691406,
1096
+ "logps/rejected": -55.63613510131836,
1097
+ "loss": 0.5328,
1098
+ "losses/dpo": 0.596257746219635,
1099
+ "losses/sft": 1.5779916048049927,
1100
+ "losses/total": 0.596257746219635,
1101
+ "ref_logps/chosen": -29.888713836669922,
1102
+ "ref_logps/rejected": -38.11284637451172,
1103
+ "rewards/accuracies": 0.7734375,
1104
+ "rewards/chosen": -0.9387915134429932,
1105
+ "rewards/margins": 0.8135374784469604,
1106
+ "rewards/rejected": -1.752328872680664,
1107
+ "step": 61
1108
+ },
1109
+ {
1110
+ "epoch": 0.4679245283018868,
1111
+ "grad_norm": 5.742541313171387,
1112
+ "learning_rate": 4.261603375527426e-06,
1113
+ "logps/chosen": -40.12831497192383,
1114
+ "logps/rejected": -48.92723846435547,
1115
+ "loss": 0.5512,
1116
+ "losses/dpo": 0.6246651411056519,
1117
+ "losses/sft": 1.4945807456970215,
1118
+ "losses/total": 0.6246651411056519,
1119
+ "ref_logps/chosen": -31.467824935913086,
1120
+ "ref_logps/rejected": -33.796714782714844,
1121
+ "rewards/accuracies": 0.7109375,
1122
+ "rewards/chosen": -0.86604905128479,
1123
+ "rewards/margins": 0.6470035314559937,
1124
+ "rewards/rejected": -1.5130527019500732,
1125
+ "step": 62
1126
+ },
1127
+ {
1128
+ "epoch": 0.47547169811320755,
1129
+ "grad_norm": 5.606804370880127,
1130
+ "learning_rate": 4.240506329113924e-06,
1131
+ "logps/chosen": -38.19676971435547,
1132
+ "logps/rejected": -51.48821258544922,
1133
+ "loss": 0.5277,
1134
+ "losses/dpo": 0.6507617235183716,
1135
+ "losses/sft": 1.7701728343963623,
1136
+ "losses/total": 0.6507617235183716,
1137
+ "ref_logps/chosen": -28.915279388427734,
1138
+ "ref_logps/rejected": -34.66787338256836,
1139
+ "rewards/accuracies": 0.7421875,
1140
+ "rewards/chosen": -0.9281493425369263,
1141
+ "rewards/margins": 0.7538847327232361,
1142
+ "rewards/rejected": -1.6820340156555176,
1143
+ "step": 63
1144
+ },
1145
+ {
1146
+ "epoch": 0.4830188679245283,
1147
+ "grad_norm": 5.0319647789001465,
1148
+ "learning_rate": 4.219409282700423e-06,
1149
+ "logps/chosen": -36.99481964111328,
1150
+ "logps/rejected": -49.86357116699219,
1151
+ "loss": 0.5064,
1152
+ "losses/dpo": 0.3545170724391937,
1153
+ "losses/sft": 1.4933311939239502,
1154
+ "losses/total": 0.3545170724391937,
1155
+ "ref_logps/chosen": -27.901212692260742,
1156
+ "ref_logps/rejected": -32.68770980834961,
1157
+ "rewards/accuracies": 0.7265625,
1158
+ "rewards/chosen": -0.9093605875968933,
1159
+ "rewards/margins": 0.8082252144813538,
1160
+ "rewards/rejected": -1.717585802078247,
1161
+ "step": 64
1162
+ },
1163
+ {
1164
+ "epoch": 0.49056603773584906,
1165
+ "grad_norm": 5.624882698059082,
1166
+ "learning_rate": 4.19831223628692e-06,
1167
+ "logps/chosen": -41.812522888183594,
1168
+ "logps/rejected": -50.7585334777832,
1169
+ "loss": 0.5876,
1170
+ "losses/dpo": 0.7280862331390381,
1171
+ "losses/sft": 1.730393648147583,
1172
+ "losses/total": 0.7280862331390381,
1173
+ "ref_logps/chosen": -32.30836868286133,
1174
+ "ref_logps/rejected": -35.13983917236328,
1175
+ "rewards/accuracies": 0.734375,
1176
+ "rewards/chosen": -0.9504156708717346,
1177
+ "rewards/margins": 0.6114538908004761,
1178
+ "rewards/rejected": -1.5618693828582764,
1179
+ "step": 65
1180
+ },
1181
+ {
1182
+ "epoch": 0.4981132075471698,
1183
+ "grad_norm": 5.693253517150879,
1184
+ "learning_rate": 4.177215189873418e-06,
1185
+ "logps/chosen": -41.0583381652832,
1186
+ "logps/rejected": -57.39263153076172,
1187
+ "loss": 0.4943,
1188
+ "losses/dpo": 0.4355122745037079,
1189
+ "losses/sft": 1.1784332990646362,
1190
+ "losses/total": 0.4355122745037079,
1191
+ "ref_logps/chosen": -31.086488723754883,
1192
+ "ref_logps/rejected": -38.95460891723633,
1193
+ "rewards/accuracies": 0.7734375,
1194
+ "rewards/chosen": -0.9971848726272583,
1195
+ "rewards/margins": 0.8466169834136963,
1196
+ "rewards/rejected": -1.8438018560409546,
1197
+ "step": 66
1198
+ },
1199
+ {
1200
+ "epoch": 0.5056603773584906,
1201
+ "grad_norm": 7.065197944641113,
1202
+ "learning_rate": 4.156118143459915e-06,
1203
+ "logps/chosen": -39.25422668457031,
1204
+ "logps/rejected": -50.87791442871094,
1205
+ "loss": 0.5339,
1206
+ "losses/dpo": 0.8317179083824158,
1207
+ "losses/sft": 1.3175883293151855,
1208
+ "losses/total": 0.8317179083824158,
1209
+ "ref_logps/chosen": -28.742538452148438,
1210
+ "ref_logps/rejected": -33.21300506591797,
1211
+ "rewards/accuracies": 0.71875,
1212
+ "rewards/chosen": -1.0511683225631714,
1213
+ "rewards/margins": 0.7153225541114807,
1214
+ "rewards/rejected": -1.7664909362792969,
1215
+ "step": 67
1216
+ },
1217
+ {
1218
+ "epoch": 0.5132075471698113,
1219
+ "grad_norm": 5.2859907150268555,
1220
+ "learning_rate": 4.135021097046414e-06,
1221
+ "logps/chosen": -38.8599853515625,
1222
+ "logps/rejected": -53.00445556640625,
1223
+ "loss": 0.4691,
1224
+ "losses/dpo": 0.5039323568344116,
1225
+ "losses/sft": 0.7760534286499023,
1226
+ "losses/total": 0.5039323568344116,
1227
+ "ref_logps/chosen": -30.572181701660156,
1228
+ "ref_logps/rejected": -35.3000602722168,
1229
+ "rewards/accuracies": 0.78125,
1230
+ "rewards/chosen": -0.828780472278595,
1231
+ "rewards/margins": 0.9416592121124268,
1232
+ "rewards/rejected": -1.770439863204956,
1233
+ "step": 68
1234
+ },
1235
+ {
1236
+ "epoch": 0.5207547169811321,
1237
+ "grad_norm": 5.869091033935547,
1238
+ "learning_rate": 4.113924050632912e-06,
1239
+ "logps/chosen": -39.38776397705078,
1240
+ "logps/rejected": -56.88161849975586,
1241
+ "loss": 0.5432,
1242
+ "losses/dpo": 0.28467729687690735,
1243
+ "losses/sft": 1.18190336227417,
1244
+ "losses/total": 0.28467729687690735,
1245
+ "ref_logps/chosen": -28.762392044067383,
1246
+ "ref_logps/rejected": -37.78593063354492,
1247
+ "rewards/accuracies": 0.7109375,
1248
+ "rewards/chosen": -1.0625369548797607,
1249
+ "rewards/margins": 0.8470318913459778,
1250
+ "rewards/rejected": -1.9095687866210938,
1251
+ "step": 69
1252
+ },
1253
+ {
1254
+ "epoch": 0.5283018867924528,
1255
+ "grad_norm": 5.994374752044678,
1256
+ "learning_rate": 4.09282700421941e-06,
1257
+ "logps/chosen": -39.55097579956055,
1258
+ "logps/rejected": -50.52848815917969,
1259
+ "loss": 0.5564,
1260
+ "losses/dpo": 0.6404341459274292,
1261
+ "losses/sft": 1.6006711721420288,
1262
+ "losses/total": 0.6404341459274292,
1263
+ "ref_logps/chosen": -29.427146911621094,
1264
+ "ref_logps/rejected": -32.81227111816406,
1265
+ "rewards/accuracies": 0.7109375,
1266
+ "rewards/chosen": -1.0123828649520874,
1267
+ "rewards/margins": 0.759239137172699,
1268
+ "rewards/rejected": -1.7716220617294312,
1269
+ "step": 70
1270
+ },
1271
+ {
1272
+ "epoch": 0.5358490566037736,
1273
+ "grad_norm": 6.391396999359131,
1274
+ "learning_rate": 4.0717299578059074e-06,
1275
+ "logps/chosen": -39.2120475769043,
1276
+ "logps/rejected": -53.28424072265625,
1277
+ "loss": 0.5533,
1278
+ "losses/dpo": 0.35934409499168396,
1279
+ "losses/sft": 1.446973443031311,
1280
+ "losses/total": 0.35934409499168396,
1281
+ "ref_logps/chosen": -28.753549575805664,
1282
+ "ref_logps/rejected": -34.92060470581055,
1283
+ "rewards/accuracies": 0.71875,
1284
+ "rewards/chosen": -1.0458496809005737,
1285
+ "rewards/margins": 0.790514349937439,
1286
+ "rewards/rejected": -1.8363640308380127,
1287
+ "step": 71
1288
+ },
1289
+ {
1290
+ "epoch": 0.5433962264150943,
1291
+ "grad_norm": 6.23837423324585,
1292
+ "learning_rate": 4.050632911392405e-06,
1293
+ "logps/chosen": -41.26081466674805,
1294
+ "logps/rejected": -57.397159576416016,
1295
+ "loss": 0.5566,
1296
+ "losses/dpo": 0.6022348403930664,
1297
+ "losses/sft": 1.4900131225585938,
1298
+ "losses/total": 0.6022348403930664,
1299
+ "ref_logps/chosen": -29.823455810546875,
1300
+ "ref_logps/rejected": -38.314422607421875,
1301
+ "rewards/accuracies": 0.6953125,
1302
+ "rewards/chosen": -1.1437358856201172,
1303
+ "rewards/margins": 0.7645378112792969,
1304
+ "rewards/rejected": -1.908273696899414,
1305
+ "step": 72
