Qin Liu commited on
Commit
8c31ff2
1 Parent(s): 0518b2b

Model save

Browse files
Files changed (4) hide show
  1. README.md +62 -0
  2. all_results.json +9 -0
  3. train_results.json +9 -0
  4. trainer_state.json +432 -0
README.md ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: llama3
3
+ library_name: peft
4
+ tags:
5
+ - trl
6
+ - dpo
7
+ - generated_from_trainer
8
+ base_model: Jackie999/llama3-sudo-5epochs-tofu_full_sft
9
+ model-index:
10
+ - name: llama3-sudo-dpo-instruct-10epochs-forget10-1sft-2lora
11
+ results: []
12
+ ---
13
+
14
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
+ should probably proofread and complete it, then remove this comment. -->
16
+
17
+ # llama3-sudo-dpo-instruct-10epochs-forget10-1sft-2lora
18
+
19
+ This model is a fine-tuned version of [Jackie999/llama3-sudo-5epochs-tofu_full_sft](https://huggingface.co/Jackie999/llama3-sudo-5epochs-tofu_full_sft) on an unknown dataset.
20
+
21
+ ## Model description
22
+
23
+ More information needed
24
+
25
+ ## Intended uses & limitations
26
+
27
+ More information needed
28
+
29
+ ## Training and evaluation data
30
+
31
+ More information needed
32
+
33
+ ## Training procedure
34
+
35
+ ### Training hyperparameters
36
+
37
+ The following hyperparameters were used during training:
38
+ - learning_rate: 5e-06
39
+ - train_batch_size: 2
40
+ - eval_batch_size: 8
41
+ - seed: 42
42
+ - distributed_type: multi-GPU
43
+ - num_devices: 4
44
+ - gradient_accumulation_steps: 4
45
+ - total_train_batch_size: 32
46
+ - total_eval_batch_size: 32
47
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
48
+ - lr_scheduler_type: cosine
49
+ - lr_scheduler_warmup_ratio: 0.1
50
+ - num_epochs: 10
51
+
52
+ ### Training results
53
+
54
+
55
+
56
+ ### Framework versions
57
+
58
+ - PEFT 0.12.0
59
+ - Transformers 4.44.0
60
+ - Pytorch 2.1.2
61
+ - Datasets 3.0.0
62
+ - Tokenizers 0.19.1
all_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 10.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.18631765481829643,
5
+ "train_runtime": 3826.8955,
6
+ "train_samples": 800,
7
+ "train_samples_per_second": 2.09,
8
+ "train_steps_per_second": 0.065
9
+ }
train_results.json ADDED
@@ -0,0 +1,9 @@
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 10.0,
3
+ "total_flos": 0.0,
4
+ "train_loss": 0.18631765481829643,
5
+ "train_runtime": 3826.8955,
6
+ "train_samples": 800,
7
+ "train_samples_per_second": 2.09,
8
+ "train_steps_per_second": 0.065
9
+ }
trainer_state.json ADDED
@@ -0,0 +1,432 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 10.0,
5
+ "eval_steps": 1000,
6
+ "global_step": 250,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.04,
13
+ "grad_norm": 0.2610610014559749,
14
+ "learning_rate": 2.0000000000000002e-07,
15
+ "logits/chosen": -1.7501684427261353,
16
+ "logits/rejected": -1.5946102142333984,
17
+ "logps/chosen": -84.6928939819336,
18
+ "logps/rejected": -66.21642303466797,
19
+ "loss": 0.6931,
20
+ "rewards/accuracies": 0.0,
21
+ "rewards/chosen": 0.0,
22
+ "rewards/margins": 0.0,
23
+ "rewards/rejected": 0.0,
24
+ "step": 1
25
+ },
26
+ {
27
+ "epoch": 0.4,
28
+ "grad_norm": 0.22905427192235048,
29
+ "learning_rate": 2.0000000000000003e-06,
30
+ "logits/chosen": -1.743150234222412,
31
+ "logits/rejected": -1.7186625003814697,
32
