nisten commited on
Commit
ad1a955
·
verified ·
1 Parent(s): 49f07c2

Upload folder using huggingface_hub

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. checkpoint-1000/config.json +30 -0
  2. checkpoint-1000/generation_config.json +7 -0
  3. checkpoint-1000/model.safetensors +3 -0
  4. checkpoint-1000/optimizer.pt +3 -0
  5. checkpoint-1000/rng_state.pth +3 -0
  6. checkpoint-1000/scheduler.pt +3 -0
  7. checkpoint-1000/trainer_state.json +749 -0
  8. checkpoint-1000/training_args.bin +3 -0
  9. checkpoint-1500/config.json +30 -0
  10. checkpoint-1500/generation_config.json +7 -0
  11. checkpoint-1500/model.safetensors +3 -0
  12. checkpoint-1500/optimizer.pt +3 -0
  13. checkpoint-1500/rng_state.pth +3 -0
  14. checkpoint-1500/scheduler.pt +3 -0
  15. checkpoint-1500/trainer_state.json +1107 -0
  16. checkpoint-1500/training_args.bin +3 -0
  17. checkpoint-2000/config.json +30 -0
  18. checkpoint-2000/generation_config.json +7 -0
  19. checkpoint-2000/model.safetensors +3 -0
  20. checkpoint-2000/optimizer.pt +3 -0
  21. checkpoint-2000/rng_state.pth +3 -0
  22. checkpoint-2000/scheduler.pt +3 -0
  23. checkpoint-2000/trainer_state.json +1465 -0
  24. checkpoint-2000/training_args.bin +3 -0
  25. checkpoint-2500/config.json +30 -0
  26. checkpoint-2500/generation_config.json +7 -0
  27. checkpoint-2500/model.safetensors +3 -0
  28. checkpoint-2500/optimizer.pt +3 -0
  29. checkpoint-2500/rng_state.pth +3 -0
  30. checkpoint-2500/scheduler.pt +3 -0
  31. checkpoint-2500/trainer_state.json +1823 -0
  32. checkpoint-2500/training_args.bin +3 -0
  33. checkpoint-3000/config.json +30 -0
  34. checkpoint-3000/generation_config.json +7 -0
  35. checkpoint-3000/model.safetensors +3 -0
  36. checkpoint-3000/optimizer.pt +3 -0
  37. checkpoint-3000/rng_state.pth +3 -0
  38. checkpoint-3000/scheduler.pt +3 -0
  39. checkpoint-3000/trainer_state.json +2181 -0
  40. checkpoint-3000/training_args.bin +3 -0
  41. checkpoint-3500/config.json +30 -0
  42. checkpoint-3500/generation_config.json +7 -0
  43. checkpoint-3500/model.safetensors +3 -0
  44. checkpoint-3500/optimizer.pt +3 -0
  45. checkpoint-3500/rng_state.pth +3 -0
  46. checkpoint-3500/scheduler.pt +3 -0
  47. checkpoint-3500/trainer_state.json +2539 -0
  48. checkpoint-3500/training_args.bin +3 -0
  49. checkpoint-4000/config.json +30 -0
  50. checkpoint-4000/generation_config.json +7 -0
checkpoint-1000/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-1000/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-1000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:83b2283690c190c43aa549a22d1a3fda798ff4530c240e2c8cc935eb7bbf3396
3
+ size 802337544
checkpoint-1000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bf1f00b3b4a003d644afab735229e0db118f4bbda037c2de97ab85a9e7b88d9e
3
+ size 2407189434
checkpoint-1000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f56790c3ee491baab329d1fe00f10cb0b9f5114649fad9b4021437c861c9b810
3
+ size 13990
checkpoint-1000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42670ac61cea9c26497d71cb2283cdd42e4f835a074a8600a4087204b15b45a5
3
+ size 1000
checkpoint-1000/trainer_state.json ADDED
@@ -0,0 +1,749 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.9072233438491821,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-1000",
4
+ "epoch": 1.4970059880239521,
5
+ "eval_steps": 500,
6
+ "global_step": 1000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ }
727
+ ],
728
+ "logging_steps": 10,
729
+ "max_steps": 9000,
730
+ "num_input_tokens_seen": 0,
731
+ "num_train_epochs": 14,
732
+ "save_steps": 500,
733
+ "stateful_callbacks": {
734
+ "TrainerControl": {
735
+ "args": {
736
+ "should_epoch_stop": false,
737
+ "should_evaluate": false,
738
+ "should_log": false,
739
+ "should_save": true,
740
+ "should_training_stop": false
741
+ },
742
+ "attributes": {}
743
+ }
744
+ },
745
+ "total_flos": 6.954005429747712e+16,
746
+ "train_batch_size": 16,
747
+ "trial_name": null,
748
+ "trial_params": null
749
+ }
checkpoint-1000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-1500/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-1500/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-1500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:17deca4da9ef56af97f362253eddabc09ca7f3579e6794f92d70a1673e3bd15d
3
+ size 802337544
checkpoint-1500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4127998d8e3413a0c6fb58f1542d1de1e36359c372bb0a354cce7caaf12c16a2
3
+ size 2407189434
checkpoint-1500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e721764cfe6d7ef80d19a2ff27d723d65ac25eb377b60bb6e4c10a66202a1bf
3
+ size 13990
checkpoint-1500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a4963c3a03730b160fe3c8dc1c59fbd690229290da46d26feac9ca8ca03aa9a5
3
+ size 1000
checkpoint-1500/trainer_state.json ADDED
@@ -0,0 +1,1107 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.7798940539360046,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-1500",
4
+ "epoch": 2.245508982035928,
5
+ "eval_steps": 500,
6
+ "global_step": 1500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 1.5119760479041915,
729
+ "grad_norm": 0.2890625,
730
+ "learning_rate": 0.00019416730477082533,
731
+ "loss": 0.9444,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 1.5269461077844313,
736
+ "grad_norm": 0.296875,
737
+ "learning_rate": 0.00019404886375423984,
738
+ "loss": 0.9829,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 1.5419161676646707,
743
+ "grad_norm": 0.27734375,
744
+ "learning_rate": 0.00019392926910386353,
745
+ "loss": 0.9532,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 1.55688622754491,
750
+ "grad_norm": 0.265625,
751
+ "learning_rate": 0.00019380852228668304,
752
+ "loss": 0.9769,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 1.5718562874251498,
757
+ "grad_norm": 0.275390625,
758
+ "learning_rate": 0.00019368662478381799,
759
+ "loss": 0.9783,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 1.5868263473053892,
764
+ "grad_norm": 0.2890625,
765
+ "learning_rate": 0.00019356357809050247,
766
+ "loss": 0.9881,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 1.6017964071856288,
771
+ "grad_norm": 0.232421875,
772
+ "learning_rate": 0.00019343938371606712,
773
+ "loss": 0.9883,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 1.6167664670658684,
778
+ "grad_norm": 0.2734375,
779
+ "learning_rate": 0.00019331404318392027,
780
+ "loss": 0.9893,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 1.6317365269461077,
785
+ "grad_norm": 0.248046875,
786
+ "learning_rate": 0.00019318755803152945,
787
+ "loss": 0.9851,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 1.6467065868263473,
792
+ "grad_norm": 0.28515625,
793
+ "learning_rate": 0.00019305992981040246,
794
+ "loss": 0.9531,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 1.6616766467065869,
799
+ "grad_norm": 0.275390625,
800
+ "learning_rate": 0.00019293116008606837,
801
+ "loss": 0.9717,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 1.6766467065868262,
806
+ "grad_norm": 0.232421875,
807
+ "learning_rate": 0.00019280125043805824,
808
+ "loss": 0.9699,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 1.6916167664670658,
813
+ "grad_norm": 0.2294921875,
814
+ "learning_rate": 0.00019267020245988592,
815
+ "loss": 0.9407,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 1.7065868263473054,
820
+ "grad_norm": 0.251953125,
821
+ "learning_rate": 0.00019253801775902824,
822
+ "loss": 0.977,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 1.7215568862275448,
827
+ "grad_norm": 0.287109375,
828
+ "learning_rate": 0.0001924046979569055,
829
+ "loss": 0.9549,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 1.7365269461077846,
834
+ "grad_norm": 0.248046875,
835
+ "learning_rate": 0.00019227024468886157,
836
+ "loss": 0.9824,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 1.751497005988024,
841
+ "grad_norm": 0.259765625,
842
+ "learning_rate": 0.00019213465960414368,
843
+ "loss": 0.9936,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 1.7664670658682635,
848
+ "grad_norm": 0.29296875,
849
+ "learning_rate": 0.00019199794436588243,
850
+ "loss": 1.0042,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 1.781437125748503,
855
+ "grad_norm": 0.23828125,
856
+ "learning_rate": 0.0001918601006510711,
857
+ "loss": 0.9629,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 1.7964071856287425,
862
+ "grad_norm": 0.2890625,
863
+ "learning_rate": 0.00019172113015054532,
864
+ "loss": 0.9522,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 1.811377245508982,
869
+ "grad_norm": 0.26953125,
870
+ "learning_rate": 0.0001915810345689622,
871
+ "loss": 0.9806,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 1.8263473053892216,
876
+ "grad_norm": 0.263671875,
877
+ "learning_rate": 0.00019143981562477947,
878
+ "loss": 0.9736,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 1.841317365269461,
883
+ "grad_norm": 0.2578125,
884
+ "learning_rate": 0.00019129747505023436,
885
+ "loss": 0.9701,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 1.8562874251497006,
890
+ "grad_norm": 0.22265625,
891
+ "learning_rate": 0.00019115401459132247,
892
+ "loss": 0.9494,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 1.8712574850299402,
897
+ "grad_norm": 0.251953125,
898
+ "learning_rate": 0.00019100943600777615,
899
+ "loss": 0.9922,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 1.8862275449101795,
904
+ "grad_norm": 0.2490234375,
905
+ "learning_rate": 0.00019086374107304312,
906
+ "loss": 0.9711,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 1.9011976047904193,
911
+ "grad_norm": 0.25390625,
912
+ "learning_rate": 0.00019071693157426457,
913
+ "loss": 0.9664,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 1.9161676646706587,
918
+ "grad_norm": 0.26171875,
919
+ "learning_rate": 0.00019056900931225333,
920
+ "loss": 0.9591,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 1.931137724550898,
925
+ "grad_norm": 0.380859375,
926
+ "learning_rate": 0.00019041997610147167,
927
+ "loss": 0.9942,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 1.9461077844311379,
932
+ "grad_norm": 0.2294921875,
933
+ "learning_rate": 0.0001902698337700092,
934