1306
+ },
1307
+ {
1308
+ "epoch": 0.5509433962264151,
1309
+ "grad_norm": 5.54686164855957,
1310
+ "learning_rate": 4.029535864978903e-06,
1311
+ "logps/chosen": -40.462684631347656,
1312
+ "logps/rejected": -56.720314025878906,
1313
+ "loss": 0.4834,
1314
+ "losses/dpo": 0.5470014810562134,
1315
+ "losses/sft": 1.4223295450210571,
1316
+ "losses/total": 0.5470014810562134,
1317
+ "ref_logps/chosen": -30.118234634399414,
1318
+ "ref_logps/rejected": -36.64280700683594,
1319
+ "rewards/accuracies": 0.75,
1320
+ "rewards/chosen": -1.0344449281692505,
1321
+ "rewards/margins": 0.9733060002326965,
1322
+ "rewards/rejected": -2.007750988006592,
1323
+ "step": 73
1324
+ },
1325
+ {
1326
+ "epoch": 0.5584905660377358,
1327
+ "grad_norm": 6.6403656005859375,
1328
+ "learning_rate": 4.008438818565401e-06,
1329
+ "logps/chosen": -43.17585754394531,
1330
+ "logps/rejected": -54.99962615966797,
1331
+ "loss": 0.6214,
1332
+ "losses/dpo": 0.8616761565208435,
1333
+ "losses/sft": 1.4494423866271973,
1334
+ "losses/total": 0.8616761565208435,
1335
+ "ref_logps/chosen": -30.735164642333984,
1336
+ "ref_logps/rejected": -36.314048767089844,
1337
+ "rewards/accuracies": 0.6875,
1338
+ "rewards/chosen": -1.2440688610076904,
1339
+ "rewards/margins": 0.6244890689849854,
1340
+ "rewards/rejected": -1.8685579299926758,
1341
+ "step": 74
1342
+ },
1343
+ {
1344
+ "epoch": 0.5660377358490566,
1345
+ "grad_norm": 5.264912128448486,
1346
+ "learning_rate": 3.9873417721518995e-06,
1347
+ "logps/chosen": -40.38694763183594,
1348
+ "logps/rejected": -57.54372024536133,
1349
+ "loss": 0.4474,
1350
+ "losses/dpo": 0.49264034628868103,
1351
+ "losses/sft": 1.4342858791351318,
1352
+ "losses/total": 0.49264034628868103,
1353
+ "ref_logps/chosen": -30.248275756835938,
1354
+ "ref_logps/rejected": -35.96062469482422,
1355
+ "rewards/accuracies": 0.8046875,
1356
+ "rewards/chosen": -1.0138667821884155,
1357
+ "rewards/margins": 1.1444426774978638,
1358
+ "rewards/rejected": -2.1583094596862793,
1359
+ "step": 75
1360
+ },
1361
+ {
1362
+ "epoch": 0.5735849056603773,
1363
+ "grad_norm": 5.621297836303711,
1364
+ "learning_rate": 3.9662447257383965e-06,
1365
+ "logps/chosen": -40.704254150390625,
1366
+ "logps/rejected": -54.67380905151367,
1367
+ "loss": 0.5208,
1368
+ "losses/dpo": 0.3526512086391449,
1369
+ "losses/sft": 1.205520749092102,
1370
+ "losses/total": 0.3526512086391449,
1371
+ "ref_logps/chosen": -30.080062866210938,
1372
+ "ref_logps/rejected": -35.63865661621094,
1373
+ "rewards/accuracies": 0.7265625,
1374
+ "rewards/chosen": -1.0624192953109741,
1375
+ "rewards/margins": 0.8410958647727966,
1376
+ "rewards/rejected": -1.903515100479126,
1377
+ "step": 76
1378
+ },
1379
+ {
1380
+ "epoch": 0.5811320754716981,
1381
+ "grad_norm": 5.440826416015625,
1382
+ "learning_rate": 3.945147679324895e-06,
1383
+ "logps/chosen": -38.52376174926758,
1384
+ "logps/rejected": -51.74415588378906,
1385
+ "loss": 0.5272,
1386
+ "losses/dpo": 0.43253982067108154,
1387
+ "losses/sft": 1.3709319829940796,
1388
+ "losses/total": 0.43253982067108154,
1389
+ "ref_logps/chosen": -29.032743453979492,
1390
+ "ref_logps/rejected": -34.911163330078125,
1391
+ "rewards/accuracies": 0.75,
1392
+ "rewards/chosen": -0.9491016268730164,
1393
+ "rewards/margins": 0.7341974377632141,
1394
+ "rewards/rejected": -1.68329918384552,
1395
+ "step": 77
1396
+ },
1397
+ {
1398
+ "epoch": 0.5886792452830188,
1399
+ "grad_norm": 4.8831377029418945,
1400
+ "learning_rate": 3.924050632911393e-06,
1401
+ "logps/chosen": -37.76980972290039,
1402
+ "logps/rejected": -57.21007537841797,
1403
+ "loss": 0.4277,
1404
+ "losses/dpo": 0.4043513536453247,
1405
+ "losses/sft": 1.341131329536438,
1406
+ "losses/total": 0.4043513536453247,
1407
+ "ref_logps/chosen": -29.5567684173584,
1408
+ "ref_logps/rejected": -37.932891845703125,
1409
+ "rewards/accuracies": 0.828125,
1410
+ "rewards/chosen": -0.8213039636611938,
1411
+ "rewards/margins": 1.1064141988754272,
1412
+ "rewards/rejected": -1.927718162536621,
1413
+ "step": 78
1414
+ },
1415
+ {
1416
+ "epoch": 0.5962264150943396,
1417
+ "grad_norm": 4.781276702880859,
1418
+ "learning_rate": 3.902953586497891e-06,
1419
+ "logps/chosen": -39.535728454589844,
1420
+ "logps/rejected": -55.912841796875,
1421
+ "loss": 0.4133,
1422
+ "losses/dpo": 0.22891421616077423,
1423
+ "losses/sft": 1.0599099397659302,
1424
+ "losses/total": 0.22891421616077423,
1425
+ "ref_logps/chosen": -30.658559799194336,
1426
+ "ref_logps/rejected": -35.42290115356445,
1427
+ "rewards/accuracies": 0.8359375,
1428
+ "rewards/chosen": -0.8877166509628296,
1429
+ "rewards/margins": 1.1612776517868042,
1430
+ "rewards/rejected": -2.048994302749634,
1431
+ "step": 79
1432
+ },
1433
+ {
1434
+ "epoch": 0.6037735849056604,
1435
+ "grad_norm": 5.942996501922607,
1436
+ "learning_rate": 3.8818565400843886e-06,
1437
+ "logps/chosen": -42.60219192504883,
1438
+ "logps/rejected": -59.31925582885742,
1439
+ "loss": 0.4876,
1440
+ "losses/dpo": 0.18772652745246887,
1441
+ "losses/sft": 1.6294547319412231,
1442
+ "losses/total": 0.18772652745246887,
1443
+ "ref_logps/chosen": -31.457292556762695,
1444
+ "ref_logps/rejected": -36.81336212158203,
1445
+ "rewards/accuracies": 0.7578125,
1446
+ "rewards/chosen": -1.114490032196045,
1447
+ "rewards/margins": 1.1360994577407837,
1448
+ "rewards/rejected": -2.250589370727539,
1449
+ "step": 80
1450
+ },
1451
+ {
1452
+ "epoch": 0.6113207547169811,
1453
+ "grad_norm": 6.1673102378845215,
1454
+ "learning_rate": 3.860759493670886e-06,
1455
+ "logps/chosen": -45.821449279785156,
1456
+ "logps/rejected": -57.79277420043945,
1457
+ "loss": 0.5474,
1458
+ "losses/dpo": 0.5481749773025513,
1459
+ "losses/sft": 1.9008748531341553,
1460
+ "losses/total": 0.5481749773025513,
1461
+ "ref_logps/chosen": -33.29500961303711,
1462
+ "ref_logps/rejected": -35.532379150390625,
1463
+ "rewards/accuracies": 0.734375,
1464
+ "rewards/chosen": -1.2526437044143677,
1465
+ "rewards/margins": 0.9733958840370178,
1466
+ "rewards/rejected": -2.2260396480560303,
1467
+ "step": 81
1468
+ },
1469
+ {
1470
+ "epoch": 0.6188679245283019,
1471
+ "grad_norm": 6.780162811279297,
1472
+ "learning_rate": 3.839662447257384e-06,
1473
+ "logps/chosen": -42.07637405395508,
1474
+ "logps/rejected": -53.19427490234375,
1475
+ "loss": 0.6369,
1476
+ "losses/dpo": 0.47502392530441284,
1477
+ "losses/sft": 1.4819515943527222,
1478
+ "losses/total": 0.47502392530441284,
1479
+ "ref_logps/chosen": -30.144392013549805,
1480
+ "ref_logps/rejected": -34.309967041015625,
1481
+ "rewards/accuracies": 0.6640625,
1482
+ "rewards/chosen": -1.193198323249817,
1483
+ "rewards/margins": 0.695232629776001,
1484
+ "rewards/rejected": -1.8884310722351074,
1485
+ "step": 82
1486
+ },
1487
+ {
1488
+ "epoch": 0.6264150943396226,
1489
+ "grad_norm": 6.206979751586914,
1490
+ "learning_rate": 3.818565400843882e-06,
1491
+ "logps/chosen": -41.30870056152344,
1492
+ "logps/rejected": -53.384010314941406,
1493
+ "loss": 0.5578,
1494
+ "losses/dpo": 0.7641366720199585,
1495
+ "losses/sft": 1.1807348728179932,
1496
+ "losses/total": 0.7641366720199585,
1497
+ "ref_logps/chosen": -30.49885368347168,
1498
+ "ref_logps/rejected": -34.54371643066406,
1499
+ "rewards/accuracies": 0.7421875,
1500
+ "rewards/chosen": -1.0809844732284546,
1501
+ "rewards/margins": 0.803044319152832,
1502
+ "rewards/rejected": -1.8840289115905762,
1503
+ "step": 83
1504
+ },
1505
+ {
1506
+ "epoch": 0.6339622641509434,
1507
+ "grad_norm": 6.179627418518066,
1508
+ "learning_rate": 3.7974683544303802e-06,
1509
+ "logps/chosen": -43.51046371459961,
1510
+ "logps/rejected": -57.6606330871582,
1511
+ "loss": 0.5393,
1512
+ "losses/dpo": 0.2461186796426773,
1513
+ "losses/sft": 1.4413666725158691,
1514
+ "losses/total": 0.2461186796426773,
1515
+ "ref_logps/chosen": -31.447532653808594,
1516
+ "ref_logps/rejected": -35.95327377319336,