+ "logps/chosen": -75.15956115722656,
33
+ "logps/rejected": -78.51848602294922,
34
+ "loss": 0.6929,
35
+ "rewards/accuracies": 0.5694444179534912,
36
+ "rewards/chosen": 0.0005810303264297545,
37
+ "rewards/margins": 0.0006510103703476489,
38
+ "rewards/rejected": -6.998008029768243e-05,
39
+ "step": 10
40
+ },
41
+ {
42
+ "epoch": 0.8,
43
+ "grad_norm": 0.22900106097369388,
44
+ "learning_rate": 4.000000000000001e-06,
45
+ "logits/chosen": -1.7664934396743774,
46
+ "logits/rejected": -1.697330117225647,
47
+ "logps/chosen": -77.00563049316406,
48
+ "logps/rejected": -74.59007263183594,
49
+ "loss": 0.6898,
50
+ "rewards/accuracies": 0.9624999761581421,
51
+ "rewards/chosen": 0.004773316439241171,
52
+ "rewards/margins": 0.006489777471870184,
53
+ "rewards/rejected": -0.001716460450552404,
54
+ "step": 20
55
+ },
56
+ {
57
+ "epoch": 1.2,
58
+ "grad_norm": 0.2398363937620789,
59
+ "learning_rate": 4.993910125649561e-06,
60
+ "logits/chosen": -1.670393705368042,
61
+ "logits/rejected": -1.7334346771240234,
62
+ "logps/chosen": -70.95411682128906,
63
+ "logps/rejected": -81.66401672363281,
64
+ "loss": 0.6758,
65
+ "rewards/accuracies": 1.0,
66
+ "rewards/chosen": 0.021032074466347694,
67
+ "rewards/margins": 0.03419991582632065,
68
+ "rewards/rejected": -0.013167845085263252,
69
+ "step": 30
70
+ },
71
+ {
72
+ "epoch": 1.6,
73
+ "grad_norm": 0.4134736462223887,
74
+ "learning_rate": 4.9453690018345144e-06,
75
+ "logits/chosen": -1.7007486820220947,
76
+ "logits/rejected": -1.7079846858978271,
77
+ "logps/chosen": -68.65242004394531,
78
+ "logps/rejected": -83.7640609741211,
79
+ "loss": 0.6394,
80
+ "rewards/accuracies": 1.0,
81
+ "rewards/chosen": 0.06118398904800415,
82
+ "rewards/margins": 0.11314834654331207,
83
+ "rewards/rejected": -0.051964350044727325,
84
+ "step": 40
85
+ },
86
+ {
87
+ "epoch": 2.0,
88
+ "grad_norm": 0.5811657939889897,
89
+ "learning_rate": 4.849231551964771e-06,
90
+ "logits/chosen": -1.680452585220337,
91
+ "logits/rejected": -1.6310945749282837,
92
+ "logps/chosen": -63.04345703125,
93
+ "logps/rejected": -102.03121948242188,
94
+ "loss": 0.5323,
95
+ "rewards/accuracies": 1.0,
96
+ "rewards/chosen": 0.1268819272518158,
97
+ "rewards/margins": 0.3797517418861389,
98
+ "rewards/rejected": -0.2528698444366455,
99
+ "step": 50
100
+ },
101
+ {
102
+ "epoch": 2.4,
103
+ "grad_norm": 0.3746410656156246,
104
+ "learning_rate": 4.707368982147318e-06,
105
+ "logits/chosen": -1.6536400318145752,
106
+ "logits/rejected": -1.496199607849121,
107
+ "logps/chosen": -55.06901931762695,
108
+ "logps/rejected": -124.32417297363281,
109
+ "loss": 0.3761,
110
+ "rewards/accuracies": 1.0,
111
+ "rewards/chosen": 0.2706597149372101,
112
+ "rewards/margins": 0.8032084703445435,
113
+ "rewards/rejected": -0.5325487852096558,
114
+ "step": 60
115
+ },
116
+ {
117
+ "epoch": 2.8,
118
+ "grad_norm": 0.2923014161860858,
119
+ "learning_rate": 4.522542485937369e-06,
120
+ "logits/chosen": -1.4761438369750977,
121
+ "logits/rejected": -1.4017894268035889,
122
+ "logps/chosen": -44.963287353515625,
123
+ "logps/rejected": -186.46328735351562,
124
+ "loss": 0.2624,
125
+ "rewards/accuracies": 1.0,
126
+ "rewards/chosen": 0.33375969529151917,
127
+ "rewards/margins": 1.4536139965057373,
128
+ "rewards/rejected": -1.1198543310165405,
129
+ "step": 70