+ "loss": 0.9391,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 1.9610778443113772,
939
+ "grad_norm": 0.298828125,
940
+ "learning_rate": 0.00019011858415956038,
941
+ "loss": 0.9993,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 1.9760479041916168,
946
+ "grad_norm": 0.29296875,
947
+ "learning_rate": 0.0001899662291254018,
948
+ "loss": 0.9571,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 1.9910179640718564,
953
+ "grad_norm": 0.283203125,
954
+ "learning_rate": 0.0001898127705363696,
955
+ "loss": 0.9835,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 2.0059880239520957,
960
+ "grad_norm": 0.27734375,
961
+ "learning_rate": 0.00018965821027483654,
962
+ "loss": 0.9305,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 2.020958083832335,
967
+ "grad_norm": 0.26171875,
968
+ "learning_rate": 0.00018950255023668876,
969
+ "loss": 0.8295,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 2.035928143712575,
974
+ "grad_norm": 0.240234375,
975
+ "learning_rate": 0.00018934579233130267,
976
+ "loss": 0.7653,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 2.0508982035928143,
981
+ "grad_norm": 0.29296875,
982
+ "learning_rate": 0.00018918793848152142,
983
+ "loss": 0.7581,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 2.065868263473054,
988
+ "grad_norm": 0.29296875,
989
+ "learning_rate": 0.00018902899062363143,
990
+ "loss": 0.7983,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 2.0808383233532934,
995
+ "grad_norm": 0.279296875,
996
+ "learning_rate": 0.0001888689507073385,
997
+ "loss": 0.8187,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 2.095808383233533,
1002
+ "grad_norm": 0.29296875,
1003
+ "learning_rate": 0.0001887078206957441,
1004
+ "loss": 0.7879,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 2.1107784431137726,
1009
+ "grad_norm": 0.2734375,
1010
+ "learning_rate": 0.000188545602565321,
1011
+ "loss": 0.8083,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 2.125748502994012,
1016
+ "grad_norm": 0.263671875,
1017
+ "learning_rate": 0.00018838229830588934,
1018
+ "loss": 0.8057,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 2.1407185628742513,
1023
+ "grad_norm": 0.244140625,
1024
+ "learning_rate": 0.00018821790992059196,
1025
+ "loss": 0.8194,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 2.155688622754491,
1030
+ "grad_norm": 0.2451171875,
1031
+ "learning_rate": 0.00018805243942587,
1032
+ "loss": 0.7958,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 2.1706586826347305,
1037
+ "grad_norm": 0.248046875,
1038
+ "learning_rate": 0.00018788588885143808,
1039
+ "loss": 0.8169,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 2.18562874251497,
1044
+ "grad_norm": 0.265625,
1045
+ "learning_rate": 0.00018771826024025946,
1046
+ "loss": 0.7722,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 2.2005988023952097,
1051
+ "grad_norm": 0.265625,
1052
+ "learning_rate": 0.0001875495556485208,
1053
+ "loss": 0.7934,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 2.215568862275449,
1058
+ "grad_norm": 0.244140625,
1059
+ "learning_rate": 0.00018737977714560738,
1060
+ "loss": 0.7915,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 2.230538922155689,
1065
+ "grad_norm": 0.310546875,
1066
+ "learning_rate": 0.00018720892681407708,
1067
+ "loss": 0.8021,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 2.245508982035928,
1072
+ "grad_norm": 0.298828125,
1073
+ "learning_rate": 0.00018703700674963547,
1074
+ "loss": 0.7987,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 2.245508982035928,
1079
+ "eval_loss": 0.7798940539360046,
1080
+ "eval_runtime": 109.62,
1081
+ "eval_samples_per_second": 9.122,
1082
+ "eval_steps_per_second": 1.14,
1083
+ "step": 1500
1084
+ }
1085
+ ],
1086
+ "logging_steps": 10,
1087
+ "max_steps": 9000,
1088
+ "num_input_tokens_seen": 0,
1089
+ "num_train_epochs": 14,
1090
+ "save_steps": 500,
1091
+ "stateful_callbacks": {
1092
+ "TrainerControl": {
1093
+ "args": {
1094
+ "should_epoch_stop": false,
1095
+ "should_evaluate": false,
1096
+ "should_log": false,
1097
+ "should_save": true,
1098
+ "should_training_stop": false
1099
+ },
1100
+ "attributes": {}
1101
+ }
1102
+ },
1103
+ "total_flos": 1.0428398435303424e+17,
1104
+ "train_batch_size": 16,
1105
+ "trial_name": null,
1106
+ "trial_params": null
1107
+ }
checkpoint-1500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-2000/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-2000/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-2000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ea1ad04a15feae5c1349d35fac34da52c910834ee59e290b444a7f2ef8ac7a96
3
+ size 802337544
checkpoint-2000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0dabd483f1784f8dc4753f8788c023fef83497a8090bd80b3e4e539a614d2bc4
3
+ size 2407189434
checkpoint-2000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d77bc69a6941ab95918ba94cb68ce1fc6e31fd48c77db098e43e275b8386fcb6
3
+ size 13990
checkpoint-2000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:75e45437b62903a953b21df5ba0bc38684d1cc900ef4c0d2bf23f2ad8aafefb4
3
+ size 1000
checkpoint-2000/trainer_state.json ADDED
@@ -0,0 +1,1465 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.6911507844924927,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-2000",
4
+ "epoch": 2.9940119760479043,
5
+ "eval_steps": 500,
6
+ "global_step": 2000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 1.5119760479041915,
729
+ "grad_norm": 0.2890625,
730
+ "learning_rate": 0.00019416730477082533,
731
+ "loss": 0.9444,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 1.5269461077844313,
736
+ "grad_norm": 0.296875,
737
+ "learning_rate": 0.00019404886375423984,
738
+ "loss": 0.9829,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 1.5419161676646707,
743
+ "grad_norm": 0.27734375,
744
+ "learning_rate": 0.00019392926910386353,
745
+ "loss": 0.9532,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 1.55688622754491,
750
+ "grad_norm": 0.265625,
751
+ "learning_rate": 0.00019380852228668304,
752
+ "loss": 0.9769,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 1.5718562874251498,
757
+ "grad_norm": 0.275390625,
758
+ "learning_rate": 0.00019368662478381799,
759
+ "loss": 0.9783,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 1.5868263473053892,
764
+ "grad_norm": 0.2890625,
765
+ "learning_rate": 0.00019356357809050247,
766
+ "loss": 0.9881,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 1.6017964071856288,
771
+ "grad_norm": 0.232421875,
772
+ "learning_rate": 0.00019343938371606712,
773
+ "loss": 0.9883,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 1.6167664670658684,
778
+ "grad_norm": 0.2734375,
779
+ "learning_rate": 0.00019331404318392027,
780
+ "loss": 0.9893,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 1.6317365269461077,
785
+ "grad_norm": 0.248046875,
786
+ "learning_rate": 0.00019318755803152945,
787
+ "loss": 0.9851,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 1.6467065868263473,
792
+ "grad_norm": 0.28515625,
793
+ "learning_rate": 0.00019305992981040246,
794
+ "loss": 0.9531,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 1.6616766467065869,
799
+ "grad_norm": 0.275390625,
800
+ "learning_rate": 0.00019293116008606837,
801
+ "loss": 0.9717,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 1.6766467065868262,
806
+ "grad_norm": 0.232421875,
807
+ "learning_rate": 0.00019280125043805824,
808
+ "loss": 0.9699,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 1.6916167664670658,
813
+ "grad_norm": 0.2294921875,
814
+ "learning_rate": 0.00019267020245988592,
815
+ "loss": 0.9407,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 1.7065868263473054,
820
+ "grad_norm": 0.251953125,
821
+ "learning_rate": 0.00019253801775902824,
822
+ "loss": 0.977,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 1.7215568862275448,
827
+ "grad_norm": 0.287109375,
828
+ "learning_rate": 0.0001924046979569055,
829
+ "loss": 0.9549,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 1.7365269461077846,
834
+ "grad_norm": 0.248046875,
835
+ "learning_rate": 0.00019227024468886157,
836
+ "loss": 0.9824,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 1.751497005988024,
841
+ "grad_norm": 0.259765625,
842
+ "learning_rate": 0.00019213465960414368,
843
+ "loss": 0.9936,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 1.7664670658682635,
848
+ "grad_norm": 0.29296875,
849
+ "learning_rate": 0.00019199794436588243,
850
+ "loss": 1.0042,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 1.781437125748503,
855
+ "grad_norm": 0.23828125,
856
+ "learning_rate": 0.0001918601006510711,
857
+ "loss": 0.9629,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 1.7964071856287425,
862
+ "grad_norm": 0.2890625,
863
+ "learning_rate": 0.00019172113015054532,
864
+ "loss": 0.9522,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 1.811377245508982,
869
+ "grad_norm": 0.26953125,
870
+ "learning_rate": 0.0001915810345689622,
871
+ "loss": 0.9806,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 1.8263473053892216,
876
+ "grad_norm": 0.263671875,
877
+ "learning_rate": 0.00019143981562477947,
878
+ "loss": 0.9736,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 1.841317365269461,
883
+ "grad_norm": 0.2578125,
884
+ "learning_rate": 0.00019129747505023436,
885
+ "loss": 0.9701,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 1.8562874251497006,
890
+ "grad_norm": 0.22265625,
891
+ "learning_rate": 0.00019115401459132247,
892
+ "loss": 0.9494,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 1.8712574850299402,
897
+ "grad_norm": 0.251953125,
898
+ "learning_rate": 0.00019100943600777615,
899
+ "loss": 0.9922,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 1.8862275449101795,
904
+ "grad_norm": 0.2490234375,
905
+ "learning_rate": 0.00019086374107304312,
906
+ "loss": 0.9711,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 1.9011976047904193,
911
+ "grad_norm": 0.25390625,
912
+ "learning_rate": 0.00019071693157426457,
913
+ "loss": 0.9664,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 1.9161676646706587,
918
+ "grad_norm": 0.26171875,
919
+ "learning_rate": 0.00019056900931225333,
920