1517
+ "rewards/accuracies": 0.71875,
1518
+ "rewards/chosen": -1.2062931060791016,
1519
+ "rewards/margins": 0.964443027973175,
1520
+ "rewards/rejected": -2.170736312866211,
1521
+ "step": 84
1522
+ },
1523
+ {
1524
+ "epoch": 0.6415094339622641,
1525
+ "grad_norm": 5.617598533630371,
1526
+ "learning_rate": 3.776371308016878e-06,
1527
+ "logps/chosen": -40.44061279296875,
1528
+ "logps/rejected": -58.36695098876953,
1529
+ "loss": 0.4637,
1530
+ "losses/dpo": 0.32039570808410645,
1531
+ "losses/sft": 1.308812141418457,
1532
+ "losses/total": 0.32039570808410645,
1533
+ "ref_logps/chosen": -30.45447540283203,
1534
+ "ref_logps/rejected": -36.56258010864258,
1535
+ "rewards/accuracies": 0.8203125,
1536
+ "rewards/chosen": -0.9986135363578796,
1537
+ "rewards/margins": 1.1818233728408813,
1538
+ "rewards/rejected": -2.180436849594116,
1539
+ "step": 85
1540
+ },
1541
+ {
1542
+ "epoch": 0.6490566037735849,
1543
+ "grad_norm": 6.853886604309082,
1544
+ "learning_rate": 3.755274261603376e-06,
1545
+ "logps/chosen": -42.52161407470703,
1546
+ "logps/rejected": -54.287841796875,
1547
+ "loss": 0.5991,
1548
+ "losses/dpo": 0.3581075370311737,
1549
+ "losses/sft": 1.5368932485580444,
1550
+ "losses/total": 0.3581075370311737,
1551
+ "ref_logps/chosen": -31.34189796447754,
1552
+ "ref_logps/rejected": -35.659629821777344,
1553
+ "rewards/accuracies": 0.671875,
1554
+ "rewards/chosen": -1.117971658706665,
1555
+ "rewards/margins": 0.7448497414588928,
1556
+ "rewards/rejected": -1.8628215789794922,
1557
+ "step": 86
1558
+ },
1559
+ {
1560
+ "epoch": 0.6566037735849056,
1561
+ "grad_norm": 6.692890167236328,
1562
+ "learning_rate": 3.7341772151898737e-06,
1563
+ "logps/chosen": -43.72019958496094,
1564
+ "logps/rejected": -58.138694763183594,
1565
+ "loss": 0.5519,
1566
+ "losses/dpo": 0.48806032538414,
1567
+ "losses/sft": 0.9289418458938599,
1568
+ "losses/total": 0.48806032538414,
1569
+ "ref_logps/chosen": -31.137048721313477,
1570
+ "ref_logps/rejected": -36.59523010253906,
1571
+ "rewards/accuracies": 0.75,
1572
+ "rewards/chosen": -1.2583153247833252,
1573
+ "rewards/margins": 0.8960307836532593,
1574
+ "rewards/rejected": -2.154345989227295,
1575
+ "step": 87
1576
+ },
1577
+ {
1578
+ "epoch": 0.6641509433962264,
1579
+ "grad_norm": 5.944717884063721,
1580
+ "learning_rate": 3.713080168776372e-06,
1581
+ "logps/chosen": -43.606327056884766,
1582
+ "logps/rejected": -54.38173294067383,
1583
+ "loss": 0.525,
1584
+ "losses/dpo": 0.4686363637447357,
1585
+ "losses/sft": 1.4131765365600586,
1586
+ "losses/total": 0.4686363637447357,
1587
+ "ref_logps/chosen": -32.12074279785156,
1588
+ "ref_logps/rejected": -35.17863464355469,
1589
+ "rewards/accuracies": 0.7421875,
1590
+ "rewards/chosen": -1.1485581398010254,
1591
+ "rewards/margins": 0.7717516422271729,
1592
+ "rewards/rejected": -1.9203099012374878,
1593
+ "step": 88
1594
+ },
1595
+ {
1596
+ "epoch": 0.6716981132075471,
1597
+ "grad_norm": 5.189677715301514,
1598
+ "learning_rate": 3.6919831223628693e-06,
1599
+ "logps/chosen": -41.43803787231445,
1600
+ "logps/rejected": -58.55611038208008,
1601
+ "loss": 0.4709,
1602
+ "losses/dpo": 0.27158284187316895,
1603
+ "losses/sft": 1.2001585960388184,
1604
+ "losses/total": 0.27158284187316895,
1605
+ "ref_logps/chosen": -30.10678482055664,
1606
+ "ref_logps/rejected": -36.89292907714844,
1607
+ "rewards/accuracies": 0.7578125,
1608
+ "rewards/chosen": -1.1331257820129395,
1609
+ "rewards/margins": 1.0331923961639404,
1610
+ "rewards/rejected": -2.16631817817688,
1611
+ "step": 89
1612
+ },
1613
+ {
1614
+ "epoch": 0.6792452830188679,
1615
+ "grad_norm": 5.98684024810791,
1616
+ "learning_rate": 3.6708860759493675e-06,
1617
+ "logps/chosen": -43.76606750488281,
1618
+ "logps/rejected": -52.34052276611328,
1619
+ "loss": 0.5686,
1620
+ "losses/dpo": 0.7345205545425415,
1621
+ "losses/sft": 1.5648329257965088,
1622
+ "losses/total": 0.7345205545425415,
1623
+ "ref_logps/chosen": -31.27000617980957,
1624
+ "ref_logps/rejected": -32.98979187011719,
1625
+ "rewards/accuracies": 0.71875,
1626
+ "rewards/chosen": -1.2496063709259033,
1627
+ "rewards/margins": 0.6854668855667114,
1628
+ "rewards/rejected": -1.9350732564926147,
1629
+ "step": 90
1630
+ },
1631
+ {
1632
+ "epoch": 0.6867924528301886,
1633
+ "grad_norm": 5.93943452835083,
1634
+ "learning_rate": 3.649789029535865e-06,
1635
+ "logps/chosen": -44.927276611328125,
1636
+ "logps/rejected": -56.36608123779297,
1637
+ "loss": 0.5478,
1638
+ "losses/dpo": 0.31034794449806213,
1639
+ "losses/sft": 1.5676113367080688,
1640
+ "losses/total": 0.31034794449806213,
1641
+ "ref_logps/chosen": -31.989208221435547,
1642
+ "ref_logps/rejected": -35.02646255493164,
1643
+ "rewards/accuracies": 0.75,
1644
+ "rewards/chosen": -1.2938066720962524,
1645
+ "rewards/margins": 0.8401551246643066,
1646
+ "rewards/rejected": -2.1339616775512695,
1647
+ "step": 91
1648
+ },
1649
+ {
1650
+ "epoch": 0.6943396226415094,
1651
+ "grad_norm": 5.51214075088501,
1652
+ "learning_rate": 3.628691983122363e-06,
1653
+ "logps/chosen": -44.41264343261719,
1654
+ "logps/rejected": -60.244510650634766,
1655
+ "loss": 0.4797,
1656
+ "losses/dpo": 0.42363566160202026,
1657
+ "losses/sft": 1.270397424697876,
1658
+ "losses/total": 0.42363566160202026,
1659
+ "ref_logps/chosen": -33.250125885009766,
1660
+ "ref_logps/rejected": -40.045772552490234,
1661
+ "rewards/accuracies": 0.75,
1662
+ "rewards/chosen": -1.1162512302398682,
1663
+ "rewards/margins": 0.9036226868629456,
1664
+ "rewards/rejected": -2.019874095916748,
1665
+ "step": 92
1666
+ },
1667
+ {
1668
+ "epoch": 0.7018867924528301,
1669
+ "grad_norm": 5.8951945304870605,
1670
+ "learning_rate": 3.607594936708861e-06,
1671
+ "logps/chosen": -41.33027648925781,
1672
+ "logps/rejected": -54.42854309082031,
1673
+ "loss": 0.5894,
1674
+ "losses/dpo": 0.31869563460350037,
1675
+ "losses/sft": 1.346259593963623,
1676
+ "losses/total": 0.31869563460350037,
1677
+ "ref_logps/chosen": -28.775463104248047,
1678
+ "ref_logps/rejected": -35.096275329589844,
1679
+ "rewards/accuracies": 0.703125,
1680
+ "rewards/chosen": -1.255481243133545,
1681
+ "rewards/margins": 0.6777457594871521,
1682
+ "rewards/rejected": -1.9332269430160522,
1683
+ "step": 93
1684
+ },
1685
+ {
1686
+ "epoch": 0.7094339622641509,
1687
+ "grad_norm": 6.491157531738281,
1688
+ "learning_rate": 3.586497890295359e-06,
1689
+ "logps/chosen": -43.51985168457031,
1690
+ "logps/rejected": -54.04357147216797,
1691
+ "loss": 0.5492,
1692
+ "losses/dpo": 0.6047073602676392,
1693
+ "losses/sft": 1.633582353591919,
1694
+ "losses/total": 0.6047073602676392,
1695
+ "ref_logps/chosen": -32.43757247924805,
1696
+ "ref_logps/rejected": -35.235679626464844,
1697
+ "rewards/accuracies": 0.6640625,
1698
+ "rewards/chosen": -1.1082279682159424,
1699
+ "rewards/margins": 0.7725614309310913,
1700
+ "rewards/rejected": -1.8807893991470337,
1701
+ "step": 94
1702
+ },
1703
+ {
1704
+ "epoch": 0.7169811320754716,
1705
+ "grad_norm": 5.706908702850342,
1706
+ "learning_rate": 3.5654008438818566e-06,
1707
+ "logps/chosen": -44.01639938354492,
1708
+ "logps/rejected": -56.538856506347656,
1709
+ "loss": 0.5286,
1710
+ "losses/dpo": 1.006984829902649,
1711
+ "losses/sft": 1.5528416633605957,
1712
+ "losses/total": 1.006984829902649,
1713
+ "ref_logps/chosen": -31.790197372436523,
1714
+ "ref_logps/rejected": -36.23029327392578,
1715
+ "rewards/accuracies": 0.7578125,
1716
+ "rewards/chosen": -1.2226204872131348,
1717
+ "rewards/margins": 0.8082359433174133,
1718
+ "rewards/rejected": -2.0308563709259033,
1719
+ "step": 95
1720
+ },
1721
+ {
1722
+ "epoch": 0.7245283018867924,
1723
+ "grad_norm": 5.353209018707275,
1724
+ "learning_rate": 3.544303797468355e-06,
1725
+ "logps/chosen": -39.303466796875,
1726
+ "logps/rejected": -55.11781311035156,
1727
+ "loss": 0.489,
1728
+ "losses/dpo": 0.3290543556213379,
1729