130
+ },
131
+ {
132
+ "epoch": 3.2,
133
+ "grad_norm": 0.1321053576972282,
134
+ "learning_rate": 4.2983495008466285e-06,
135
+ "logits/chosen": -1.4355762004852295,
136
+ "logits/rejected": -1.3916115760803223,
137
+ "logps/chosen": -38.39271926879883,
138
+ "logps/rejected": -254.19973754882812,
139
+ "loss": 0.2324,
140
+ "rewards/accuracies": 1.0,
141
+ "rewards/chosen": 0.35416102409362793,
142
+ "rewards/margins": 2.1357216835021973,
143
+ "rewards/rejected": -1.7815606594085693,
144
+ "step": 80
145
+ },
146
+ {
147
+ "epoch": 3.6,
148
+ "grad_norm": 0.14508831317113868,
149
+ "learning_rate": 4.039153688314146e-06,
150
+ "logits/chosen": -1.5442588329315186,
151
+ "logits/rejected": -1.526135802268982,
152
+ "logps/chosen": -35.97533416748047,
153
+ "logps/rejected": -263.40643310546875,
154
+ "loss": 0.2095,
155
+ "rewards/accuracies": 1.0,
156
+ "rewards/chosen": 0.41692012548446655,
157
+ "rewards/margins": 2.3173232078552246,
158
+ "rewards/rejected": -1.9004030227661133,
159
+ "step": 90
160
+ },
161
+ {
162
+ "epoch": 4.0,
163
+ "grad_norm": 0.44893168632741065,
164
+ "learning_rate": 3.7500000000000005e-06,
165
+ "logits/chosen": -1.2732983827590942,
166
+ "logits/rejected": -1.2810883522033691,
167
+ "logps/chosen": -31.883533477783203,
168
+ "logps/rejected": -319.82061767578125,
169
+ "loss": 0.1721,
170
+ "rewards/accuracies": 1.0,
171
+ "rewards/chosen": 0.4404378831386566,
172
+ "rewards/margins": 2.844158411026001,
173
+ "rewards/rejected": -2.4037206172943115,
174
+ "step": 100
175
+ },
176
+ {
177
+ "epoch": 4.4,
178
+ "grad_norm": 0.6002566776493338,
179
+ "learning_rate": 3.436516483539781e-06,
180
+ "logits/chosen": -1.155807614326477,
181
+ "logits/rejected": -1.1776975393295288,
182
+ "logps/chosen": -31.725173950195312,
183
+ "logps/rejected": -360.02349853515625,
184
+ "loss": 0.1083,
185
+ "rewards/accuracies": 1.0,
186
+ "rewards/chosen": 0.40672487020492554,
187
+ "rewards/margins": 3.2172622680664062,
188
+ "rewards/rejected": -2.810537338256836,
189
+ "step": 110
190
+ },
191
+ {
192
+ "epoch": 4.8,
193
+ "grad_norm": 0.15892065424589785,
194
+ "learning_rate": 3.1048047389991693e-06,
195
+ "logits/chosen": -1.228723406791687,
196
+ "logits/rejected": -1.1159226894378662,
197
+ "logps/chosen": -44.75030517578125,
198
+ "logps/rejected": -419.255126953125,
199
+ "loss": 0.0409,
200
+ "rewards/accuracies": 1.0,
201
+ "rewards/chosen": 0.3626783490180969,
202
+ "rewards/margins": 3.842852830886841,
203
+ "rewards/rejected": -3.4801743030548096,
204
+ "step": 120
205
+ },
206
+ {
207
+ "epoch": 5.2,
208
+ "grad_norm": 0.04827040736700255,
209
+ "learning_rate": 2.761321158169134e-06,
210
+ "logits/chosen": -1.0756232738494873,
211
+ "logits/rejected": -1.0372257232666016,
212
+ "logps/chosen": -64.86322021484375,
213
+ "logps/rejected": -753.9734497070312,
214
+ "loss": 0.0049,
215
+ "rewards/accuracies": 1.0,
216
+ "rewards/chosen": 0.14987702667713165,
217
+ "rewards/margins": 6.945639133453369,
218
+ "rewards/rejected": -6.795762062072754,
219
+ "step": 130
220
+ },
221
+ {
222
+ "epoch": 5.6,
223
+ "grad_norm": 0.02673618186874915,
224
+ "learning_rate": 2.4127512582437486e-06,
225
+ "logits/chosen": -1.1154285669326782,
226
+ "logits/rejected": -0.9958986043930054,
227
+ "logps/chosen": -69.85175323486328,