+ "loss": 0.9591,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 1.931137724550898,
925
+ "grad_norm": 0.380859375,
926
+ "learning_rate": 0.00019041997610147167,
927
+ "loss": 0.9942,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 1.9461077844311379,
932
+ "grad_norm": 0.2294921875,
933
+ "learning_rate": 0.0001902698337700092,
934
+ "loss": 0.9391,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 1.9610778443113772,
939
+ "grad_norm": 0.298828125,
940
+ "learning_rate": 0.00019011858415956038,
941
+ "loss": 0.9993,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 1.9760479041916168,
946
+ "grad_norm": 0.29296875,
947
+ "learning_rate": 0.0001899662291254018,
948
+ "loss": 0.9571,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 1.9910179640718564,
953
+ "grad_norm": 0.283203125,
954
+ "learning_rate": 0.0001898127705363696,
955
+ "loss": 0.9835,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 2.0059880239520957,
960
+ "grad_norm": 0.27734375,
961
+ "learning_rate": 0.00018965821027483654,
962
+ "loss": 0.9305,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 2.020958083832335,
967
+ "grad_norm": 0.26171875,
968
+ "learning_rate": 0.00018950255023668876,
969
+ "loss": 0.8295,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 2.035928143712575,
974
+ "grad_norm": 0.240234375,
975
+ "learning_rate": 0.00018934579233130267,
976
+ "loss": 0.7653,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 2.0508982035928143,
981
+ "grad_norm": 0.29296875,
982
+ "learning_rate": 0.00018918793848152142,
983
+ "loss": 0.7581,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 2.065868263473054,
988
+ "grad_norm": 0.29296875,
989
+ "learning_rate": 0.00018902899062363143,
990
+ "loss": 0.7983,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 2.0808383233532934,
995
+ "grad_norm": 0.279296875,
996
+ "learning_rate": 0.0001888689507073385,
997
+ "loss": 0.8187,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 2.095808383233533,
1002
+ "grad_norm": 0.29296875,
1003
+ "learning_rate": 0.0001887078206957441,
1004
+ "loss": 0.7879,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 2.1107784431137726,
1009
+ "grad_norm": 0.2734375,
1010
+ "learning_rate": 0.000188545602565321,
1011
+ "loss": 0.8083,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 2.125748502994012,
1016
+ "grad_norm": 0.263671875,
1017
+ "learning_rate": 0.00018838229830588934,
1018
+ "loss": 0.8057,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 2.1407185628742513,
1023
+ "grad_norm": 0.244140625,
1024
+ "learning_rate": 0.00018821790992059196,
1025
+ "loss": 0.8194,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 2.155688622754491,
1030
+ "grad_norm": 0.2451171875,
1031
+ "learning_rate": 0.00018805243942587,
1032
+ "loss": 0.7958,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 2.1706586826347305,
1037
+ "grad_norm": 0.248046875,
1038
+ "learning_rate": 0.00018788588885143808,
1039
+ "loss": 0.8169,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 2.18562874251497,
1044
+ "grad_norm": 0.265625,
1045
+ "learning_rate": 0.00018771826024025946,
1046
+ "loss": 0.7722,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 2.2005988023952097,
1051
+ "grad_norm": 0.265625,
1052
+ "learning_rate": 0.0001875495556485208,
1053
+ "loss": 0.7934,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 2.215568862275449,
1058
+ "grad_norm": 0.244140625,
1059
+ "learning_rate": 0.00018737977714560738,
1060
+ "loss": 0.7915,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 2.230538922155689,
1065
+ "grad_norm": 0.310546875,
1066
+ "learning_rate": 0.00018720892681407708,
1067
+ "loss": 0.8021,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 2.245508982035928,
1072
+ "grad_norm": 0.298828125,
1073
+ "learning_rate": 0.00018703700674963547,
1074
+ "loss": 0.7987,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 2.245508982035928,
1079
+ "eval_loss": 0.7798940539360046,
1080
+ "eval_runtime": 109.62,
1081
+ "eval_samples_per_second": 9.122,
1082
+ "eval_steps_per_second": 1.14,
1083
+ "step": 1500
1084
+ },
1085
+ {
1086
+ "epoch": 2.2604790419161676,
1087
+ "grad_norm": 0.259765625,
1088
+ "learning_rate": 0.00018686401906110964,
1089
+ "loss": 0.7979,
1090
+ "step": 1510
1091
+ },
1092
+ {
1093
+ "epoch": 2.2754491017964074,
1094
+ "grad_norm": 0.265625,
1095
+ "learning_rate": 0.00018668996587042252,
1096
+ "loss": 0.8255,
1097
+ "step": 1520
1098
+ },
1099
+ {
1100
+ "epoch": 2.2904191616766467,
1101
+ "grad_norm": 0.2470703125,
1102
+ "learning_rate": 0.00018651484931256685,
1103
+ "loss": 0.8252,
1104
+ "step": 1530
1105
+ },
1106
+ {
1107
+ "epoch": 2.305389221556886,
1108
+ "grad_norm": 0.25,
1109
+ "learning_rate": 0.00018633867153557905,
1110
+ "loss": 0.8455,
1111
+ "step": 1540
1112
+ },
1113
+ {
1114
+ "epoch": 2.320359281437126,
1115
+ "grad_norm": 0.302734375,
1116
+ "learning_rate": 0.00018616143470051263,
1117
+ "loss": 0.8118,
1118
+ "step": 1550
1119
+ },
1120
+ {
1121
+ "epoch": 2.3353293413173652,
1122
+ "grad_norm": 0.283203125,
1123
+ "learning_rate": 0.00018598314098141206,
1124
+ "loss": 0.8122,
1125
+ "step": 1560
1126
+ },
1127
+ {
1128
+ "epoch": 2.3502994011976046,
1129
+ "grad_norm": 0.24609375,
1130
+ "learning_rate": 0.00018580379256528576,
1131
+ "loss": 0.7965,
1132
+ "step": 1570
1133
+ },
1134
+ {
1135
+ "epoch": 2.3652694610778444,
1136
+ "grad_norm": 0.279296875,
1137
+ "learning_rate": 0.00018562339165207936,
1138
+ "loss": 0.8309,
1139
+ "step": 1580
1140
+ },
1141
+ {
1142
+ "epoch": 2.3802395209580838,
1143
+ "grad_norm": 0.31640625,
1144
+ "learning_rate": 0.00018544194045464886,
1145
+ "loss": 0.8046,
1146
+ "step": 1590
1147
+ },
1148
+ {
1149
+ "epoch": 2.3952095808383236,
1150
+ "grad_norm": 0.2734375,
1151
+ "learning_rate": 0.0001852594411987334,
1152
+ "loss": 0.8467,
1153
+ "step": 1600
1154
+ },
1155
+ {
1156
+ "epoch": 2.410179640718563,
1157
+ "grad_norm": 0.275390625,
1158
+ "learning_rate": 0.00018507589612292783,
1159
+ "loss": 0.8566,
1160
+ "step": 1610
1161
+ },
1162
+ {
1163
+ "epoch": 2.4251497005988023,
1164
+ "grad_norm": 0.30859375,
1165
+ "learning_rate": 0.00018489130747865548,
1166
+ "loss": 0.8297,
1167
+ "step": 1620
1168
+ },
1169
+ {
1170
+ "epoch": 2.440119760479042,
1171
+ "grad_norm": 0.359375,
1172
+ "learning_rate": 0.00018470567753014035,
1173
+ "loss": 0.7823,
1174
+ "step": 1630
1175
+ },
1176
+ {
1177
+ "epoch": 2.4550898203592815,
1178
+ "grad_norm": 0.29296875,
1179
+ "learning_rate": 0.0001845190085543795,
1180
+ "loss": 0.8558,
1181
+ "step": 1640
1182
+ },
1183
+ {
1184
+ "epoch": 2.470059880239521,
1185
+ "grad_norm": 0.3046875,
1186
+ "learning_rate": 0.0001843313028411149,
1187
+ "loss": 0.8262,
1188
+ "step": 1650
1189
+ },
1190
+ {
1191
+ "epoch": 2.4850299401197606,
1192
+ "grad_norm": 0.2578125,
1193
+ "learning_rate": 0.00018414256269280564,
1194
+ "loss": 0.7982,
1195
+ "step": 1660
1196
+ },
1197
+ {
1198
+ "epoch": 2.5,
1199
+ "grad_norm": 0.314453125,
1200
+ "learning_rate": 0.00018395279042459937,
1201
+ "loss": 0.8182,
1202
+ "step": 1670
1203
+ },
1204
+ {
1205
+ "epoch": 2.5149700598802394,
1206
+ "grad_norm": 0.328125,
1207
+ "learning_rate": 0.00018376198836430417,
1208
+ "loss": 0.8275,
1209
+ "step": 1680
1210
+ },
1211
+ {
1212
+ "epoch": 2.529940119760479,
1213
+ "grad_norm": 0.271484375,
1214
+ "learning_rate": 0.00018357015885235982,
1215
+ "loss": 0.8102,
1216
+ "step": 1690
1217
+ },
1218
+ {
1219
+ "epoch": 2.5449101796407185,
1220
+ "grad_norm": 0.3125,
1221
+ "learning_rate": 0.0001833773042418092,
1222
+ "loss": 0.8145,
1223
+ "step": 1700
1224
+ },
1225
+ {
1226
+ "epoch": 2.5598802395209583,
1227
+ "grad_norm": 1.40625,
1228
+ "learning_rate": 0.00018318342689826938,
1229
+ "loss": 0.8279,
1230
+ "step": 1710
1231
+ },
1232
+ {
1233
+ "epoch": 2.5748502994011977,
1234
+ "grad_norm": 0.275390625,
1235
+ "learning_rate": 0.00018298852919990252,
1236
+ "loss": 0.8354,
1237
+ "step": 1720
1238
+ },
1239
+ {
1240
+ "epoch": 2.589820359281437,
1241
+ "grad_norm": 0.3125,
1242
+ "learning_rate": 0.0001827926135373869,
1243
+ "loss": 0.8106,
1244
+ "step": 1730
1245
+ },
1246
+ {
1247
+ "epoch": 2.6047904191616764,
1248
+ "grad_norm": 0.267578125,
1249
+ "learning_rate": 0.00018259568231388738,
1250
+ "loss": 0.7983,
1251
+ "step": 1740
1252
+ },
1253
+ {
1254
+ "epoch": 2.6197604790419162,
1255
+ "grad_norm": 0.251953125,
1256
+ "learning_rate": 0.00018239773794502607,
1257
+ "loss": 0.8183,
1258
+ "step": 1750
1259
+ },
1260
+ {
1261
+ "epoch": 2.6347305389221556,
1262
+ "grad_norm": 0.294921875,
1263
+ "learning_rate": 0.00018219878285885267,
1264
+ "loss": 0.8462,
1265
+ "step": 1760
1266
+ },
1267
+ {
1268
+ "epoch": 2.6497005988023954,
1269
+ "grad_norm": 0.283203125,
1270
+ "learning_rate": 0.0001819988194958146,
1271
+ "loss": 0.8227,
1272
+ "step": 1770
1273
+ },
1274
+ {
1275
+ "epoch": 2.6646706586826348,
1276
+ "grad_norm": 0.2431640625,
1277
+ "learning_rate": 0.0001817978503087272,
1278
+ "loss": 0.8104,
1279
+ "step": 1780
1280
+ },
1281
+ {
1282
+ "epoch": 2.679640718562874,
1283
+ "grad_norm": 0.271484375,
1284
+ "learning_rate": 0.0001815958777627435,
1285
+ "loss": 0.7972,
1286
+ "step": 1790
1287
+ },
1288
+ {
1289
+ "epoch": 2.694610778443114,
1290
+ "grad_norm": 0.283203125,
1291
+ "learning_rate": 0.00018139290433532416,
1292
+ "loss": 0.8339,
1293
+ "step": 1800
1294
+ },
1295
+ {
1296
+ "epoch": 2.7095808383233533,
1297
+ "grad_norm": 0.298828125,
1298
+ "learning_rate": 0.00018118893251620682,
1299
+ "loss": 0.8723,
1300
+ "step": 1810
1301
+ },
1302
+ {
1303
+ "epoch": 2.724550898203593,
1304
+ "grad_norm": 0.3203125,
1305
+ "learning_rate": 0.00018098396480737585,
1306
+ "loss": 0.8544,
1307
+ "step": 1820
1308
+ },
1309
+ {
1310
+ "epoch": 2.7395209580838324,
1311
+ "grad_norm": 0.265625,
1312
+ "learning_rate": 0.0001807780037230315,
1313
+ "loss": 0.8557,
1314
+ "step": 1830
1315
+ },
1316
+ {
1317
+ "epoch": 2.754491017964072,
1318
+ "grad_norm": 0.279296875,
1319
+ "learning_rate": 0.00018057105178955905,
1320
+ "loss": 0.8283,
1321
+ "step": 1840
1322
+ },
1323
+ {
1324
+ "epoch": 2.769461077844311,
1325
+ "grad_norm": 0.3828125,
1326
+ "learning_rate": 0.00018036311154549784,