+ "losses/sft": 1.5605956315994263,
1730
+ "losses/total": 0.3290543556213379,
1731
+ "ref_logps/chosen": -30.248308181762695,
1732
+ "ref_logps/rejected": -37.410919189453125,
1733
+ "rewards/accuracies": 0.765625,
1734
+ "rewards/chosen": -0.9055157899856567,
1735
+ "rewards/margins": 0.8651739358901978,
1736
+ "rewards/rejected": -1.770689606666565,
1737
+ "step": 96
1738
+ },
1739
+ {
1740
+ "epoch": 0.7320754716981132,
1741
+ "grad_norm": 5.129034996032715,
1742
+ "learning_rate": 3.523206751054853e-06,
1743
+ "logps/chosen": -39.52154541015625,
1744
+ "logps/rejected": -57.900733947753906,
1745
+ "loss": 0.4432,
1746
+ "losses/dpo": 0.5874547958374023,
1747
+ "losses/sft": 2.055988073348999,
1748
+ "losses/total": 0.5874547958374023,
1749
+ "ref_logps/chosen": -30.57717514038086,
1750
+ "ref_logps/rejected": -37.92079544067383,
1751
+ "rewards/accuracies": 0.828125,
1752
+ "rewards/chosen": -0.8944366574287415,
1753
+ "rewards/margins": 1.1035571098327637,
1754
+ "rewards/rejected": -1.99799382686615,
1755
+ "step": 97
1756
+ },
1757
+ {
1758
+ "epoch": 0.7396226415094339,
1759
+ "grad_norm": 5.658941745758057,
1760
+ "learning_rate": 3.5021097046413504e-06,
1761
+ "logps/chosen": -42.83544158935547,
1762
+ "logps/rejected": -51.84044647216797,
1763
+ "loss": 0.5452,
1764
+ "losses/dpo": 0.711673378944397,
1765
+ "losses/sft": 1.195967674255371,
1766
+ "losses/total": 0.711673378944397,
1767
+ "ref_logps/chosen": -32.588661193847656,
1768
+ "ref_logps/rejected": -34.962364196777344,
1769
+ "rewards/accuracies": 0.7734375,
1770
+ "rewards/chosen": -1.0246777534484863,
1771
+ "rewards/margins": 0.6631301641464233,
1772
+ "rewards/rejected": -1.6878077983856201,
1773
+ "step": 98
1774
+ },
1775
+ {
1776
+ "epoch": 0.7471698113207547,
1777
+ "grad_norm": 5.521385669708252,
1778
+ "learning_rate": 3.4810126582278487e-06,
1779
+ "logps/chosen": -38.521541595458984,
1780
+ "logps/rejected": -50.79652786254883,
1781
+ "loss": 0.5304,
1782
+ "losses/dpo": 0.3267762064933777,
1783
+ "losses/sft": 1.4650697708129883,
1784
+ "losses/total": 0.3267762064933777,
1785
+ "ref_logps/chosen": -28.887462615966797,
1786
+ "ref_logps/rejected": -33.21889877319336,
1787
+ "rewards/accuracies": 0.71875,
1788
+ "rewards/chosen": -0.9634078741073608,
1789
+ "rewards/margins": 0.7943546772003174,
1790
+ "rewards/rejected": -1.7577626705169678,
1791
+ "step": 99
1792
+ },
1793
+ {
1794
+ "epoch": 0.7547169811320755,
1795
+ "grad_norm": 6.092041492462158,
1796
+ "learning_rate": 3.459915611814346e-06,
1797
+ "logps/chosen": -43.435054779052734,
1798
+ "logps/rejected": -48.984378814697266,
1799
+ "loss": 0.6331,
1800
+ "losses/dpo": 0.2698668837547302,
1801
+ "losses/sft": 1.1696505546569824,
1802
+ "losses/total": 0.2698668837547302,
1803
+ "ref_logps/chosen": -31.87053871154785,
1804
+ "ref_logps/rejected": -31.63926887512207,
1805
+ "rewards/accuracies": 0.6953125,
1806
+ "rewards/chosen": -1.1564514636993408,
1807
+ "rewards/margins": 0.5780597925186157,
1808
+ "rewards/rejected": -1.734511375427246,
1809
+ "step": 100
1810
+ },
1811
+ {
1812
+ "epoch": 0.7622641509433963,
1813
+ "grad_norm": 5.298862934112549,
1814
+ "learning_rate": 3.4388185654008443e-06,
1815
+ "logps/chosen": -38.33399200439453,
1816
+ "logps/rejected": -53.81504821777344,
1817
+ "loss": 0.478,
1818
+ "losses/dpo": 0.4309498071670532,
1819
+ "losses/sft": 1.7226063013076782,
1820
+ "losses/total": 0.4309498071670532,
1821
+ "ref_logps/chosen": -29.21530532836914,
1822
+ "ref_logps/rejected": -36.157432556152344,
1823
+ "rewards/accuracies": 0.7890625,
1824
+ "rewards/chosen": -0.9118687510490417,
1825
+ "rewards/margins": 0.8538926839828491,
1826
+ "rewards/rejected": -1.765761375427246,
1827
+ "step": 101
1828
+ },
1829
+ {
1830
+ "epoch": 0.769811320754717,
1831
+ "grad_norm": 5.454709529876709,
1832
+ "learning_rate": 3.417721518987342e-06,
1833
+ "logps/chosen": -36.00552749633789,
1834
+ "logps/rejected": -47.35252380371094,
1835
+ "loss": 0.5585,
1836
+ "losses/dpo": 0.3503888249397278,
1837
+ "losses/sft": 1.201081395149231,
1838
+ "losses/total": 0.3503888249397278,
1839
+ "ref_logps/chosen": -27.280101776123047,
1840
+ "ref_logps/rejected": -31.31577491760254,
1841
+ "rewards/accuracies": 0.75,
1842
+ "rewards/chosen": -0.8725426197052002,
1843
+ "rewards/margins": 0.7311323881149292,
1844
+ "rewards/rejected": -1.6036750078201294,
1845
+ "step": 102
1846
+ },
1847
+ {
1848
+ "epoch": 0.7773584905660378,
1849
+ "grad_norm": 5.91898250579834,
1850
+ "learning_rate": 3.39662447257384e-06,
1851
+ "logps/chosen": -41.72618103027344,
1852
+ "logps/rejected": -53.16327667236328,
1853
+ "loss": 0.5765,
1854
+ "losses/dpo": 0.3589654564857483,
1855
+ "losses/sft": 1.56475031375885,
1856
+ "losses/total": 0.3589654564857483,
1857
+ "ref_logps/chosen": -30.899045944213867,
1858
+ "ref_logps/rejected": -34.665802001953125,
1859
+ "rewards/accuracies": 0.6875,
1860
+ "rewards/chosen": -1.0827136039733887,
1861
+ "rewards/margins": 0.7670339941978455,
1862
+ "rewards/rejected": -1.8497475385665894,
1863
+ "step": 103
1864
+ },
1865
+ {
1866
+ "epoch": 0.7849056603773585,
1867
+ "grad_norm": 5.455533981323242,
1868
+ "learning_rate": 3.3755274261603377e-06,
1869
+ "logps/chosen": -41.35075378417969,
1870
+ "logps/rejected": -52.81592559814453,
1871
+ "loss": 0.5401,
1872
+ "losses/dpo": 0.39715278148651123,
1873
+ "losses/sft": 1.4784698486328125,
1874
+ "losses/total": 0.39715278148651123,
1875
+ "ref_logps/chosen": -30.155914306640625,
1876
+ "ref_logps/rejected": -34.45831298828125,
1877
+ "rewards/accuracies": 0.734375,
1878
+ "rewards/chosen": -1.1194841861724854,
1879
+ "rewards/margins": 0.7162774801254272,
1880
+ "rewards/rejected": -1.835761547088623,
1881
+ "step": 104
1882
+ },
1883
+ {
1884
+ "epoch": 0.7924528301886793,
1885
+ "grad_norm": 4.831524848937988,
1886
+ "learning_rate": 3.354430379746836e-06,
1887
+ "logps/chosen": -38.7393684387207,
1888
+ "logps/rejected": -53.984886169433594,
1889
+ "loss": 0.4876,
1890
+ "losses/dpo": 0.5145635604858398,
1891
+ "losses/sft": 1.3864595890045166,
1892
+ "losses/total": 0.5145635604858398,
1893
+ "ref_logps/chosen": -27.76490020751953,
1894
+ "ref_logps/rejected": -34.16275405883789,
1895
+ "rewards/accuracies": 0.7109375,
1896
+ "rewards/chosen": -1.0974467992782593,
1897
+ "rewards/margins": 0.8847663402557373,
1898
+ "rewards/rejected": -1.9822131395339966,
1899
+ "step": 105
1900
+ },
1901
+ {
1902
+ "epoch": 0.8,
1903
+ "grad_norm": 4.85020637512207,
1904
+ "learning_rate": 3.3333333333333333e-06,
1905
+ "logps/chosen": -34.77644729614258,
1906
+ "logps/rejected": -51.18222427368164,
1907
+ "loss": 0.5472,
1908
+ "losses/dpo": 0.6919156908988953,
1909
+ "losses/sft": 1.51556396484375,
1910
+ "losses/total": 0.6919156908988953,
1911
+ "ref_logps/chosen": -24.479114532470703,
1912
+ "ref_logps/rejected": -33.03002166748047,
1913
+ "rewards/accuracies": 0.703125,
1914
+ "rewards/chosen": -1.0297331809997559,
1915
+ "rewards/margins": 0.785487174987793,
1916
+ "rewards/rejected": -1.8152204751968384,
1917
+ "step": 106
1918
+ },
1919
+ {
1920
+ "epoch": 0.8075471698113208,
1921
+ "grad_norm": 5.955616474151611,
1922
+ "learning_rate": 3.3122362869198316e-06,
1923
+ "logps/chosen": -43.87641143798828,
1924
+ "logps/rejected": -54.51424789428711,
1925
+ "loss": 0.6114,
1926
+ "losses/dpo": 0.756654679775238,
1927
+ "losses/sft": 1.6842637062072754,
1928
+ "losses/total": 0.756654679775238,
1929
+ "ref_logps/chosen": -31.69654083251953,
1930
+ "ref_logps/rejected": -36.67472839355469,
1931
+ "rewards/accuracies": 0.734375,
1932
+ "rewards/chosen": -1.217987060546875,
1933
+ "rewards/margins": 0.565964937210083,
1934
+ "rewards/rejected": -1.7839521169662476,
1935
+ "step": 107
1936
+ },
1937
+ {
1938
+ "epoch": 0.8150943396226416,
1939
+ "grad_norm": 5.920444965362549,
1940
+ "learning_rate": 3.2911392405063294e-06,