228
+ "logps/rejected": -913.9327392578125,
229
+ "loss": 0.0029,
230
+ "rewards/accuracies": 1.0,
231
+ "rewards/chosen": 0.10866590589284897,
232
+ "rewards/margins": 8.513258934020996,
233
+ "rewards/rejected": -8.404594421386719,
234
+ "step": 140
235
+ },
236
+ {
237
+ "epoch": 6.0,
238
+ "grad_norm": 0.017203740876186722,
239
+ "learning_rate": 2.0658795558326745e-06,
240
+ "logits/chosen": -1.0685889720916748,
241
+ "logits/rejected": -1.033384919166565,
242
+ "logps/chosen": -67.58888244628906,
243
+ "logps/rejected": -917.8318481445312,
244
+ "loss": 0.0024,
245
+ "rewards/accuracies": 1.0,
246
+ "rewards/chosen": 0.08431224524974823,
247
+ "rewards/margins": 8.512849807739258,
248
+ "rewards/rejected": -8.42853832244873,
249
+ "step": 150
250
+ },
251
+ {
252
+ "epoch": 6.4,
253
+ "grad_norm": 0.015005056648195798,
254
+ "learning_rate": 1.7274575140626318e-06,
255
+ "logits/chosen": -1.1218775510787964,
256
+ "logits/rejected": -1.0821073055267334,
257
+ "logps/chosen": -61.10993576049805,
258
+ "logps/rejected": -929.9596557617188,
259
+ "loss": 0.0019,
260
+ "rewards/accuracies": 1.0,
261
+ "rewards/chosen": 0.15528056025505066,
262
+ "rewards/margins": 8.694746971130371,
263
+ "rewards/rejected": -8.539467811584473,
264
+ "step": 160
265
+ },
266
+ {
267
+ "epoch": 6.8,
268
+ "grad_norm": 0.011209472305113497,
269
+ "learning_rate": 1.4040721330273063e-06,
270
+ "logits/chosen": -1.1669288873672485,
271
+ "logits/rejected": -1.057388424873352,
272
+ "logps/chosen": -59.08354568481445,
273
+ "logps/rejected": -995.8665771484375,
274
+ "loss": 0.0016,
275
+ "rewards/accuracies": 1.0,
276
+ "rewards/chosen": 0.21780769526958466,
277
+ "rewards/margins": 9.462162971496582,
278
+ "rewards/rejected": -9.244354248046875,
279
+ "step": 170
280
+ },
281
+ {
282
+ "epoch": 7.2,
283
+ "grad_norm": 0.012213732988109802,
284
+ "learning_rate": 1.1020177413231334e-06,
285
+ "logits/chosen": -1.0734723806381226,
286
+ "logits/rejected": -1.0714576244354248,
287
+ "logps/chosen": -54.94775390625,
288
+ "logps/rejected": -925.8068237304688,
289
+ "loss": 0.0017,
290
+ "rewards/accuracies": 1.0,
291
+ "rewards/chosen": 0.20917268097400665,
292
+ "rewards/margins": 8.696084976196289,
293
+ "rewards/rejected": -8.486912727355957,
294
+ "step": 180
295
+ },
296
+ {
297
+ "epoch": 7.6,
298
+ "grad_norm": 0.0137000842262227,
299
+ "learning_rate": 8.271734841028553e-07,
300
+ "logits/chosen": -1.1287630796432495,
301
+ "logits/rejected": -1.0409098863601685,
302
+ "logps/chosen": -55.30584716796875,
303
+ "logps/rejected": -984.7487182617188,
304
+ "loss": 0.0016,
305
+ "rewards/accuracies": 1.0,
306
+ "rewards/chosen": 0.21431966125965118,
307
+ "rewards/margins": 9.310848236083984,
308
+ "rewards/rejected": -9.096528053283691,
309
+ "step": 190
310
+ },
311
+ {
312
+ "epoch": 8.0,
313
+ "grad_norm": 0.012982424059208061,
314
+ "learning_rate": 5.848888922025553e-07,
315
+ "logits/chosen": -1.1048905849456787,
316
+ "logits/rejected": -1.0343353748321533,
317
+ "logps/chosen": -55.2299919128418,
318
+ "logps/rejected": -963.7037353515625,
319
+ "loss": 0.0016,
320
+ "rewards/accuracies": 1.0,
321
+ "rewards/chosen": 0.2215685397386551,
322
+ "rewards/margins": 9.095995903015137,
323
+ "rewards/rejected": -8.87442684173584,
324
+ "step": 200