1327
+ "loss": 0.7906,
1328
+ "step": 1850
1329
+ },
1330
+ {
1331
+ "epoch": 2.784431137724551,
1332
+ "grad_norm": 0.2490234375,
1333
+ "learning_rate": 0.0001801541855415102,
1334
+ "loss": 0.8036,
1335
+ "step": 1860
1336
+ },
1337
+ {
1338
+ "epoch": 2.7994011976047903,
1339
+ "grad_norm": 0.298828125,
1340
+ "learning_rate": 0.00017994427634035015,
1341
+ "loss": 0.7828,
1342
+ "step": 1870
1343
+ },
1344
+ {
1345
+ "epoch": 2.81437125748503,
1346
+ "grad_norm": 0.306640625,
1347
+ "learning_rate": 0.00017973338651683176,
1348
+ "loss": 0.7915,
1349
+ "step": 1880
1350
+ },
1351
+ {
1352
+ "epoch": 2.8293413173652695,
1353
+ "grad_norm": 0.228515625,
1354
+ "learning_rate": 0.00017952151865779792,
1355
+ "loss": 0.8141,
1356
+ "step": 1890
1357
+ },
1358
+ {
1359
+ "epoch": 2.844311377245509,
1360
+ "grad_norm": 0.349609375,
1361
+ "learning_rate": 0.00017930867536208826,
1362
+ "loss": 0.8155,
1363
+ "step": 1900
1364
+ },
1365
+ {
1366
+ "epoch": 2.8592814371257482,
1367
+ "grad_norm": 0.310546875,
1368
+ "learning_rate": 0.00017909485924050758,
1369
+ "loss": 0.8004,
1370
+ "step": 1910
1371
+ },
1372
+ {
1373
+ "epoch": 2.874251497005988,
1374
+ "grad_norm": 0.2470703125,
1375
+ "learning_rate": 0.00017888007291579357,
1376
+ "loss": 0.803,
1377
+ "step": 1920
1378
+ },
1379
+ {
1380
+ "epoch": 2.8892215568862274,
1381
+ "grad_norm": 0.275390625,
1382
+ "learning_rate": 0.00017866431902258478,
1383
+ "loss": 0.804,
1384
+ "step": 1930
1385
+ },
1386
+ {
1387
+ "epoch": 2.904191616766467,
1388
+ "grad_norm": 0.310546875,
1389
+ "learning_rate": 0.00017844760020738827,
1390
+ "loss": 0.8154,
1391
+ "step": 1940
1392
+ },
1393
+ {
1394
+ "epoch": 2.9191616766467066,
1395
+ "grad_norm": 0.251953125,
1396
+ "learning_rate": 0.00017822991912854713,
1397
+ "loss": 0.8257,
1398
+ "step": 1950
1399
+ },
1400
+ {
1401
+ "epoch": 2.934131736526946,
1402
+ "grad_norm": 0.3125,
1403
+ "learning_rate": 0.00017801127845620793,
1404
+ "loss": 0.8386,
1405
+ "step": 1960
1406
+ },
1407
+ {
1408
+ "epoch": 2.9491017964071857,
1409
+ "grad_norm": 0.275390625,
1410
+ "learning_rate": 0.0001777916808722879,
1411
+ "loss": 0.8003,
1412
+ "step": 1970
1413
+ },
1414
+ {
1415
+ "epoch": 2.964071856287425,
1416
+ "grad_norm": 0.28515625,
1417
+ "learning_rate": 0.000177571129070442,
1418
+ "loss": 0.8055,
1419
+ "step": 1980
1420
+ },
1421
+ {
1422
+ "epoch": 2.979041916167665,
1423
+ "grad_norm": 0.271484375,
1424
+ "learning_rate": 0.00017734962575603,
1425
+ "loss": 0.8233,
1426
+ "step": 1990
1427
+ },
1428
+ {
1429
+ "epoch": 2.9940119760479043,
1430
+ "grad_norm": 0.255859375,
1431
+ "learning_rate": 0.00017712717364608328,
1432
+ "loss": 0.8106,
1433
+ "step": 2000
1434
+ },
1435
+ {
1436
+ "epoch": 2.9940119760479043,
1437
+ "eval_loss": 0.6911507844924927,
1438
+ "eval_runtime": 109.6711,
1439
+ "eval_samples_per_second": 9.118,
1440
+ "eval_steps_per_second": 1.14,
1441
+ "step": 2000
1442
+ }
1443
+ ],
1444
+ "logging_steps": 10,
1445
+ "max_steps": 9000,
1446
+ "num_input_tokens_seen": 0,
1447
+ "num_train_epochs": 14,
1448
+ "save_steps": 500,
1449
+ "stateful_callbacks": {
1450
+ "TrainerControl": {
1451
+ "args": {
1452
+ "should_epoch_stop": false,
1453
+ "should_evaluate": false,
1454
+ "should_log": false,
1455
+ "should_save": true,
1456
+ "should_training_stop": false
1457
+ },
1458
+ "attributes": {}
1459
+ }
1460
+ },
1461
+ "total_flos": 1.3908010859495424e+17,
1462
+ "train_batch_size": 16,
1463
+ "trial_name": null,
1464
+ "trial_params": null
1465
+ }
checkpoint-2000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-2500/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-2500/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-2500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52a16fe3362916915dc5d75ad40ef5e7b0c09db26414471854e035adc89d33a4
3
+ size 802337544
checkpoint-2500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be83f8da8b0cb4c5561c901003da5411475cad4e09175269dcefd49854843c1d
3
+ size 2407189434
checkpoint-2500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:df83af33d9dd44ccaa55fb5093a41268b23107a4fc8be829b05e8a59317adb75
3
+ size 13990
checkpoint-2500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:25287cc2d60649428b47465bf2d1382b4eb9beedc34f870ac3e779ed505ba32c
3
+ size 1000
checkpoint-2500/trainer_state.json ADDED
@@ -0,0 +1,1823 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.5838193893432617,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-2500",
4
+ "epoch": 3.7425149700598803,
5
+ "eval_steps": 500,
6
+ "global_step": 2500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 1.5119760479041915,
729
+ "grad_norm": 0.2890625,
730
+ "learning_rate": 0.00019416730477082533,
731
+ "loss": 0.9444,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 1.5269461077844313,
736
+ "grad_norm": 0.296875,
737
+ "learning_rate": 0.00019404886375423984,
738
+ "loss": 0.9829,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 1.5419161676646707,
743
+ "grad_norm": 0.27734375,
744
+ "learning_rate": 0.00019392926910386353,
745
+ "loss": 0.9532,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 1.55688622754491,
750
+ "grad_norm": 0.265625,
751
+ "learning_rate": 0.00019380852228668304,
752
+ "loss": 0.9769,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 1.5718562874251498,
757
+ "grad_norm": 0.275390625,
758
+ "learning_rate": 0.00019368662478381799,
759
+ "loss": 0.9783,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 1.5868263473053892,
764
+ "grad_norm": 0.2890625,
765
+ "learning_rate": 0.00019356357809050247,
766
+ "loss": 0.9881,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 1.6017964071856288,
771
+ "grad_norm": 0.232421875,
772
+ "learning_rate": 0.00019343938371606712,
773
+ "loss": 0.9883,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 1.6167664670658684,
778
+ "grad_norm": 0.2734375,
779
+ "learning_rate": 0.00019331404318392027,
780
+ "loss": 0.9893,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 1.6317365269461077,
785
+ "grad_norm": 0.248046875,
786
+ "learning_rate": 0.00019318755803152945,
787
+ "loss": 0.9851,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 1.6467065868263473,
792
+ "grad_norm": 0.28515625,
793
+ "learning_rate": 0.00019305992981040246,
794
+ "loss": 0.9531,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 1.6616766467065869,
799
+ "grad_norm": 0.275390625,
800
+ "learning_rate": 0.00019293116008606837,
801
+ "loss": 0.9717,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 1.6766467065868262,
806
+ "grad_norm": 0.232421875,
807
+ "learning_rate": 0.00019280125043805824,
808
+ "loss": 0.9699,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 1.6916167664670658,
813
+ "grad_norm": 0.2294921875,
814
+ "learning_rate": 0.00019267020245988592,
815
+ "loss": 0.9407,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 1.7065868263473054,
820
+ "grad_norm": 0.251953125,
821
+ "learning_rate": 0.00019253801775902824,
822
+ "loss": 0.977,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 1.7215568862275448,
827
+ "grad_norm": 0.287109375,
828
+ "learning_rate": 0.0001924046979569055,
829
+ "loss": 0.9549,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 1.7365269461077846,
834
+ "grad_norm": 0.248046875,
835
+ "learning_rate": 0.00019227024468886157,
836
+ "loss": 0.9824,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 1.751497005988024,
841
+ "grad_norm": 0.259765625,
842
+ "learning_rate": 0.00019213465960414368,
843
+ "loss": 0.9936,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 1.7664670658682635,
848
+ "grad_norm": 0.29296875,
849
+ "learning_rate": 0.00019199794436588243,
850
+ "loss": 1.0042,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 1.781437125748503,
855
+ "grad_norm": 0.23828125,
856
+ "learning_rate": 0.0001918601006510711,
857
+ "loss": 0.9629,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 1.7964071856287425,
862
+ "grad_norm": 0.2890625,
863
+ "learning_rate": 0.00019172113015054532,
864
+ "loss": 0.9522,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 1.811377245508982,
869
+ "grad_norm": 0.26953125,
870
+ "learning_rate": 0.0001915810345689622,
871
+ "loss": 0.9806,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 1.8263473053892216,
876
+ "grad_norm": 0.263671875,
877
+ "learning_rate": 0.00019143981562477947,
878
+ "loss": 0.9736,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 1.841317365269461,
883
+ "grad_norm": 0.2578125,
884
+ "learning_rate": 0.00019129747505023436,
885
+ "loss": 0.9701,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 1.8562874251497006,
890
+ "grad_norm": 0.22265625,
891
+ "learning_rate": 0.00019115401459132247,
892
+ "loss": 0.9494,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 1.8712574850299402,
897
+ "grad_norm": 0.251953125,
898
+ "learning_rate": 0.00019100943600777615,
899
+ "loss": 0.9922,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 1.8862275449101795,
904
+ "grad_norm": 0.2490234375,
905
+ "learning_rate": 0.00019086374107304312,
906
+ "loss": 0.9711,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 1.9011976047904193,
911
+ "grad_norm": 0.25390625,
912
+ "learning_rate": 0.00019071693157426457,
913
+ "loss": 0.9664,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 1.9161676646706587,
918
+ "grad_norm": 0.26171875,
919
+ "learning_rate": 0.00019056900931225333,
920
+ "loss": 0.9591,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 1.931137724550898,
925
+ "grad_norm": 0.380859375,
926
+ "learning_rate": 0.00019041997610147167,
927
+ "loss": 0.9942,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 1.9461077844311379,
932
+ "grad_norm": 0.2294921875,
933
+ "learning_rate": 0.0001902698337700092,
934
+ "loss": 0.9391,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 1.9610778443113772,
939
+ "grad_norm": 0.298828125,
940
+ "learning_rate": 0.00019011858415956038,
941
+ "loss": 0.9993,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 1.9760479041916168,
946
+ "grad_norm": 0.29296875,
947
+ "learning_rate": 0.0001899662291254018,
948
+ "loss": 0.9571,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 1.9910179640718564,
953