1941
+ "logps/chosen": -40.982208251953125,
1942
+ "logps/rejected": -52.83149337768555,
1943
+ "loss": 0.5517,
1944
+ "losses/dpo": 0.65492182970047,
1945
+ "losses/sft": 1.6355092525482178,
1946
+ "losses/total": 0.65492182970047,
1947
+ "ref_logps/chosen": -29.412181854248047,
1948
+ "ref_logps/rejected": -34.126094818115234,
1949
+ "rewards/accuracies": 0.6796875,
1950
+ "rewards/chosen": -1.157002329826355,
1951
+ "rewards/margins": 0.7135379910469055,
1952
+ "rewards/rejected": -1.8705402612686157,
1953
+ "step": 108
1954
+ },
1955
+ {
1956
+ "epoch": 0.8226415094339623,
1957
+ "grad_norm": 5.697230339050293,
1958
+ "learning_rate": 3.270042194092827e-06,
1959
+ "logps/chosen": -42.04249954223633,
1960
+ "logps/rejected": -56.36012268066406,
1961
+ "loss": 0.5426,
1962
+ "losses/dpo": 0.3828045725822449,
1963
+ "losses/sft": 1.5319883823394775,
1964
+ "losses/total": 0.3828045725822449,
1965
+ "ref_logps/chosen": -31.600698471069336,
1966
+ "ref_logps/rejected": -36.47235107421875,
1967
+ "rewards/accuracies": 0.7265625,
1968
+ "rewards/chosen": -1.0441800355911255,
1969
+ "rewards/margins": 0.9445971846580505,
1970
+ "rewards/rejected": -1.9887771606445312,
1971
+ "step": 109
1972
+ },
1973
+ {
1974
+ "epoch": 0.8301886792452831,
1975
+ "grad_norm": 6.10832405090332,
1976
+ "learning_rate": 3.248945147679325e-06,
1977
+ "logps/chosen": -39.39381408691406,
1978
+ "logps/rejected": -50.33586120605469,
1979
+ "loss": 0.6518,
1980
+ "losses/dpo": 0.5053750872612,
1981
+ "losses/sft": 1.3503327369689941,
1982
+ "losses/total": 0.5053750872612,
1983
+ "ref_logps/chosen": -27.708271026611328,
1984
+ "ref_logps/rejected": -33.47583770751953,
1985
+ "rewards/accuracies": 0.640625,
1986
+ "rewards/chosen": -1.1685543060302734,
1987
+ "rewards/margins": 0.5174481272697449,
1988
+ "rewards/rejected": -1.6860023736953735,
1989
+ "step": 110
1990
+ },
1991
+ {
1992
+ "epoch": 0.8377358490566038,
1993
+ "grad_norm": 5.662174224853516,
1994
+ "learning_rate": 3.2278481012658232e-06,
1995
+ "logps/chosen": -41.917667388916016,
1996
+ "logps/rejected": -55.30170440673828,
1997
+ "loss": 0.5537,
1998
+ "losses/dpo": 0.9183385372161865,
1999
+ "losses/sft": 1.6308963298797607,
2000
+ "losses/total": 0.9183385372161865,
2001
+ "ref_logps/chosen": -29.052471160888672,
2002
+ "ref_logps/rejected": -34.99309539794922,
2003
+ "rewards/accuracies": 0.7109375,
2004
+ "rewards/chosen": -1.2865197658538818,
2005
+ "rewards/margins": 0.7443405985832214,
2006
+ "rewards/rejected": -2.030860424041748,
2007
+ "step": 111
2008
+ },
2009
+ {
2010
+ "epoch": 0.8452830188679246,
2011
+ "grad_norm": 5.386070728302002,
2012
+ "learning_rate": 3.206751054852321e-06,
2013
+ "logps/chosen": -43.091156005859375,
2014
+ "logps/rejected": -57.210243225097656,
2015
+ "loss": 0.5028,
2016
+ "losses/dpo": 0.3924194872379303,
2017
+ "losses/sft": 1.2852116823196411,
2018
+ "losses/total": 0.3924194872379303,
2019
+ "ref_logps/chosen": -30.833717346191406,
2020
+ "ref_logps/rejected": -36.638153076171875,
2021
+ "rewards/accuracies": 0.78125,
2022
+ "rewards/chosen": -1.2257441282272339,
2023
+ "rewards/margins": 0.8314655423164368,
2024
+ "rewards/rejected": -2.0572097301483154,
2025
+ "step": 112
2026
+ },
2027
+ {
2028
+ "epoch": 0.8528301886792453,
2029
+ "grad_norm": 6.078062057495117,
2030
+ "learning_rate": 3.185654008438819e-06,
2031
+ "logps/chosen": -44.7150764465332,
2032
+ "logps/rejected": -60.601261138916016,
2033
+ "loss": 0.5275,
2034
+ "losses/dpo": 0.3586404323577881,
2035
+ "losses/sft": 1.5925896167755127,
2036
+ "losses/total": 0.3586404323577881,
2037
+ "ref_logps/chosen": -31.377941131591797,
2038
+ "ref_logps/rejected": -38.768470764160156,
2039
+ "rewards/accuracies": 0.7265625,
2040
+ "rewards/chosen": -1.3337135314941406,
2041
+ "rewards/margins": 0.8495657444000244,
2042
+ "rewards/rejected": -2.183279037475586,
2043
+ "step": 113
2044
+ },
2045
+ {
2046
+ "epoch": 0.8603773584905661,
2047
+ "grad_norm": 5.526185035705566,
2048
+ "learning_rate": 3.164556962025317e-06,
2049
+ "logps/chosen": -42.03477478027344,
2050
+ "logps/rejected": -54.82069778442383,
2051
+ "loss": 0.5588,
2052
+ "losses/dpo": 1.1230486631393433,
2053
+ "losses/sft": 1.6963893175125122,
2054
+ "losses/total": 1.1230486631393433,
2055
+ "ref_logps/chosen": -28.538129806518555,
2056
+ "ref_logps/rejected": -33.739173889160156,
2057
+ "rewards/accuracies": 0.71875,
2058
+ "rewards/chosen": -1.3496648073196411,
2059
+ "rewards/margins": 0.75848788022995,
2060
+ "rewards/rejected": -2.1081528663635254,
2061
+ "step": 114
2062
+ },
2063
+ {
2064
+ "epoch": 0.8679245283018868,
2065
+ "grad_norm": 5.497169017791748,
2066
+ "learning_rate": 3.1434599156118145e-06,
2067
+ "logps/chosen": -42.89957046508789,
2068
+ "logps/rejected": -57.447914123535156,
2069
+ "loss": 0.523,
2070
+ "losses/dpo": 0.7913863658905029,
2071
+ "losses/sft": 1.356689453125,
2072
+ "losses/total": 0.7913863658905029,
2073
+ "ref_logps/chosen": -30.25906753540039,
2074
+ "ref_logps/rejected": -36.00640869140625,
2075
+ "rewards/accuracies": 0.78125,
2076
+ "rewards/chosen": -1.2640503644943237,
2077
+ "rewards/margins": 0.8801001906394958,
2078
+ "rewards/rejected": -2.144150733947754,
2079
+ "step": 115
2080
+ },
2081
+ {
2082
+ "epoch": 0.8754716981132076,
2083
+ "grad_norm": 5.6127824783325195,
2084
+ "learning_rate": 3.1223628691983127e-06,
2085
+ "logps/chosen": -46.81511688232422,
2086
+ "logps/rejected": -56.43955993652344,
2087
+ "loss": 0.5401,
2088
+ "losses/dpo": 0.7304984927177429,
2089
+ "losses/sft": 1.7529406547546387,
2090
+ "losses/total": 0.7304984927177429,
2091
+ "ref_logps/chosen": -32.37680435180664,
2092
+ "ref_logps/rejected": -34.742347717285156,
2093
+ "rewards/accuracies": 0.7421875,
2094
+ "rewards/chosen": -1.4438307285308838,
2095
+ "rewards/margins": 0.7258903980255127,
2096
+ "rewards/rejected": -2.1697211265563965,
2097
+ "step": 116
2098
+ },
2099
+ {
2100
+ "epoch": 0.8830188679245283,
2101
+ "grad_norm": 6.019126892089844,
2102
+ "learning_rate": 3.10126582278481e-06,
2103
+ "logps/chosen": -44.45075988769531,
2104
+ "logps/rejected": -56.06095886230469,
2105
+ "loss": 0.5493,
2106
+ "losses/dpo": 0.5076150298118591,
2107
+ "losses/sft": 1.855008602142334,
2108
+ "losses/total": 0.5076150298118591,
2109
+ "ref_logps/chosen": -29.896549224853516,
2110
+ "ref_logps/rejected": -33.72525405883789,
2111
+ "rewards/accuracies": 0.65625,
2112
+ "rewards/chosen": -1.4554212093353271,
2113
+ "rewards/margins": 0.7781496644020081,
2114
+ "rewards/rejected": -2.2335708141326904,
2115
+ "step": 117
2116
+ },
2117
+ {
2118
+ "epoch": 0.8905660377358491,
2119
+ "grad_norm": 4.897617816925049,
2120
+ "learning_rate": 3.0801687763713083e-06,
2121
+ "logps/chosen": -44.17365264892578,
2122
+ "logps/rejected": -59.890846252441406,
2123
+ "loss": 0.4334,
2124
+ "losses/dpo": 0.41634368896484375,
2125
+ "losses/sft": 1.3092498779296875,
2126
+ "losses/total": 0.41634368896484375,
2127
+ "ref_logps/chosen": -30.777435302734375,
2128
+ "ref_logps/rejected": -36.164852142333984,
2129
+ "rewards/accuracies": 0.859375,
2130
+ "rewards/chosen": -1.3396217823028564,
2131
+ "rewards/margins": 1.0329774618148804,
2132
+ "rewards/rejected": -2.3725991249084473,
2133
+ "step": 118
2134
+ },
2135
+ {
2136
+ "epoch": 0.8981132075471698,
2137
+ "grad_norm": 5.326719760894775,
2138
+ "learning_rate": 3.059071729957806e-06,
2139
+ "logps/chosen": -41.951332092285156,
2140
+ "logps/rejected": -55.92094039916992,
2141
+ "loss": 0.4896,
2142
+ "losses/dpo": 0.36023202538490295,
2143
+ "losses/sft": 1.5721536874771118,
2144
+ "losses/total": 0.36023202538490295,
2145
+ "ref_logps/chosen": -29.51433563232422,
2146
+ "ref_logps/rejected": -34.74650573730469,
2147
+ "rewards/accuracies": 0.796875,
2148
+ "rewards/chosen": -1.2436996698379517,
2149
+ "rewards/margins": 0.8737441301345825,
2150