325
+ },
326
+ {
327
+ "epoch": 8.4,
328
+ "grad_norm": 0.011333873276454402,
329
+ "learning_rate": 3.798797596089351e-07,
330
+ "logits/chosen": -1.0944920778274536,
331
+ "logits/rejected": -0.9962091445922852,
332
+ "logps/chosen": -53.680267333984375,
333
+ "logps/rejected": -962.94775390625,
334
+ "loss": 0.0015,
335
+ "rewards/accuracies": 1.0,
336
+ "rewards/chosen": 0.24401147663593292,
337
+ "rewards/margins": 9.146904945373535,
338
+ "rewards/rejected": -8.902894973754883,
339
+ "step": 210
340
+ },
341
+ {
342
+ "epoch": 8.8,
343
+ "grad_norm": 0.010235633073822109,
344
+ "learning_rate": 2.1613635589349756e-07,
345
+ "logits/chosen": -1.069654107093811,
346
+ "logits/rejected": -1.0441746711730957,
347
+ "logps/chosen": -53.099769592285156,
348
+ "logps/rejected": -953.6365356445312,
349
+ "loss": 0.0015,
350
+ "rewards/accuracies": 1.0,
351
+ "rewards/chosen": 0.22151072323322296,
352
+ "rewards/margins": 8.963228225708008,
353
+ "rewards/rejected": -8.741719245910645,
354
+ "step": 220
355
+ },
356
+ {
357
+ "epoch": 9.2,
358
+ "grad_norm": 0.012374131573979113,
359
+ "learning_rate": 9.684576015420277e-08,
360
+ "logits/chosen": -1.023950219154358,
361
+ "logits/rejected": -1.0566022396087646,
362
+ "logps/chosen": -50.87507247924805,
363
+ "logps/rejected": -911.796875,
364
+ "loss": 0.0015,
365
+ "rewards/accuracies": 1.0,
366
+ "rewards/chosen": 0.20660237967967987,
367
+ "rewards/margins": 8.510995864868164,
368
+ "rewards/rejected": -8.30439281463623,
369
+ "step": 230
370
+ },
371
+ {
372
+ "epoch": 9.6,
373
+ "grad_norm": 0.010557040037916637,
374
+ "learning_rate": 2.4329828146074096e-08,
375
+ "logits/chosen": -1.1207894086837769,
376
+ "logits/rejected": -1.097846269607544,
377
+ "logps/chosen": -54.16496658325195,
378
+ "logps/rejected": -937.8318481445312,
379
+ "loss": 0.0014,
380
+ "rewards/accuracies": 1.0,
381
+ "rewards/chosen": 0.23018613457679749,
382
+ "rewards/margins": 8.850129127502441,
383
+ "rewards/rejected": -8.619942665100098,
384
+ "step": 240
385
+ },
386
+ {
387
+ "epoch": 10.0,
388
+ "grad_norm": 0.01132325198281179,
389
+ "learning_rate": 0.0,
390
+ "logits/chosen": -1.1102557182312012,
391
+ "logits/rejected": -1.01675546169281,
392
+ "logps/chosen": -56.481834411621094,
393
+ "logps/rejected": -970.00634765625,
394
+ "loss": 0.0015,
395
+ "rewards/accuracies": 1.0,
396
+ "rewards/chosen": 0.22424273192882538,
397
+ "rewards/margins": 9.186958312988281,
398
+ "rewards/rejected": -8.962717056274414,
399
+ "step": 250
400
+ },
401
+ {
402
+ "epoch": 10.0,
403
+ "step": 250,
404
+ "total_flos": 0.0,
405
+ "train_loss": 0.18631765481829643,
406
+ "train_runtime": 3826.8955,
407
+ "train_samples_per_second": 2.09,
408
+ "train_steps_per_second": 0.065
409
+ }
410
+ ],
411
+ "logging_steps": 10,
412
+ "max_steps": 250,
413
+ "num_input_tokens_seen": 0,
414
+ "num_train_epochs": 10,
415
+ "save_steps": 1000,
416
+ "stateful_callbacks": {
417
+ "TrainerControl": {
418
+ "args": {
419
+ "should_epoch_stop": false,
420
+ "should_evaluate": false,
421
+ "should_log": false,
422
+ "should_save": true,
423
+ "should_training_stop": true
424
+ },
425
+ "attributes": {}
426
+ }
427
+ },
428
+ "total_flos": 0.0,
429
+ "train_batch_size": 2,
430
+ "trial_name": null,
431
+ "trial_params": null
432
+ }