+ "grad_norm": 0.283203125,
954
+ "learning_rate": 0.0001898127705363696,
955
+ "loss": 0.9835,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 2.0059880239520957,
960
+ "grad_norm": 0.27734375,
961
+ "learning_rate": 0.00018965821027483654,
962
+ "loss": 0.9305,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 2.020958083832335,
967
+ "grad_norm": 0.26171875,
968
+ "learning_rate": 0.00018950255023668876,
969
+ "loss": 0.8295,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 2.035928143712575,
974
+ "grad_norm": 0.240234375,
975
+ "learning_rate": 0.00018934579233130267,
976
+ "loss": 0.7653,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 2.0508982035928143,
981
+ "grad_norm": 0.29296875,
982
+ "learning_rate": 0.00018918793848152142,
983
+ "loss": 0.7581,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 2.065868263473054,
988
+ "grad_norm": 0.29296875,
989
+ "learning_rate": 0.00018902899062363143,
990
+ "loss": 0.7983,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 2.0808383233532934,
995
+ "grad_norm": 0.279296875,
996
+ "learning_rate": 0.0001888689507073385,
997
+ "loss": 0.8187,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 2.095808383233533,
1002
+ "grad_norm": 0.29296875,
1003
+ "learning_rate": 0.0001887078206957441,
1004
+ "loss": 0.7879,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 2.1107784431137726,
1009
+ "grad_norm": 0.2734375,
1010
+ "learning_rate": 0.000188545602565321,
1011
+ "loss": 0.8083,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 2.125748502994012,
1016
+ "grad_norm": 0.263671875,
1017
+ "learning_rate": 0.00018838229830588934,
1018
+ "loss": 0.8057,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 2.1407185628742513,
1023
+ "grad_norm": 0.244140625,
1024
+ "learning_rate": 0.00018821790992059196,
1025
+ "loss": 0.8194,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 2.155688622754491,
1030
+ "grad_norm": 0.2451171875,
1031
+ "learning_rate": 0.00018805243942587,
1032
+ "loss": 0.7958,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 2.1706586826347305,
1037
+ "grad_norm": 0.248046875,
1038
+ "learning_rate": 0.00018788588885143808,
1039
+ "loss": 0.8169,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 2.18562874251497,
1044
+ "grad_norm": 0.265625,
1045
+ "learning_rate": 0.00018771826024025946,
1046
+ "loss": 0.7722,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 2.2005988023952097,
1051
+ "grad_norm": 0.265625,
1052
+ "learning_rate": 0.0001875495556485208,
1053
+ "loss": 0.7934,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 2.215568862275449,
1058
+ "grad_norm": 0.244140625,
1059
+ "learning_rate": 0.00018737977714560738,
1060
+ "loss": 0.7915,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 2.230538922155689,
1065
+ "grad_norm": 0.310546875,
1066
+ "learning_rate": 0.00018720892681407708,
1067
+ "loss": 0.8021,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 2.245508982035928,
1072
+ "grad_norm": 0.298828125,
1073
+ "learning_rate": 0.00018703700674963547,
1074
+ "loss": 0.7987,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 2.245508982035928,
1079
+ "eval_loss": 0.7798940539360046,
1080
+ "eval_runtime": 109.62,
1081
+ "eval_samples_per_second": 9.122,
1082
+ "eval_steps_per_second": 1.14,
1083
+ "step": 1500
1084
+ },
1085
+ {
1086
+ "epoch": 2.2604790419161676,
1087
+ "grad_norm": 0.259765625,
1088
+ "learning_rate": 0.00018686401906110964,
1089
+ "loss": 0.7979,
1090
+ "step": 1510
1091
+ },
1092
+ {
1093
+ "epoch": 2.2754491017964074,
1094
+ "grad_norm": 0.265625,
1095
+ "learning_rate": 0.00018668996587042252,
1096
+ "loss": 0.8255,
1097
+ "step": 1520
1098
+ },
1099
+ {
1100
+ "epoch": 2.2904191616766467,
1101
+ "grad_norm": 0.2470703125,
1102
+ "learning_rate": 0.00018651484931256685,
1103
+ "loss": 0.8252,
1104
+ "step": 1530
1105
+ },
1106
+ {
1107
+ "epoch": 2.305389221556886,
1108
+ "grad_norm": 0.25,
1109
+ "learning_rate": 0.00018633867153557905,
1110
+ "loss": 0.8455,
1111
+ "step": 1540
1112
+ },
1113
+ {
1114
+ "epoch": 2.320359281437126,
1115
+ "grad_norm": 0.302734375,
1116
+ "learning_rate": 0.00018616143470051263,
1117
+ "loss": 0.8118,
1118
+ "step": 1550
1119
+ },
1120
+ {
1121
+ "epoch": 2.3353293413173652,
1122
+ "grad_norm": 0.283203125,
1123
+ "learning_rate": 0.00018598314098141206,
1124
+ "loss": 0.8122,
1125
+ "step": 1560
1126
+ },
1127
+ {
1128
+ "epoch": 2.3502994011976046,
1129
+ "grad_norm": 0.24609375,
1130
+ "learning_rate": 0.00018580379256528576,
1131
+ "loss": 0.7965,
1132
+ "step": 1570
1133
+ },
1134
+ {
1135
+ "epoch": 2.3652694610778444,
1136
+ "grad_norm": 0.279296875,
1137
+ "learning_rate": 0.00018562339165207936,
1138
+ "loss": 0.8309,
1139
+ "step": 1580
1140
+ },
1141
+ {
1142
+ "epoch": 2.3802395209580838,
1143
+ "grad_norm": 0.31640625,
1144
+ "learning_rate": 0.00018544194045464886,
1145
+ "loss": 0.8046,
1146
+ "step": 1590
1147
+ },
1148
+ {
1149
+ "epoch": 2.3952095808383236,
1150
+ "grad_norm": 0.2734375,
1151
+ "learning_rate": 0.0001852594411987334,
1152
+ "loss": 0.8467,
1153
+ "step": 1600
1154
+ },
1155
+ {
1156
+ "epoch": 2.410179640718563,
1157
+ "grad_norm": 0.275390625,
1158
+ "learning_rate": 0.00018507589612292783,
1159
+ "loss": 0.8566,
1160
+ "step": 1610
1161
+ },
1162
+ {
1163
+ "epoch": 2.4251497005988023,
1164
+ "grad_norm": 0.30859375,
1165
+ "learning_rate": 0.00018489130747865548,
1166
+ "loss": 0.8297,
1167
+ "step": 1620
1168
+ },
1169
+ {
1170
+ "epoch": 2.440119760479042,
1171
+ "grad_norm": 0.359375,
1172
+ "learning_rate": 0.00018470567753014035,
1173
+ "loss": 0.7823,
1174
+ "step": 1630
1175
+ },
1176
+ {
1177
+ "epoch": 2.4550898203592815,
1178
+ "grad_norm": 0.29296875,
1179
+ "learning_rate": 0.0001845190085543795,
1180
+ "loss": 0.8558,
1181
+ "step": 1640
1182
+ },
1183
+ {
1184
+ "epoch": 2.470059880239521,
1185
+ "grad_norm": 0.3046875,
1186
+ "learning_rate": 0.0001843313028411149,
1187
+ "loss": 0.8262,
1188
+ "step": 1650
1189
+ },
1190
+ {
1191
+ "epoch": 2.4850299401197606,
1192
+ "grad_norm": 0.2578125,
1193
+ "learning_rate": 0.00018414256269280564,
1194
+ "loss": 0.7982,
1195
+ "step": 1660
1196
+ },
1197
+ {
1198
+ "epoch": 2.5,
1199
+ "grad_norm": 0.314453125,
1200
+ "learning_rate": 0.00018395279042459937,
1201
+ "loss": 0.8182,
1202
+ "step": 1670
1203
+ },
1204
+ {
1205
+ "epoch": 2.5149700598802394,
1206
+ "grad_norm": 0.328125,
1207
+ "learning_rate": 0.00018376198836430417,
1208
+ "loss": 0.8275,
1209
+ "step": 1680
1210
+ },
1211
+ {
1212
+ "epoch": 2.529940119760479,
1213
+ "grad_norm": 0.271484375,
1214
+ "learning_rate": 0.00018357015885235982,
1215
+ "loss": 0.8102,
1216
+ "step": 1690
1217
+ },
1218
+ {
1219
+ "epoch": 2.5449101796407185,
1220
+ "grad_norm": 0.3125,
1221
+ "learning_rate": 0.0001833773042418092,
1222
+ "loss": 0.8145,
1223
+ "step": 1700
1224
+ },
1225
+ {
1226
+ "epoch": 2.5598802395209583,
1227
+ "grad_norm": 1.40625,
1228
+ "learning_rate": 0.00018318342689826938,
1229
+ "loss": 0.8279,
1230
+ "step": 1710
1231
+ },
1232
+ {
1233
+ "epoch": 2.5748502994011977,
1234
+ "grad_norm": 0.275390625,
1235
+ "learning_rate": 0.00018298852919990252,
1236
+ "loss": 0.8354,
1237
+ "step": 1720
1238
+ },
1239
+ {
1240
+ "epoch": 2.589820359281437,
1241
+ "grad_norm": 0.3125,
1242
+ "learning_rate": 0.0001827926135373869,
1243
+ "loss": 0.8106,
1244
+ "step": 1730
1245
+ },
1246
+ {
1247
+ "epoch": 2.6047904191616764,
1248
+ "grad_norm": 0.267578125,
1249
+ "learning_rate": 0.00018259568231388738,
1250
+ "loss": 0.7983,
1251
+ "step": 1740
1252
+ },
1253
+ {
1254
+ "epoch": 2.6197604790419162,
1255
+ "grad_norm": 0.251953125,
1256
+ "learning_rate": 0.00018239773794502607,
1257
+ "loss": 0.8183,
1258
+ "step": 1750
1259
+ },
1260
+ {
1261
+ "epoch": 2.6347305389221556,
1262
+ "grad_norm": 0.294921875,
1263
+ "learning_rate": 0.00018219878285885267,
1264
+ "loss": 0.8462,
1265
+ "step": 1760
1266
+ },
1267
+ {
1268
+ "epoch": 2.6497005988023954,
1269
+ "grad_norm": 0.283203125,
1270
+ "learning_rate": 0.0001819988194958146,
1271
+ "loss": 0.8227,
1272
+ "step": 1770
1273
+ },
1274
+ {
1275
+ "epoch": 2.6646706586826348,
1276
+ "grad_norm": 0.2431640625,
1277
+ "learning_rate": 0.0001817978503087272,
1278
+ "loss": 0.8104,
1279
+ "step": 1780
1280
+ },
1281
+ {
1282
+ "epoch": 2.679640718562874,
1283
+ "grad_norm": 0.271484375,
1284
+ "learning_rate": 0.0001815958777627435,
1285
+ "loss": 0.7972,
1286
+ "step": 1790
1287
+ },
1288
+ {
1289
+ "epoch": 2.694610778443114,
1290
+ "grad_norm": 0.283203125,
1291
+ "learning_rate": 0.00018139290433532416,
1292
+ "loss": 0.8339,
1293
+ "step": 1800
1294
+ },
1295
+ {
1296
+ "epoch": 2.7095808383233533,
1297
+ "grad_norm": 0.298828125,
1298
+ "learning_rate": 0.00018118893251620682,
1299
+ "loss": 0.8723,
1300
+ "step": 1810
1301
+ },
1302
+ {
1303
+ "epoch": 2.724550898203593,
1304
+ "grad_norm": 0.3203125,
1305
+ "learning_rate": 0.00018098396480737585,
1306
+ "loss": 0.8544,
1307
+ "step": 1820
1308
+ },
1309
+ {
1310
+ "epoch": 2.7395209580838324,
1311
+ "grad_norm": 0.265625,
1312
+ "learning_rate": 0.0001807780037230315,
1313
+ "loss": 0.8557,
1314
+ "step": 1830
1315
+ },
1316
+ {
1317
+ "epoch": 2.754491017964072,
1318
+ "grad_norm": 0.279296875,
1319
+ "learning_rate": 0.00018057105178955905,
1320
+ "loss": 0.8283,
1321
+ "step": 1840
1322
+ },
1323
+ {
1324
+ "epoch": 2.769461077844311,
1325
+ "grad_norm": 0.3828125,
1326
+ "learning_rate": 0.00018036311154549784,
1327
+ "loss": 0.7906,
1328
+ "step": 1850
1329
+ },
1330
+ {
1331
+ "epoch": 2.784431137724551,
1332
+ "grad_norm": 0.2490234375,
1333
+ "learning_rate": 0.0001801541855415102,
1334
+ "loss": 0.8036,
1335
+ "step": 1860
1336
+ },
1337
+ {
1338
+ "epoch": 2.7994011976047903,
1339
+ "grad_norm": 0.298828125,
1340
+ "learning_rate": 0.00017994427634035015,
1341
+ "loss": 0.7828,
1342
+ "step": 1870
1343
+ },
1344
+ {
1345
+ "epoch": 2.81437125748503,
1346
+ "grad_norm": 0.306640625,
1347
+ "learning_rate": 0.00017973338651683176,
1348
+ "loss": 0.7915,
1349
+ "step": 1880
1350
+ },
1351
+ {
1352
+ "epoch": 2.8293413173652695,
1353
+ "grad_norm": 0.228515625,
1354
+ "learning_rate": 0.00017952151865779792,
1355
+ "loss": 0.8141,
1356
+ "step": 1890
1357
+ },
1358
+ {
1359