+ "rewards/rejected": -2.1174440383911133,
2151
+ "step": 119
2152
+ },
2153
+ {
2154
+ "epoch": 0.9056603773584906,
2155
+ "grad_norm": 5.39622688293457,
2156
+ "learning_rate": 3.037974683544304e-06,
2157
+ "logps/chosen": -42.2480354309082,
2158
+ "logps/rejected": -58.74956512451172,
2159
+ "loss": 0.51,
2160
+ "losses/dpo": 0.6462827920913696,
2161
+ "losses/sft": 1.5755528211593628,
2162
+ "losses/total": 0.6462827920913696,
2163
+ "ref_logps/chosen": -28.274633407592773,
2164
+ "ref_logps/rejected": -35.56834411621094,
2165
+ "rewards/accuracies": 0.7421875,
2166
+ "rewards/chosen": -1.397340178489685,
2167
+ "rewards/margins": 0.9207824468612671,
2168
+ "rewards/rejected": -2.318122625350952,
2169
+ "step": 120
2170
+ }
2171
+ ],
2172
+ "logging_steps": 1.0,
2173
+ "max_steps": 264,
2174
+ "num_input_tokens_seen": 0,
2175
+ "num_train_epochs": 2,
2176
+ "save_steps": 40,
2177
+ "total_flos": 0.0,
2178
+ "train_batch_size": 2,
2179
+ "trial_name": null,
2180
+ "trial_params": null
2181
+ }
ppllava_vicuna7b_image_video_dpo/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e5bcbbe0bee03d46b03751b272dfb59ac4be134ad0b8658d4151abc7a66c48e7
3
+ size 6776
{checkpoint-160 β†’ ppllava_vicuna7b_image_video_multiimage}/config.json RENAMED
@@ -5,7 +5,7 @@
5
  "btadapter": true,
6
  "btadapter_depth": 4,
7
  "clip_post_pretrain": null,
8
- "clip_weight": "/group/40043/ruyangliu/weight/CLIP-L334",
9
  "frame_shape": [
10
  24,
11
  24
 
5
  "btadapter": true,
6
  "btadapter_depth": 4,
7
  "clip_post_pretrain": null,
8
+ "clip_weight": "openai/clip-vit-large-patch14-336",
9
  "frame_shape": [
10
  24,
11
  24
ppllava_vicuna7b_image_video_multiimage/generation_config.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "_from_model_config": true,
3
+ "bos_token_id": 1,
4
+ "eos_token_id": 2,
5
+ "pad_token_id": 0,
6
+ "transformers_version": "4.40.0"
7
+ }
ppllava_vicuna7b_image_video_multiimage/model-00001-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d8c057eac2df8c4c8cfadb3144ef2e025e407a296c82cf2ed7c56d3a3db0dc6b
3
+ size 4980641554
ppllava_vicuna7b_image_video_multiimage/model-00002-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:43c52ddb0a40ff8df8793124c2d9fa09624616ab1fc536d7ad30a596419531b3
3
+ size 4957878536
ppllava_vicuna7b_image_video_multiimage/model-00003-of-00003.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c37654ae18b3fb67c178d41c678d47be54f220c9c1cf226a3fe70f73e2b3642e
3
+ size 4580905400
ppllava_vicuna7b_image_video_multiimage/model.safetensors.index.json ADDED
The diff for this file is too large to render. See raw diff
 
ppllava_vicuna7b_image_video_multiimage/trainer_state.json ADDED
@@ -0,0 +1,821 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 0.9999115122555526,
5
+ "eval_steps": 500,
6
+ "global_step": 5650,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.008848774444739404,
13
+ "grad_norm": 1.8227072749903463,
14
+ "learning_rate": 5.882352941176471e-06,
15
+ "loss": 1.434,
16
+ "step": 50
17
+ },
18
+ {
19
+ "epoch": 0.017697548889478807,
20
+ "grad_norm": 1.593675125457607,
21
+ "learning_rate": 1.1764705882352942e-05,
22
+ "loss": 1.1202,
23
+ "step": 100
24
+ },
25
+ {
26
+ "epoch": 0.026546323334218212,
27
+ "grad_norm": 1.5547518309505592,
28
+ "learning_rate": 1.7647058823529414e-05,
29
+ "loss": 1.0971,
30
+ "step": 150
31
+ },
32
+ {
33
+ "epoch": 0.035395097778957614,
34
+ "grad_norm": 1.3952753982651918,
35
+ "learning_rate": 1.9998521094455198e-05,
36
+ "loss": 1.0463,
37
+ "step": 200
38
+ },
39
+ {
40
+ "epoch": 0.044243872223697016,
41
+ "grad_norm": 1.6576864383381864,
42
+ "learning_rate": 1.9989484922416503e-05,
43
+ "loss": 1.0808,
44
+ "step": 250
45
+ },
46
+ {
47
+ "epoch": 0.053092646668436425,
48
+ "grad_norm": 1.340907083636223,
49
+ "learning_rate": 1.9972241607451552e-05,
50
+ "loss": 1.0342,
51
+ "step": 300
52
+ },
53
+ {
54
+ "epoch": 0.061941421113175826,
55
+ "grad_norm": 1.269578433993962,
56
+ "learning_rate": 1.9946805316291817e-05,
57
+ "loss": 1.0199,
58
+ "step": 350
59
+ },
60
+ {
61
+ "epoch": 0.07079019555791523,
62
+ "grad_norm": 1.2537035911999723,
63
+ "learning_rate": 1.9913196946839304e-05,
64
+ "loss": 1.0137,
65
+ "step": 400
66
+ },
67
+ {
68
+ "epoch": 0.07963897000265463,
69
+ "grad_norm": 1.1857124728289088,
70
+ "learning_rate": 1.987144411099731e-05,
71
+ "loss": 1.0133,
72
+ "step": 450
73
+ },
74
+ {
75
+ "epoch": 0.08848774444739403,
76
+ "grad_norm": 1.2181869575632758,
77
+ "learning_rate": 1.9821581111985072e-05,
78
+ "loss": 1.0178,
79
+ "step": 500
80
+ },
81
+ {
82
+ "epoch": 0.09733651889213343,
83
+ "grad_norm": 1.2987295471871965,
84
+ "learning_rate": 1.9763648916154982e-05,
85
+ "loss": 1.0127,
86
+ "step": 550
87
+ },
88
+ {
89
+ "epoch": 0.10618529333687285,
90
+ "grad_norm": 1.2413868753158877,
91
+ "learning_rate": 1.9697695119335547e-05,
92
+ "loss": 0.9979,
93
+ "step": 600
94
+ },
95
+ {
96
+ "epoch": 0.11503406778161225,
97
+ "grad_norm": 1.2626131743513744,
98
+ "learning_rate": 1.9623773907727682e-05,
99
+ "loss": 0.9965,
100
+ "step": 650
101
+ },
102
+ {
103
+ "epoch": 0.12388284222635165,
104
+ "grad_norm": 1.3730231292537942,
105
+ "learning_rate": 1.954194601338651e-05,
106
+ "loss": 0.9942,
107
+ "step": 700
108
+ },
109
+ {
110
+ "epoch": 0.13273161667109104,
111
+ "grad_norm": 1.2218007272454348,
112
+ "learning_rate": 1.9452278664325227e-05,
113
+ "loss": 1.0036,
114
+ "step": 750
115
+ },
116
+ {
117
+ "epoch": 0.14158039111583046,
118
+ "grad_norm": 1.1454037410098823,
119
+ "learning_rate": 1.9354845529282042e-05,
120
+ "loss": 0.9868,
121
+ "step": 800
122
+ },
123
+ {
124
+ "epoch": 0.15042916556056987,
125
+ "grad_norm": 1.199534918146064,
126
+ "learning_rate": 1.9249726657195534e-05,
127
+ "loss": 0.9972,
128
+ "step": 850
129
+ },
130
+ {
131
+ "epoch": 0.15927794000530926,
132
+ "grad_norm": 1.2920998134175072,
133
+ "learning_rate": 1.9137008411438213e-05,
134
+ "loss": 1.0239,
135
+ "step": 900
136
+ },
137
+ {
138
+ "epoch": 0.16812671445004868,
139
+ "grad_norm": 1.1321328064281995,
140
+ "learning_rate": 1.901678339886223e-05,
141
+ "loss": 0.9807,
142
+ "step": 950
143
+ },
144
+ {
145
+ "epoch": 0.17697548889478806,
146
+ "grad_norm": 1.1146456739633037,
147
+ "learning_rate": 1.8889150393715627e-05,
148
+ "loss": 0.981,
149
+ "step": 1000
150
+ },
151
+ {
152
+ "epoch": 0.18582426333952748,
153
+ "grad_norm": 1.161097648736237,
154
+ "learning_rate": 1.8754214256491564e-05,
155
+ "loss": 0.9826,
156
+ "step": 1050
157
+ },
158
+ {
159
+ "epoch": 0.19467303778426687,
160
+ "grad_norm": 1.2010813609189326,
161
+ "learning_rate": 1.8612085847777215e-05,
162
+ "loss": 0.9846,
163
+ "step": 1100
164
+ },
165
+ {
166
+ "epoch": 0.20352181222900628,
167
+ "grad_norm": 1.203856802982565,
168
+ "learning_rate": 1.8462881937173144e-05,
169
+ "loss": 0.9789,
170
+ "step": 1150
171
+ },
172
+ {
173
+ "epoch": 0.2123705866737457,
174
+ "grad_norm": 1.1809801509975393,
175
+ "learning_rate": 1.8306725107357933e-05,
176
+ "loss": 0.9785,
177
+ "step": 1200
178
+ },
179
+ {
180
+ "epoch": 0.22121936111848509,
181
+ "grad_norm": 1.1856255544481202,
182
+ "learning_rate": 1.8143743653376944e-05,
183
+ "loss": 0.9724,
184
+ "step": 1250
185
+ },
186
+ {
187
+ "epoch": 0.2300681355632245,
188
+ "grad_norm": 1.2932019902094527,
189
+ "learning_rate": 1.7974071477237887e-05,
190
+ "loss": 0.9741,
191
+ "step": 1300
192
+ },
193
+ {
194
+ "epoch": 0.2389169100079639,
195
+ "grad_norm": 1.1399596376970142,
196
+ "learning_rate": 1.7797847977899873e-05,
197
+ "loss": 0.9787,
198
+ "step": 1350
199
+ },
200
+ {
201
+ "epoch": 0.2477656844527033,
202
+ "grad_norm": 1.1851681853908578,
203
+ "learning_rate": 1.7615217936746246e-05,
204
+ "loss": 0.9712,
205
+ "step": 1400
206
+ },
207
+ {
208
+ "epoch": 0.2566144588974427,
209
+ "grad_norm": 1.212090367995841,
210
+ "learning_rate": 1.742633139863538e-05,
211
+ "loss": 0.9729,
212
+ "step": 1450
213
+ },
214
+ {
215
+ "epoch": 0.2654632333421821,
216
+ "grad_norm": 1.0975454688592081,
217
+ "learning_rate": 1.7231343548627085e-05,
218
+ "loss": 0.9714,
219
+ "step": 1500
220
+ },
221
+ {
222
+ "epoch": 0.2743120077869215,
223
+ "grad_norm": 1.0110033370546834,
224
+ "learning_rate": 1.7030414584485938e-05,
225
+ "loss": 0.9591,
226
+ "step": 1550
227
+ },
228
+ {
229
+ "epoch": 0.2831607822316609,
230
+ "grad_norm": 1.0352711739713445,
231
+ "learning_rate": 1.6823709585066308e-05,
232
+ "loss": 0.9719,
233
+ "step": 1600
234
+ },
235
+ {
236
+ "epoch": 0.29200955667640033,
237
+ "grad_norm": 1.1174206790465606,
238
+ "learning_rate": 1.6611398374687172e-05,
239
+ "loss": 0.9673,
240
+ "step": 1650
241
+ },
242
+ {
243
+ "epoch": 0.30085833112113974,
244
+ "grad_norm": 1.1508488673423878,
245
+ "learning_rate": 1.6393655383608132e-05,
246
+ "loss": 0.9579,
247
+ "step": 1700
248
+ },
249
+ {
250
+ "epoch": 0.3097071055658791,
251
+ "grad_norm": 1.1140112909261894,
252
+ "learning_rate": 1.6170659504721365e-05,
253
+ "loss": 0.9773,
254
+ "step": 1750
255
+ },
256
+ {
257
+ "epoch": 0.3185558800106185,
258
+ "grad_norm": 1.078883305222083,
259
+ "learning_rate": 1.594259394657707e-05,
260
+ "loss": 0.963,
261
+ "step": 1800
262
+ },
263
+ {
264
+ "epoch": 0.32740465445535794,
265
+ "grad_norm": 1.0741496670790676,
266
+ "learning_rate": 1.570964608286336e-05,
267
+ "loss": 0.9665,
268
+ "step": 1850
269
+ },
270
+ {
271
+ "epoch": 0.33625342890009735,
272
+ "grad_norm": 1.0674741658785543,
273
+ "learning_rate": 1.5472007298464117e-05,
274
+ "loss": 0.9577,
275
+ "step": 1900
276
+ },
277
+ {
278
+ "epoch": 0.34510220334483677,
279
+ "grad_norm": 1.1266524576573997,
280
+ "learning_rate": 1.5229872832221336e-05,
281
+ "loss": 0.9578,
282
+ "step": 1950
283
+ },
284
+ {
285
+ "epoch": 0.3539509777895761,
286
+ "grad_norm": 1.0507368907995636,
287
+ "learning_rate": 1.4983441616531152e-05,
288
+ "loss": 0.9543,
289
+ "step": 2000
290
+ },
291
+ {
292
+ "epoch": 0.36279975223431554,
293
+ "grad_norm": 1.5910538700413814,
294
+ "learning_rate": 1.4732916113905336e-05,
295
+ "loss": 0.9499,
296
+ "step": 2050
297
+ },
298
+ {
299
+ "epoch": 0.37164852667905496,
300
+ "grad_norm": 1.1710473762069435,
301
+ "learning_rate": 1.4478502150632503e-05,
302
+ "loss": 0.9928,
303
+ "step": 2100
304
+ },
305
+ {
306
+ "epoch": 0.3804973011237944,
307
+ "grad_norm": 1.1721776444324115,
308
+ "learning_rate": 1.4220408747675714e-05,
309
+ "loss": 0.9509,
310
+ "step": 2150
311
+ },
312
+ {
313
+ "epoch": 0.38934607556853373,
314
+ "grad_norm": 1.1265584958834658,
315
+ "learning_rate": 1.3958847948945428e-05,
316
+ "loss": 0.9437,
317
+ "step": 2200
318
+ },
319
+ {
320
+ "epoch": 0.39819485001327315,
321
+ "grad_norm": 1.1519073177115475,
322
+ "learning_rate": 1.369403464708884e-05,
323
+ "loss": 0.9445,
324
+ "step": 2250
325
+ },
326
+ {
327
+ "epoch": 0.40704362445801257,
328
+ "grad_norm": 1.1172434119258432,
329
+ "learning_rate": 1.3426186406938769e-05,
330
+ "loss": 1.0387,
331
+ "step": 2300
332
+ },
333
+ {
334
+ "epoch": 0.415892398902752,
335
+ "grad_norm": 1.138922531256483,
336
+ "learning_rate": 1.315552328676714e-05,
337
+ "loss": 0.9391,
338
+ "step": 2350
339
+ },
340
+ {
341
+ "epoch": 0.4247411733474914,
342
+ "grad_norm": 1.089137186693905,
343
+ "learning_rate": 1.2882267657489908e-05,
344
+ "loss": 0.9457,
345
+ "step": 2400
346
+ },
347
+ {
348
+ "epoch": 0.43358994779223076,
349
+ "grad_norm": 1.0358420925020666,
350
+ "learning_rate": 1.2606644019971967e-05,
351
+ "loss": 0.9972,
352
+ "step": 2450
353
+ },
354
+ {
355
+ "epoch": 0.44243872223697017,
356
+ "grad_norm": 1.0748089642780165,
357
+ "learning_rate": 1.2328878820582122e-05,
358
+ "loss": 0.926,
359
+ "step": 2500
360
+ },
361
+ {
362
+ "epoch": 0.4512874966817096,
363
+ "grad_norm": 1.1178495139589024,
364
+ "learning_rate": 1.204920026514971e-05,
365
+ "loss": 0.9371,
366
+ "step": 2550
367
+ },
368
+ {
369
+ "epoch": 0.460136271126449,
370
+ "grad_norm": 1.0570225052003097,
371
+ "learning_rate": 1.1767838131475654e-05,
372
+ "loss": 0.9299,
373
+ "step": 2600
374
+ },
375
+ {
376
+ "epoch": 0.46898504557118836,
377
+ "grad_norm": 1.198704612437538,
378
+ "learning_rate": 1.1485023580552039e-05,
379
+ "loss": 0.9333,
380
+ "step": 2650
381
+ },
382
+ {
383
+ "epoch": 0.4778338200159278,
384
+ "grad_norm": 1.2153247727284249,
385
+ "learning_rate": 1.1200988966645286e-05,
386
+ "loss": 0.9325,
387
+ "step": 2700
388
+ },
389
+ {
390
+ "epoch": 0.4866825944606672,
391
+ "grad_norm": 1.0862037277462553,
392
+ "learning_rate": 1.091596764639895e-05,
393
+ "loss": 0.9341,
394
+ "step": 2750
395
+ },
396
+ {
397
+ "epoch": 0.4955313689054066,
398
+ "grad_norm": 1.0724182576148855,
399
+ "learning_rate": 1.0630193787112994e-05,
400
+ "loss": 0.9063,
401
+ "step": 2800
402
+ },
403
+ {
404
+ "epoch": 0.504380143350146,
405
+ "grad_norm": 1.0396985853342051,
406
+ "learning_rate": 1.034390217435704e-05,
407
+ "loss": 0.9293,
408
+ "step": 2850
409
+ },
410
+ {
411
+ "epoch": 0.5132289177948854,
412
+ "grad_norm": 1.0749902902208996,
413
+ "learning_rate": 1.005732801907567e-05,
414
+ "loss": 0.9214,
415
+ "step": 2900
416
+ },
417
+ {
418
+ "epoch": 0.5220776922396249,
419
+ "grad_norm": 1.085293805471844,
420
+ "learning_rate": 9.770706764344235e-06,
421
+ "loss": 1.0245,
422
+ "step": 2950
423
+ },
424
+ {
425
+ "epoch": 0.5309264666843642,
426
+ "grad_norm": 1.1358464620386077,
427
+ "learning_rate": 9.484273891933982e-06,
428
+ "loss": 0.9297,
429
+ "step": 3000
430
+ },
431
+ {
432
+ "epoch": 0.5397752411291036,
433
+ "grad_norm": 1.102492622904414,
434
+ "learning_rate": 9.198264728845332e-06,
435
+ "loss": 0.9157,
436
+ "step": 3050
437
+ },
438
+ {
439
+ "epoch": 0.548624015573843,
440
+ "grad_norm": 1.1636978192620964,
441
+ "learning_rate": 8.912914253968391e-06,
442
+ "loss": 0.9236,
443
+ "step": 3100
444
+ },
445
+ {
446
+ "epoch": 0.5574727900185824,
447
+ "grad_norm": 1.1333308119371828,
448
+ "learning_rate": 8.628456905029383e-06,
449
+ "loss": 0.9158,
450
+ "step": 3150
451
+ },
452
+ {
453
+ "epoch": 0.5663215644633218,
454
+ "grad_norm": 1.117341944549429,
455
+ "learning_rate": 8.345126385981737e-06,
456
+ "loss": 0.9102,
457
+ "step": 3200
458
+ },
459
+ {
460
+ "epoch": 0.5751703389080612,
461
+ "grad_norm": 1.0840357862773122,
462
+ "learning_rate": 8.063155475000037e-06,
463
+ "loss": 0.9546,
464
+ "step": 3250
465
+ },
466
+ {
467
+ "epoch": 0.5840191133528007,
468
+ "grad_norm": 1.1791446775850642,
469
+ "learning_rate": 7.782775833234522e-06,
470
+ "loss": 0.924,
471
+ "step": 3300
472
+ },
473
+ {
474
+ "epoch": 0.5928678877975401,
475
+ "grad_norm": 1.1948088039686837,
476
+ "learning_rate": 7.504217814483364e-06,
477
+ "loss": 0.9135,
478
+ "step": 3350
479
+ },
480
+ {
481
+ "epoch": 0.6017166622422795,
482
+ "grad_norm": 1.107584300567853,
483
+ "learning_rate": 7.227710275938987e-06,
484
+ "loss": 0.9088,
485
+ "step": 3400
486
+ },
487
+ {
488
+ "epoch": 0.6105654366870189,
489
+ "grad_norm": 1.167878367521536,
490
+ "learning_rate": 6.953480390164001e-06,
491
+ "loss": 0.9394,
492
+ "step": 3450
493
+ },
494
+ {
495
+ "epoch": 0.6194142111317582,
496
+ "grad_norm": 1.0387264654842252,
497
+ "learning_rate": 6.68175345845119e-06,
498
+ "loss": 0.9022,
499
+ "step": 3500
500
+ },
501
+ {
502
+ "epoch": 0.6282629855764976,
503
+ "grad_norm": 1.2650051247263,
504
+ "learning_rate": 6.412752725720864e-06,
505
+ "loss": 0.9135,
506
+ "step": 3550
507
+ },
508
+ {
509
+ "epoch": 0.637111760021237,
510
+ "grad_norm": 1.20136444284978,
511
+ "learning_rate": 6.146699197107715e-06,
512
+ "loss": 0.9068,
513
+ "step": 3600
514
+ },
515
+ {
516
+ "epoch": 0.6459605344659765,
517
+ "grad_norm": 1.0811422393549932,
518
+ "learning_rate": 5.883811456387821e-06,
519
+ "loss": 0.9082,