+ "epoch": 2.844311377245509,
1360
+ "grad_norm": 0.349609375,
1361
+ "learning_rate": 0.00017930867536208826,
1362
+ "loss": 0.8155,
1363
+ "step": 1900
1364
+ },
1365
+ {
1366
+ "epoch": 2.8592814371257482,
1367
+ "grad_norm": 0.310546875,
1368
+ "learning_rate": 0.00017909485924050758,
1369
+ "loss": 0.8004,
1370
+ "step": 1910
1371
+ },
1372
+ {
1373
+ "epoch": 2.874251497005988,
1374
+ "grad_norm": 0.2470703125,
1375
+ "learning_rate": 0.00017888007291579357,
1376
+ "loss": 0.803,
1377
+ "step": 1920
1378
+ },
1379
+ {
1380
+ "epoch": 2.8892215568862274,
1381
+ "grad_norm": 0.275390625,
1382
+ "learning_rate": 0.00017866431902258478,
1383
+ "loss": 0.804,
1384
+ "step": 1930
1385
+ },
1386
+ {
1387
+ "epoch": 2.904191616766467,
1388
+ "grad_norm": 0.310546875,
1389
+ "learning_rate": 0.00017844760020738827,
1390
+ "loss": 0.8154,
1391
+ "step": 1940
1392
+ },
1393
+ {
1394
+ "epoch": 2.9191616766467066,
1395
+ "grad_norm": 0.251953125,
1396
+ "learning_rate": 0.00017822991912854713,
1397
+ "loss": 0.8257,
1398
+ "step": 1950
1399
+ },
1400
+ {
1401
+ "epoch": 2.934131736526946,
1402
+ "grad_norm": 0.3125,
1403
+ "learning_rate": 0.00017801127845620793,
1404
+ "loss": 0.8386,
1405
+ "step": 1960
1406
+ },
1407
+ {
1408
+ "epoch": 2.9491017964071857,
1409
+ "grad_norm": 0.275390625,
1410
+ "learning_rate": 0.0001777916808722879,
1411
+ "loss": 0.8003,
1412
+ "step": 1970
1413
+ },
1414
+ {
1415
+ "epoch": 2.964071856287425,
1416
+ "grad_norm": 0.28515625,
1417
+ "learning_rate": 0.000177571129070442,
1418
+ "loss": 0.8055,
1419
+ "step": 1980
1420
+ },
1421
+ {
1422
+ "epoch": 2.979041916167665,
1423
+ "grad_norm": 0.271484375,
1424
+ "learning_rate": 0.00017734962575603,
1425
+ "loss": 0.8233,
1426
+ "step": 1990
1427
+ },
1428
+ {
1429
+ "epoch": 2.9940119760479043,
1430
+ "grad_norm": 0.255859375,
1431
+ "learning_rate": 0.00017712717364608328,
1432
+ "loss": 0.8106,
1433
+ "step": 2000
1434
+ },
1435
+ {
1436
+ "epoch": 2.9940119760479043,
1437
+ "eval_loss": 0.6911507844924927,
1438
+ "eval_runtime": 109.6711,
1439
+ "eval_samples_per_second": 9.118,
1440
+ "eval_steps_per_second": 1.14,
1441
+ "step": 2000
1442
+ },
1443
+ {
1444
+ "epoch": 3.0089820359281436,
1445
+ "grad_norm": 0.37890625,
1446
+ "learning_rate": 0.00017690377546927133,
1447
+ "loss": 0.7276,
1448
+ "step": 2010
1449
+ },
1450
+ {
1451
+ "epoch": 3.0239520958083834,
1452
+ "grad_norm": 0.27734375,
1453
+ "learning_rate": 0.00017667943396586848,
1454
+ "loss": 0.6882,
1455
+ "step": 2020
1456
+ },
1457
+ {
1458
+ "epoch": 3.038922155688623,
1459
+ "grad_norm": 0.291015625,
1460
+ "learning_rate": 0.0001764541518877202,
1461
+ "loss": 0.6786,
1462
+ "step": 2030
1463
+ },
1464
+ {
1465
+ "epoch": 3.053892215568862,
1466
+ "grad_norm": 0.31640625,
1467
+ "learning_rate": 0.00017622793199820934,
1468
+ "loss": 0.6352,
1469
+ "step": 2040
1470
+ },
1471
+ {
1472
+ "epoch": 3.068862275449102,
1473
+ "grad_norm": 0.294921875,
1474
+ "learning_rate": 0.00017600077707222224,
1475
+ "loss": 0.6648,
1476
+ "step": 2050
1477
+ },
1478
+ {
1479
+ "epoch": 3.0838323353293413,
1480
+ "grad_norm": 0.294921875,
1481
+ "learning_rate": 0.00017577268989611472,
1482
+ "loss": 0.6601,
1483
+ "step": 2060
1484
+ },
1485
+ {
1486
+ "epoch": 3.0988023952095807,
1487
+ "grad_norm": 0.361328125,
1488
+ "learning_rate": 0.00017554367326767792,
1489
+ "loss": 0.6645,
1490
+ "step": 2070
1491
+ },
1492
+ {
1493
+ "epoch": 3.1137724550898205,
1494
+ "grad_norm": 0.251953125,
1495
+ "learning_rate": 0.00017531372999610384,
1496
+ "loss": 0.6631,
1497
+ "step": 2080
1498
+ },
1499
+ {
1500
+ "epoch": 3.12874251497006,
1501
+ "grad_norm": 0.30078125,
1502
+ "learning_rate": 0.0001750828629019511,
1503
+ "loss": 0.6451,
1504
+ "step": 2090
1505
+ },
1506
+ {
1507
+ "epoch": 3.143712574850299,
1508
+ "grad_norm": 0.365234375,
1509
+ "learning_rate": 0.00017485107481711012,
1510
+ "loss": 0.6703,
1511
+ "step": 2100
1512
+ },
1513
+ {
1514
+ "epoch": 3.158682634730539,
1515
+ "grad_norm": 0.341796875,
1516
+ "learning_rate": 0.00017461836858476856,
1517
+ "loss": 0.6598,
1518
+ "step": 2110
1519
+ },
1520
+ {
1521
+ "epoch": 3.1736526946107784,
1522
+ "grad_norm": 0.287109375,
1523
+ "learning_rate": 0.00017438474705937639,
1524
+ "loss": 0.6689,
1525
+ "step": 2120
1526
+ },
1527
+ {
1528
+ "epoch": 3.1886227544910177,
1529
+ "grad_norm": 0.3125,
1530
+ "learning_rate": 0.0001741502131066107,
1531
+ "loss": 0.6741,
1532
+ "step": 2130
1533
+ },
1534
+ {
1535
+ "epoch": 3.2035928143712575,
1536
+ "grad_norm": 0.31640625,
1537
+ "learning_rate": 0.000173914769603341,
1538
+ "loss": 0.6691,
1539
+ "step": 2140
1540
+ },
1541
+ {
1542
+ "epoch": 3.218562874251497,
1543
+ "grad_norm": 0.322265625,
1544
+ "learning_rate": 0.00017367841943759338,
1545
+ "loss": 0.6702,
1546
+ "step": 2150
1547
+ },
1548
+ {
1549
+ "epoch": 3.2335329341317367,
1550
+ "grad_norm": 0.265625,
1551
+ "learning_rate": 0.00017344116550851543,
1552
+ "loss": 0.6451,
1553
+ "step": 2160
1554
+ },
1555
+ {
1556
+ "epoch": 3.248502994011976,
1557
+ "grad_norm": 0.328125,
1558
+ "learning_rate": 0.00017320301072634066,
1559
+ "loss": 0.6507,
1560
+ "step": 2170
1561
+ },
1562
+ {
1563
+ "epoch": 3.2634730538922154,
1564
+ "grad_norm": 0.32421875,
1565
+ "learning_rate": 0.00017296395801235265,
1566
+ "loss": 0.695,
1567
+ "step": 2180
1568
+ },
1569
+ {
1570
+ "epoch": 3.2784431137724552,
1571
+ "grad_norm": 0.31640625,
1572
+ "learning_rate": 0.00017272401029884933,
1573
+ "loss": 0.6798,
1574
+ "step": 2190
1575
+ },
1576
+ {
1577
+ "epoch": 3.2934131736526946,
1578
+ "grad_norm": 0.330078125,
1579
+ "learning_rate": 0.000172483170529107,
1580
+ "loss": 0.6526,
1581
+ "step": 2200
1582
+ },
1583
+ {
1584
+ "epoch": 3.308383233532934,
1585
+ "grad_norm": 0.35546875,
1586
+ "learning_rate": 0.00017224144165734417,
1587
+ "loss": 0.6538,
1588
+ "step": 2210
1589
+ },
1590
+ {
1591
+ "epoch": 3.3233532934131738,
1592
+ "grad_norm": 0.291015625,
1593
+ "learning_rate": 0.00017199882664868538,
1594
+ "loss": 0.6777,
1595
+ "step": 2220
1596
+ },
1597
+ {
1598
+ "epoch": 3.338323353293413,
1599
+ "grad_norm": 0.28515625,
1600
+ "learning_rate": 0.00017175532847912487,
1601
+ "loss": 0.6762,
1602
+ "step": 2230
1603
+ },
1604
+ {
1605
+ "epoch": 3.3532934131736525,
1606
+ "grad_norm": 0.3203125,
1607
+ "learning_rate": 0.00017151095013548994,
1608
+ "loss": 0.673,
1609
+ "step": 2240
1610
+ },
1611
+ {
1612
+ "epoch": 3.3682634730538923,
1613
+ "grad_norm": 0.318359375,
1614
+ "learning_rate": 0.00017126569461540443,
1615
+ "loss": 0.6757,
1616
+ "step": 2250
1617
+ },
1618
+ {
1619
+ "epoch": 3.3832335329341316,
1620
+ "grad_norm": 0.3046875,
1621
+ "learning_rate": 0.00017101956492725185,
1622
+ "loss": 0.6563,
1623
+ "step": 2260
1624
+ },
1625
+ {
1626
+ "epoch": 3.3982035928143715,
1627
+ "grad_norm": 0.267578125,
1628
+ "learning_rate": 0.00017077256409013866,
1629
+ "loss": 0.6877,
1630
+ "step": 2270
1631
+ },
1632
+ {
1633
+ "epoch": 3.413173652694611,
1634
+ "grad_norm": 0.328125,
1635
+ "learning_rate": 0.000170524695133857,
1636
+ "loss": 0.67,
1637
+ "step": 2280
1638
+ },
1639
+ {
1640
+ "epoch": 3.42814371257485,
1641
+ "grad_norm": 0.326171875,
1642
+ "learning_rate": 0.00017027596109884768,
1643
+ "loss": 0.6808,
1644
+ "step": 2290
1645
+ },
1646
+ {
1647
+ "epoch": 3.44311377245509,
1648
+ "grad_norm": 0.318359375,
1649
+ "learning_rate": 0.00017002636503616282,
1650
+ "loss": 0.6941,
1651
+ "step": 2300
1652
+ },
1653
+ {
1654
+ "epoch": 3.4580838323353293,
1655
+ "grad_norm": 0.279296875,
1656
+ "learning_rate": 0.00016977591000742854,
1657
+ "loss": 0.6798,
1658
+ "step": 2310
1659
+ },
1660
+ {
1661
+ "epoch": 3.4730538922155687,
1662
+ "grad_norm": 0.287109375,
1663
+ "learning_rate": 0.0001695245990848072,
1664
+ "loss": 0.6718,
1665
+ "step": 2320
1666
+ },
1667
+ {
1668
+ "epoch": 3.4880239520958085,
1669
+ "grad_norm": 0.30859375,
1670
+ "learning_rate": 0.00016927243535095997,
1671
+ "loss": 0.6483,
1672
+ "step": 2330
1673
+ },
1674
+ {
1675
+ "epoch": 3.502994011976048,
1676
+ "grad_norm": 0.333984375,
1677
+ "learning_rate": 0.00016901942189900867,
1678
+ "loss": 0.7177,
1679
+ "step": 2340
1680
+ },
1681
+ {
1682
+ "epoch": 3.5179640718562872,
1683
+ "grad_norm": 0.34765625,
1684
+ "learning_rate": 0.00016876556183249822,
1685
+ "loss": 0.6833,
1686
+ "step": 2350
1687
+ },
1688
+ {
1689
+ "epoch": 3.532934131736527,
1690
+ "grad_norm": 0.365234375,
1691
+ "learning_rate": 0.00016851085826535838,
1692
+ "loss": 0.6826,
1693
+ "step": 2360
1694
+ },
1695
+ {
1696
+ "epoch": 3.5479041916167664,
1697
+ "grad_norm": 0.396484375,
1698
+ "learning_rate": 0.00016825531432186543,
1699
+ "loss": 0.6835,
1700
+ "step": 2370
1701
+ },
1702
+ {
1703
+ "epoch": 3.562874251497006,
1704
+ "grad_norm": 0.359375,
1705
+ "learning_rate": 0.00016799893313660408,
1706
+ "loss": 0.6791,
1707
+ "step": 2380
1708
+ },
1709
+ {
1710
+ "epoch": 3.5778443113772456,
1711
+ "grad_norm": 0.291015625,
1712
+ "learning_rate": 0.0001677417178544289,
1713
+ "loss": 0.6787,
1714
+ "step": 2390
1715
+ },
1716
+ {
1717
+ "epoch": 3.592814371257485,
1718
+ "grad_norm": 0.296875,
1719
+ "learning_rate": 0.00016748367163042576,
1720
+ "loss": 0.6542,
1721
+ "step": 2400
1722
+ },
1723
+ {
1724
+ "epoch": 3.6077844311377243,
1725
+ "grad_norm": 0.30078125,
1726
+ "learning_rate": 0.00016722479762987317,
1727
+ "loss": 0.6805,
1728
+ "step": 2410
1729
+ },
1730
+ {
1731
+ "epoch": 3.622754491017964,
1732
+ "grad_norm": 0.314453125,
1733
+ "learning_rate": 0.0001669650990282033,
1734
+ "loss": 0.6859,
1735
+ "step": 2420
1736
+ },
1737
+ {
1738
+ "epoch": 3.6377245508982035,
1739
+ "grad_norm": 0.32421875,
1740
+ "learning_rate": 0.00016670457901096328,
1741
+ "loss": 0.6633,
1742
+ "step": 2430
1743
+ },
1744
+ {
1745
+ "epoch": 3.6526946107784433,
1746
+ "grad_norm": 0.32421875,
1747
+ "learning_rate": 0.00016644324077377592,
1748
+ "loss": 0.6958,
1749
+ "step": 2440
1750
+ },
1751
+ {
1752
+ "epoch": 3.6676646706586826,
1753
+ "grad_norm": 0.3125,
1754
+ "learning_rate": 0.00016618108752230052,
1755
+ "loss": 0.6965,
1756
+ "step": 2450
1757
+ },
1758
+ {
1759
+ "epoch": 3.682634730538922,
1760
+ "grad_norm": 0.3046875,