520
+ "step": 3650
521
+ },
522
+ {
523
+ "epoch": 0.6548093089107159,
524
+ "grad_norm": 1.1656545210876348,
525
+ "learning_rate": 5.6243054863949675e-06,
526
+ "loss": 0.8898,
527
+ "step": 3700
528
+ },
529
+ {
530
+ "epoch": 0.6636580833554553,
531
+ "grad_norm": 1.1852994908957295,
532
+ "learning_rate": 5.368394491573876e-06,
533
+ "loss": 0.9026,
534
+ "step": 3750
535
+ },
536
+ {
537
+ "epoch": 0.6725068578001947,
538
+ "grad_norm": 1.161821140479422,
539
+ "learning_rate": 5.116288722816087e-06,
540
+ "loss": 0.8838,
541
+ "step": 3800
542
+ },
543
+ {
544
+ "epoch": 0.6813556322449341,
545
+ "grad_norm": 1.2037332656822164,
546
+ "learning_rate": 4.868195304722391e-06,
547
+ "loss": 0.9025,
548
+ "step": 3850
549
+ },
550
+ {
551
+ "epoch": 0.6902044066896735,
552
+ "grad_norm": 1.1242488366634837,
553
+ "learning_rate": 4.6243180654337975e-06,
554
+ "loss": 0.931,
555
+ "step": 3900
556
+ },
557
+ {
558
+ "epoch": 0.6990531811344128,
559
+ "grad_norm": 1.1696361772435924,
560
+ "learning_rate": 4.384857369170772e-06,
561
+ "loss": 0.9338,
562
+ "step": 3950
563
+ },
564
+ {
565
+ "epoch": 0.7079019555791523,
566
+ "grad_norm": 1.2264055771278481,
567
+ "learning_rate": 4.1500099516183555e-06,
568
+ "loss": 0.8993,
569
+ "step": 4000
570
+ },
571
+ {
572
+ "epoch": 0.7167507300238917,
573
+ "grad_norm": 1.1225371954007977,
574
+ "learning_rate": 3.919968758292425e-06,
575
+ "loss": 0.9044,
576
+ "step": 4050
577
+ },
578
+ {
579
+ "epoch": 0.7255995044686311,
580
+ "grad_norm": 1.1128045918827218,
581
+ "learning_rate": 3.6949227860198712e-06,
582
+ "loss": 0.8963,
583
+ "step": 4100
584
+ },
585
+ {
586
+ "epoch": 0.7344482789133705,
587
+ "grad_norm": 1.1426952334649678,
588
+ "learning_rate": 3.475056927662912e-06,
589
+ "loss": 0.8955,
590
+ "step": 4150
591
+ },
592
+ {
593
+ "epoch": 0.7432970533581099,
594
+ "grad_norm": 1.0853524038336615,
595
+ "learning_rate": 3.2605518202151577e-06,
596
+ "loss": 0.8973,
597
+ "step": 4200
598
+ },
599
+ {
600
+ "epoch": 0.7521458278028493,
601
+ "grad_norm": 1.1322485210453683,
602
+ "learning_rate": 3.0515836963942056e-06,
603
+ "loss": 0.8944,
604
+ "step": 4250
605
+ },
606
+ {
607
+ "epoch": 0.7609946022475887,
608
+ "grad_norm": 1.2182044740120312,
609
+ "learning_rate": 2.8483242398526723e-06,
610
+ "loss": 0.8872,
611
+ "step": 4300
612
+ },
613
+ {
614
+ "epoch": 0.7698433766923282,
615
+ "grad_norm": 1.0357920295682677,
616
+ "learning_rate": 2.650940444126654e-06,
617
+ "loss": 0.8856,
618
+ "step": 4350
619
+ },
620
+ {
621
+ "epoch": 0.7786921511370675,
622
+ "grad_norm": 0.984081860446035,
623
+ "learning_rate": 2.4595944754374723e-06,
624
+ "loss": 0.8818,
625
+ "step": 4400
626
+ },
627
+ {
628
+ "epoch": 0.7875409255818069,
629
+ "grad_norm": 1.1062257149827126,
630
+ "learning_rate": 2.27444353945945e-06,
631
+ "loss": 0.8883,
632
+ "step": 4450
633
+ },
634
+ {
635
+ "epoch": 0.7963897000265463,
636
+ "grad_norm": 1.1408547470520316,
637
+ "learning_rate": 2.0956397521631666e-06,
638
+ "loss": 0.8729,
639
+ "step": 4500
640
+ },
641
+ {
642
+ "epoch": 0.8052384744712857,
643
+ "grad_norm": 1.2286350695548351,
644
+ "learning_rate": 1.9233300148402767e-06,
645
+ "loss": 0.8782,
646
+ "step": 4550
647
+ },
648
+ {
649
+ "epoch": 0.8140872489160251,
650
+ "grad_norm": 1.1745122607432803,
651
+ "learning_rate": 1.757655893412622e-06,
652
+ "loss": 0.8763,
653
+ "step": 4600
654
+ },
655
+ {
656
+ "epoch": 0.8229360233607645,
657
+ "grad_norm": 1.291525874104284,
658
+ "learning_rate": 1.5987535021247668e-06,
659
+ "loss": 0.8817,
660
+ "step": 4650
661
+ },
662
+ {
663
+ "epoch": 0.831784797805504,
664
+ "grad_norm": 1.1484560802799162,
665
+ "learning_rate": 1.4467533917154842e-06,
666
+ "loss": 0.8914,
667
+ "step": 4700
668
+ },
669
+ {
670
+ "epoch": 0.8406335722502434,
671
+ "grad_norm": 1.0916994218547142,
672
+ "learning_rate": 1.3017804421601298e-06,
673
+ "loss": 0.9154,
674
+ "step": 4750
675
+ },
676
+ {
677
+ "epoch": 0.8494823466949828,
678
+ "grad_norm": 1.2124618779904544,
679
+ "learning_rate": 1.1639537600719764e-06,
680
+ "loss": 0.8821,
681
+ "step": 4800
682
+ },
683
+ {
684
+ "epoch": 0.8583311211397221,
685
+ "grad_norm": 1.1869802843495634,
686
+ "learning_rate": 1.0333865808468203e-06,
687
+ "loss": 0.8821,
688
+ "step": 4850
689
+ },
690
+ {
691
+ "epoch": 0.8671798955844615,
692
+ "grad_norm": 1.134603833533901,
693
+ "learning_rate": 9.101861756312369e-07,
694
+ "loss": 0.8799,
695
+ "step": 4900
696
+ },
697
+ {
698
+ "epoch": 0.8760286700292009,
699
+ "grad_norm": 1.23950870941106,
700
+ "learning_rate": 7.944537631909666e-07,
701
+ "loss": 0.8874,
702
+ "step": 4950
703
+ },
704
+ {
705
+ "epoch": 0.8848774444739403,
706
+ "grad_norm": 1.0710206299690663,
707
+ "learning_rate": 6.862844267517643e-07,
708
+ "loss": 0.9178,
709
+ "step": 5000
710
+ },
711
+ {
712
+ "epoch": 0.8937262189186798,
713
+ "grad_norm": 1.1079780487199702,
714
+ "learning_rate": 5.857670358811096e-07,
715
+ "loss": 0.9139,
716
+ "step": 5050
717
+ },
718
+ {
719
+ "epoch": 0.9025749933634192,
720
+ "grad_norm": 1.224157319904236,
721
+ "learning_rate": 4.929841734749063e-07,
722
+ "loss": 0.883,
723
+ "step": 5100
724
+ },
725
+ {
726
+ "epoch": 0.9114237678081586,
727
+ "grad_norm": 1.2973098944898664,
728
+ "learning_rate": 4.0801206790916815e-07,
729
+ "loss": 0.8748,
730
+ "step": 5150
731
+ },
732
+ {
733
+ "epoch": 0.920272542252898,
734
+ "grad_norm": 1.2788191274563776,
735
+ "learning_rate": 3.309205304124552e-07,
736
+ "loss": 0.9109,
737
+ "step": 5200
738
+ },
739
+ {
740
+ "epoch": 0.9291213166976374,
741
+ "grad_norm": 1.1179855803922911,
742
+ "learning_rate": 2.6177289771049274e-07,
743
+ "loss": 0.8944,
744
+ "step": 5250
745
+ },
746
+ {
747
+ "epoch": 0.9379700911423767,
748
+ "grad_norm": 1.1595081612750888,
749
+ "learning_rate": 2.0062597999009114e-07,
750
+ "loss": 0.8857,
751
+ "step": 5300
752
+ },
753
+ {
754
+ "epoch": 0.9468188655871161,
755
+ "grad_norm": 1.0486044192293602,
756
+ "learning_rate": 1.4753001422514125e-07,
757
+ "loss": 0.8827,
758
+ "step": 5350
759
+ },
760
+ {
761
+ "epoch": 0.9556676400318556,
762
+ "grad_norm": 1.1477897530038004,
763
+ "learning_rate": 1.0252862290301092e-07,
764
+ "loss": 0.8769,
765
+ "step": 5400
766
+ },
767
+ {
768
+ "epoch": 0.964516414476595,
769
+ "grad_norm": 1.1863014296064434,
770
+ "learning_rate": 6.565877818526245e-08,
771
+ "loss": 0.8754,
772
+ "step": 5450
773
+ },
774
+ {
775
+ "epoch": 0.9733651889213344,
776
+ "grad_norm": 1.2034312267234064,
777
+ "learning_rate": 3.6950771532126004e-08,
778
+ "loss": 0.8723,
779
+ "step": 5500
780
+ },
781
+ {
782
+ "epoch": 0.9822139633660738,
783
+ "grad_norm": 1.1269276845827294,
784
+ "learning_rate": 1.6428188815703627e-08,
785
+ "loss": 0.9178,
786
+ "step": 5550
787
+ },
788
+ {
789
+ "epoch": 0.9910627378108132,
790
+ "grad_norm": 1.2360290358413641,
791
+ "learning_rate": 4.1078909423253325e-09,
792
+ "loss": 0.8848,
793
+ "step": 5600
794
+ },
795
+ {
796
+ "epoch": 0.9999115122555526,
797
+ "grad_norm": 1.271675294761002,
798
+ "learning_rate": 0.0,
799
+ "loss": 0.8882,
800
+ "step": 5650
801
+ },
802
+ {
803
+ "epoch": 0.9999115122555526,
804
+ "step": 5650,
805
+ "total_flos": 3.3418410989715456e+16,
806
+ "train_loss": 0.94595458984375,
807
+ "train_runtime": 88600.3448,
808
+ "train_samples_per_second": 0.51,
809
+ "train_steps_per_second": 0.064
810
+ }
811
+ ],
812
+ "logging_steps": 50,
813
+ "max_steps": 5650,
814
+ "num_input_tokens_seen": 0,
815
+ "num_train_epochs": 1,
816
+ "save_steps": 500,
817
+ "total_flos": 3.3418410989715456e+16,
818
+ "train_batch_size": 4,
819
+ "trial_name": null,
820
+ "trial_params": null
821
+ }
ppllava_vicuna7b_image_video_multiimage/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6c0ae664795f9653af20676a1ee3c7d002c5f4756dea1b52099a6d533323fc48
3
+ size 7096