1761
+ "learning_rate": 0.00016591812247219377,
1762
+ "loss": 0.6851,
1763
+ "step": 2460
1764
+ },
1765
+ {
1766
+ "epoch": 3.697604790419162,
1767
+ "grad_norm": 0.400390625,
1768
+ "learning_rate": 0.00016565434884907002,
1769
+ "loss": 0.6669,
1770
+ "step": 2470
1771
+ },
1772
+ {
1773
+ "epoch": 3.712574850299401,
1774
+ "grad_norm": 0.306640625,
1775
+ "learning_rate": 0.0001653897698884619,
1776
+ "loss": 0.6672,
1777
+ "step": 2480
1778
+ },
1779
+ {
1780
+ "epoch": 3.727544910179641,
1781
+ "grad_norm": 0.30078125,
1782
+ "learning_rate": 0.00016512438883578044,
1783
+ "loss": 0.7049,
1784
+ "step": 2490
1785
+ },
1786
+ {
1787
+ "epoch": 3.7425149700598803,
1788
+ "grad_norm": 0.30859375,
1789
+ "learning_rate": 0.0001648582089462756,
1790
+ "loss": 0.6821,
1791
+ "step": 2500
1792
+ },
1793
+ {
1794
+ "epoch": 3.7425149700598803,
1795
+ "eval_loss": 0.5838193893432617,
1796
+ "eval_runtime": 109.8377,
1797
+ "eval_samples_per_second": 9.104,
1798
+ "eval_steps_per_second": 1.138,
1799
+ "step": 2500
1800
+ }
1801
+ ],
1802
+ "logging_steps": 10,
1803
+ "max_steps": 9000,
1804
+ "num_input_tokens_seen": 0,
1805
+ "num_train_epochs": 14,
1806
+ "save_steps": 500,
1807
+ "stateful_callbacks": {
1808
+ "TrainerControl": {
1809
+ "args": {
1810
+ "should_epoch_stop": false,
1811
+ "should_evaluate": false,
1812
+ "should_log": false,
1813
+ "should_save": true,
1814
+ "should_training_stop": false
1815
+ },
1816
+ "attributes": {}
1817
+ }
1818
+ },
1819
+ "total_flos": 1.7382403865051136e+17,
1820
+ "train_batch_size": 16,
1821
+ "trial_name": null,
1822
+ "trial_params": null
1823
+ }
checkpoint-2500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-3000/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-3000/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-3000/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:934de6ff7a1b01db74ef49b5e5e6f4d53426ee825586e588cbe50d8e2bbc3569
3
+ size 802337544
checkpoint-3000/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b91b3b7e32d3071cc3dfde4214de43638e048a379f2ae76999ec2e7fed6a12f2
3
+ size 2407189434
checkpoint-3000/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:31d69d8d5303b7ea25dd0db67bae0f13352c8bdadb4cb9c5111205c24fdcf26f
3
+ size 13990
checkpoint-3000/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:75f3beb3f58cdbe1a0ab21470e56d6a585cfcabbe063c7a960acbddbe0d6e3b2
3
+ size 1000
checkpoint-3000/trainer_state.json ADDED
@@ -0,0 +1,2181 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.49006855487823486,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-3000",
4
+ "epoch": 4.491017964071856,
5
+ "eval_steps": 500,
6
+ "global_step": 3000,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 1.5119760479041915,
729
+ "grad_norm": 0.2890625,
730
+ "learning_rate": 0.00019416730477082533,
731
+ "loss": 0.9444,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 1.5269461077844313,
736
+ "grad_norm": 0.296875,
737
+ "learning_rate": 0.00019404886375423984,
738
+ "loss": 0.9829,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 1.5419161676646707,
743
+ "grad_norm": 0.27734375,
744
+ "learning_rate": 0.00019392926910386353,
745
+ "loss": 0.9532,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 1.55688622754491,
750
+ "grad_norm": 0.265625,
751
+ "learning_rate": 0.00019380852228668304,
752
+ "loss": 0.9769,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 1.5718562874251498,
757
+ "grad_norm": 0.275390625,
758
+ "learning_rate": 0.00019368662478381799,
759
+ "loss": 0.9783,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 1.5868263473053892,
764
+ "grad_norm": 0.2890625,
765
+ "learning_rate": 0.00019356357809050247,
766
+ "loss": 0.9881,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 1.6017964071856288,
771
+ "grad_norm": 0.232421875,
772
+ "learning_rate": 0.00019343938371606712,
773
+ "loss": 0.9883,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 1.6167664670658684,
778
+ "grad_norm": 0.2734375,
779
+ "learning_rate": 0.00019331404318392027,
780
+ "loss": 0.9893,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 1.6317365269461077,
785
+ "grad_norm": 0.248046875,
786
+ "learning_rate": 0.00019318755803152945,
787
+ "loss": 0.9851,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 1.6467065868263473,
792
+ "grad_norm": 0.28515625,
793
+ "learning_rate": 0.00019305992981040246,
794
+ "loss": 0.9531,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 1.6616766467065869,
799
+ "grad_norm": 0.275390625,
800
+ "learning_rate": 0.00019293116008606837,
801
+ "loss": 0.9717,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 1.6766467065868262,
806
+ "grad_norm": 0.232421875,
807
+ "learning_rate": 0.00019280125043805824,
808
+ "loss": 0.9699,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 1.6916167664670658,
813
+ "grad_norm": 0.2294921875,
814
+ "learning_rate": 0.00019267020245988592,
815
+ "loss": 0.9407,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 1.7065868263473054,
820
+ "grad_norm": 0.251953125,
821
+ "learning_rate": 0.00019253801775902824,
822
+ "loss": 0.977,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 1.7215568862275448,
827
+ "grad_norm": 0.287109375,
828
+ "learning_rate": 0.0001924046979569055,
829
+ "loss": 0.9549,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 1.7365269461077846,
834
+ "grad_norm": 0.248046875,
835
+ "learning_rate": 0.00019227024468886157,
836
+ "loss": 0.9824,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 1.751497005988024,
841
+ "grad_norm": 0.259765625,
842
+ "learning_rate": 0.00019213465960414368,
843
+ "loss": 0.9936,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 1.7664670658682635,
848
+ "grad_norm": 0.29296875,
849
+ "learning_rate": 0.00019199794436588243,
850
+ "loss": 1.0042,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 1.781437125748503,
855
+ "grad_norm": 0.23828125,
856
+ "learning_rate": 0.0001918601006510711,
857
+ "loss": 0.9629,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 1.7964071856287425,
862
+ "grad_norm": 0.2890625,
863
+ "learning_rate": 0.00019172113015054532,
864
+ "loss": 0.9522,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 1.811377245508982,
869
+ "grad_norm": 0.26953125,
870
+ "learning_rate": 0.0001915810345689622,
871
+ "loss": 0.9806,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 1.8263473053892216,
876
+ "grad_norm": 0.263671875,
877
+ "learning_rate": 0.00019143981562477947,
878
+ "loss": 0.9736,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 1.841317365269461,
883
+ "grad_norm": 0.2578125,
884
+ "learning_rate": 0.00019129747505023436,
885
+ "loss": 0.9701,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 1.8562874251497006,
890
+ "grad_norm": 0.22265625,
891
+ "learning_rate": 0.00019115401459132247,
892
+ "loss": 0.9494,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 1.8712574850299402,
897
+ "grad_norm": 0.251953125,
898
+ "learning_rate": 0.00019100943600777615,
899
+ "loss": 0.9922,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 1.8862275449101795,
904
+ "grad_norm": 0.2490234375,
905
+ "learning_rate": 0.00019086374107304312,
906
+ "loss": 0.9711,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 1.9011976047904193,
911
+ "grad_norm": 0.25390625,
912
+ "learning_rate": 0.00019071693157426457,
913
+ "loss": 0.9664,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 1.9161676646706587,
918
+ "grad_norm": 0.26171875,
919
+ "learning_rate": 0.00019056900931225333,
920
+ "loss": 0.9591,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 1.931137724550898,
925
+ "grad_norm": 0.380859375,
926
+ "learning_rate": 0.00019041997610147167,
927
+ "loss": 0.9942,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 1.9461077844311379,
932
+ "grad_norm": 0.2294921875,
933
+ "learning_rate": 0.0001902698337700092,
934
+ "loss": 0.9391,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 1.9610778443113772,
939
+ "grad_norm": 0.298828125,
940
+ "learning_rate": 0.00019011858415956038,
941
+ "loss": 0.9993,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 1.9760479041916168,
946
+ "grad_norm": 0.29296875,
947
+ "learning_rate": 0.0001899662291254018,
948
+ "loss": 0.9571,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 1.9910179640718564,
953
+ "grad_norm": 0.283203125,
954
+ "learning_rate": 0.0001898127705363696,
955
+ "loss": 0.9835,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 2.0059880239520957,
960
+ "grad_norm": 0.27734375,
961
+ "learning_rate": 0.00018965821027483654,
962
+ "loss": 0.9305,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 2.020958083832335,
967
+ "grad_norm": 0.26171875,
968
+ "learning_rate": 0.00018950255023668876,
969
+ "loss": 0.8295,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 2.035928143712575,
974
+ "grad_norm": 0.240234375,
975
+ "learning_rate": 0.00018934579233130267,
976
+ "loss": 0.7653,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 2.0508982035928143,
981
+ "grad_norm": 0.29296875,
982
+ "learning_rate": 0.00018918793848152142,
983
+ "loss": 0.7581,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 2.065868263473054,
988
+ "grad_norm": 0.29296875,
989
+ "learning_rate": 0.00018902899062363143,
990
+ "loss": 0.7983,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 2.0808383233532934,
995
+ "grad_norm": 0.279296875,
996
+ "learning_rate": 0.0001888689507073385,
997
+ "loss": 0.8187,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 2.095808383233533,
1002
+ "grad_norm": 0.29296875,
1003
+ "learning_rate": 0.0001887078206957441,
1004
+ "loss": 0.7879,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 2.1107784431137726,
1009
+ "grad_norm": 0.2734375,
1010
+ "learning_rate": 0.000188545602565321,
1011
+ "loss": 0.8083,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 2.125748502994012,
1016
+ "grad_norm": 0.263671875,
1017
+ "learning_rate": 0.00018838229830588934,
1018
+ "loss": 0.8057,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 2.1407185628742513,
1023
+ "grad_norm": 0.244140625,
1024
+ "learning_rate": 0.00018821790992059196,
1025
+ "loss": 0.8194,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 2.155688622754491,
1030
+ "grad_norm": 0.2451171875,
1031
+ "learning_rate": 0.00018805243942587,
1032
+ "loss": 0.7958,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 2.1706586826347305,
1037
+ "grad_norm": 0.248046875,
1038
+ "learning_rate": 0.00018788588885143808,
1039
+ "loss": 0.8169,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 2.18562874251497,
1044
+ "grad_norm": 0.265625,
1045
+ "learning_rate": 0.00018771826024025946,
1046
+ "loss": 0.7722,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 2.2005988023952097,
1051
+ "grad_norm": 0.265625,
1052
+ "learning_rate": 0.0001875495556485208,
1053
+ "loss": 0.7934,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 2.215568862275449,
1058
+ "grad_norm": 0.244140625,
1059
+ "learning_rate": 0.00018737977714560738,
1060
+ "loss": 0.7915,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 2.230538922155689,
1065
+ "grad_norm": 0.310546875,
1066
+ "learning_rate": 0.00018720892681407708,
1067
+ "loss": 0.8021,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 2.245508982035928,
1072
+ "grad_norm": 0.298828125,
1073
+ "learning_rate": 0.00018703700674963547,
1074
+ "loss": 0.7987,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 2.245508982035928,
1079
+ "eval_loss": 0.7798940539360046,
1080
+ "eval_runtime": 109.62,
1081
+ "eval_samples_per_second": 9.122,
1082
+ "eval_steps_per_second": 1.14,
1083
+ "step": 1500
1084
+ },
1085
+ {
1086
+ "epoch": 2.2604790419161676,
1087
+ "grad_norm": 0.259765625,
1088
+ "learning_rate": 0.00018686401906110964,
1089
+ "loss": 0.7979,
1090
+ "step": 1510
1091
+ },
1092
+ {
1093
+ "epoch": 2.2754491017964074,
1094
+ "grad_norm": 0.265625,
1095
+ "learning_rate": 0.00018668996587042252,
1096
+ "loss": 0.8255,
1097
+ "step": 1520
1098
+ },
1099
+ {
1100
+ "epoch": 2.2904191616766467,
1101
+ "grad_norm": 0.2470703125,
1102
+ "learning_rate": 0.00018651484931256685,
1103
+ "loss": 0.8252,
1104
+ "step": 1530
1105
+ },
1106
+ {
1107
+ "epoch": 2.305389221556886,
1108
+ "grad_norm": 0.25,
1109
+ "learning_rate": 0.00018633867153557905,
1110
+ "loss": 0.8455,
1111
+ "step": 1540
1112
+ },
1113
+ {
1114
+ "epoch": 2.320359281437126,
1115
+ "grad_norm": 0.302734375,
1116
+ "learning_rate": 0.00018616143470051263,
1117
+ "loss": 0.8118,
1118
+ "step": 1550
1119
+ },
1120
+ {
1121
+ "epoch": 2.3353293413173652,
1122
+ "grad_norm": 0.283203125,
1123
+ "learning_rate": 0.00018598314098141206,
1124
+ "loss": 0.8122,
1125
+ "step": 1560
1126
+ },
1127
+ {
1128
+ "epoch": 2.3502994011976046,
1129
+ "grad_norm": 0.24609375,
1130
+ "learning_rate": 0.00018580379256528576,
1131
+ "loss": 0.7965,
1132
+ "step": 1570
1133
+ },
1134
+ {
1135
+ "epoch": 2.3652694610778444,
1136
+ "grad_norm": 0.279296875,
1137
+ "learning_rate": 0.00018562339165207936,
1138
+ "loss": 0.8309,
1139
+ "step": 1580
1140
+ },
1141
+ {
1142
+ "epoch": 2.3802395209580838,
1143
+ "grad_norm": 0.31640625,
1144
+ "learning_rate": 0.00018544194045464886,
1145
+ "loss": 0.8046,
1146
+ "step": 1590
1147
+ },
1148
+ {
1149
+ "epoch": 2.3952095808383236,
1150
+ "grad_norm": 0.2734375,
1151
+ "learning_rate": 0.0001852594411987334,
1152
+ "loss": 0.8467,
1153
+ "step": 1600
1154
+ },
1155
+ {
1156
+ "epoch": 2.410179640718563,
1157
+ "grad_norm": 0.275390625,
1158
+ "learning_rate": 0.00018507589612292783,
1159
+ "loss": 0.8566,
1160
+ "step": 1610
1161
+ },
1162
+ {
1163
+ "epoch": 2.4251497005988023,
1164
+ "grad_norm": 0.30859375,
1165
+ "learning_rate": 0.00018489130747865548,
1166
+ "loss": 0.8297,
1167
+ "step": 1620
1168
+ },
1169
+ {
1170
+ "epoch": 2.440119760479042,
1171
+ "grad_norm": 0.359375,
1172
+ "learning_rate": 0.00018470567753014035,
1173
+ "loss": 0.7823,
1174
+ "step": 1630
1175
+ },
1176
+ {
1177
+ "epoch": 2.4550898203592815,
1178
+ "grad_norm": 0.29296875,
1179
+ "learning_rate": 0.0001845190085543795,
1180
+ "loss": 0.8558,
1181
+ "step": 1640
1182
+ },
1183
+ {
1184
+ "epoch": 2.470059880239521,
1185
+ "grad_norm": 0.3046875,
1186
+ "learning_rate": 0.0001843313028411149,
1187
+ "loss": 0.8262,
1188
+ "step": 1650
1189
+ },
1190
+ {
1191
+ "epoch": 2.4850299401197606,
1192
+ "grad_norm": 0.2578125,
1193
+ "learning_rate": 0.00018414256269280564,
1194
+ "loss": 0.7982,
1195
+ "step": 1660
1196
+ },
1197
+ {
1198
+ "epoch": 2.5,
1199
+ "grad_norm": 0.314453125,
1200
+ "learning_rate": 0.00018395279042459937,
1201
+ "loss": 0.8182,
1202
+ "step": 1670
1203
+ },
1204
+ {
1205
+ "epoch": 2.5149700598802394,
1206
+ "grad_norm": 0.328125,
1207
+ "learning_rate": 0.00018376198836430417,
1208
+ "loss": 0.8275,
1209
+ "step": 1680
1210
+ },
1211
+ {
1212
+ "epoch": 2.529940119760479,
1213
+ "grad_norm": 0.271484375,
1214
+ "learning_rate": 0.00018357015885235982,
1215
+ "loss": 0.8102,
1216
+ "step": 1690
1217
+ },
1218
+ {
1219
+ "epoch": 2.5449101796407185,
1220
+ "grad_norm": 0.3125,
1221
+ "learning_rate": 0.0001833773042418092,
1222
+ "loss": 0.8145,
1223
+ "step": 1700
1224
+ },
1225
+ {
1226
+ "epoch": 2.5598802395209583,
1227
+ "grad_norm": 1.40625,
1228
+ "learning_rate": 0.00018318342689826938,
1229
+ "loss": 0.8279,
1230
+ "step": 1710
1231
+ },
1232
+ {
1233
+ "epoch": 2.5748502994011977,
1234
+ "grad_norm": 0.275390625,
1235
+ "learning_rate": 0.00018298852919990252,
1236
+ "loss": 0.8354,
1237
+ "step": 1720
1238
+ },
1239
+ {
1240
+ "epoch": 2.589820359281437,
1241
+ "grad_norm": 0.3125,
1242
+ "learning_rate": 0.0001827926135373869,
1243
+ "loss": 0.8106,
1244
+ "step": 1730
1245
+ },
1246
+ {
1247
+ "epoch": 2.6047904191616764,
1248
+ "grad_norm": 0.267578125,
1249
+ "learning_rate": 0.00018259568231388738,
1250
+ "loss": 0.7983,
1251
+ "step": 1740
1252
+ },
1253
+ {
1254
+ "epoch": 2.6197604790419162,
1255
+ "grad_norm": 0.251953125,
1256
+ "learning_rate": 0.00018239773794502607,
1257
+ "loss": 0.8183,
1258
+ "step": 1750
1259
+ },
1260
+ {
1261
+ "epoch": 2.6347305389221556,
1262
+ "grad_norm": 0.294921875,
1263
+ "learning_rate": 0.00018219878285885267,
1264
+ "loss": 0.8462,
1265
+ "step": 1760
1266
+ },
1267
+ {
1268
+ "epoch": 2.6497005988023954,
1269
+ "grad_norm": 0.283203125,
1270
+ "learning_rate": 0.0001819988194958146,
1271
+ "loss": 0.8227,
1272
+ "step": 1770
1273
+ },
1274
+ {
1275
+ "epoch": 2.6646706586826348,
1276
+ "grad_norm": 0.2431640625,
1277
+ "learning_rate": 0.0001817978503087272,
1278
+ "loss": 0.8104,
1279
+ "step": 1780
1280
+ },
1281
+ {
1282
+ "epoch": 2.679640718562874,
1283
+ "grad_norm": 0.271484375,
1284
+ "learning_rate": 0.0001815958777627435,
1285
+ "loss": 0.7972,
1286
+ "step": 1790
1287
+ },
1288
+ {
1289
+ "epoch": 2.694610778443114,
1290
+ "grad_norm": 0.283203125,
1291
+ "learning_rate": 0.00018139290433532416,
1292
+ "loss": 0.8339,
1293
+ "step": 1800
1294
+ },
1295
+ {
1296
+ "epoch": 2.7095808383233533,
1297
+ "grad_norm": 0.298828125,
1298
+ "learning_rate": 0.00018118893251620682,
1299
+ "loss": 0.8723,
1300
+ "step": 1810
1301
+ },
1302
+ {
1303
+ "epoch": 2.724550898203593,
1304
+ "grad_norm": 0.3203125,
1305
+ "learning_rate": 0.00018098396480737585,
1306
+ "loss": 0.8544,
1307
+ "step": 1820
1308
+ },
1309
+ {
1310
+ "epoch": 2.7395209580838324,
1311
+ "grad_norm": 0.265625,
1312
+ "learning_rate": 0.0001807780037230315,
1313
+ "loss": 0.8557,
1314
+ "step": 1830
1315
+ },
1316
+ {
1317
+ "epoch": 2.754491017964072,
1318
+ "grad_norm": 0.279296875,
1319
+ "learning_rate": 0.00018057105178955905,
1320
+ "loss": 0.8283,
1321
+ "step": 1840
1322
+ },
1323
+ {
1324
+ "epoch": 2.769461077844311,
1325
+ "grad_norm": 0.3828125,
1326
+ "learning_rate": 0.00018036311154549784,
1327
+ "loss": 0.7906,
1328
+ "step": 1850
1329
+ },
1330
+ {
1331
+ "epoch": 2.784431137724551,
1332
+ "grad_norm": 0.2490234375,
1333
+ "learning_rate": 0.0001801541855415102,
1334
+ "loss": 0.8036,
1335
+ "step": 1860
1336
+ },
1337
+ {
1338
+ "epoch": 2.7994011976047903,
1339
+ "grad_norm": 0.298828125,
1340
+ "learning_rate": 0.00017994427634035015,
1341
+ "loss": 0.7828,
1342
+ "step": 1870
1343
+ },
1344
+ {
1345
+ "epoch": 2.81437125748503,
1346
+ "grad_norm": 0.306640625,
1347
+ "learning_rate": 0.00017973338651683176,
1348
+ "loss": 0.7915,
1349
+ "step": 1880
1350
+ },
1351
+ {
1352
+ "epoch": 2.8293413173652695,
1353
+ "grad_norm": 0.228515625,
1354
+ "learning_rate": 0.00017952151865779792,
1355
+ "loss": 0.8141,
1356
+ "step": 1890
1357
+ },
1358
+ {
1359
+ "epoch": 2.844311377245509,
1360
+ "grad_norm": 0.349609375,
1361
+ "learning_rate": 0.00017930867536208826,
1362
+ "loss": 0.8155,
1363
+ "step": 1900
1364
+ },
1365
+ {
1366
+ "epoch": 2.8592814371257482,
1367
+ "grad_norm": 0.310546875,
1368
+ "learning_rate": 0.00017909485924050758,
1369
+ "loss": 0.8004,
1370
+ "step": 1910
1371
+ },
1372
+ {
1373
+ "epoch": 2.874251497005988,
1374
+ "grad_norm": 0.2470703125,
1375
+ "learning_rate": 0.00017888007291579357,
1376
+ "loss": 0.803,
1377
+ "step": 1920
1378
+ },
1379
+ {
1380
+ "epoch": 2.8892215568862274,
1381
+ "grad_norm": 0.275390625,
1382
+ "learning_rate": 0.00017866431902258478,
1383
+ "loss": 0.804,
1384
+ "step": 1930
1385
+ },
1386
+ {
1387
+ "epoch": 2.904191616766467,
1388
+ "grad_norm": 0.310546875,
1389
+ "learning_rate": 0.00017844760020738827,
1390
+ "loss": 0.8154,
1391
+ "step": 1940
1392
+ },
1393
+ {
1394
+ "epoch": 2.9191616766467066,
1395
+ "grad_norm": 0.251953125,
1396
+ "learning_rate": 0.00017822991912854713,
1397
+ "loss": 0.8257,
1398
+ "step": 1950
1399
+ },
1400
+ {
1401
+ "epoch": 2.934131736526946,
1402
+ "grad_norm": 0.3125,
1403
+ "learning_rate": 0.00017801127845620793,
1404
+ "loss": 0.8386,
1405
+ "step": 1960
1406
+ },
1407
+ {
1408
+ "epoch": 2.9491017964071857,
1409
+ "grad_norm": 0.275390625,
1410
+ "learning_rate": 0.0001777916808722879,
1411
+ "loss": 0.8003,
1412
+ "step": 1970
1413
+ },
1414
+ {
1415
+ "epoch": 2.964071856287425,
1416
+ "grad_norm": 0.28515625,
1417
+ "learning_rate": 0.000177571129070442,
1418
+ "loss": 0.8055,
1419
+ "step": 1980
1420
+ },
1421
+ {
1422
+ "epoch": 2.979041916167665,
1423
+ "grad_norm": 0.271484375,
1424
+ "learning_rate": 0.00017734962575603,
1425
+ "loss": 0.8233,
1426
+ "step": 1990
1427
+ },
1428
+ {
1429
+ "epoch": 2.9940119760479043,
1430
+ "grad_norm": 0.255859375,
1431
+ "learning_rate": 0.00017712717364608328,
1432
+ "loss": 0.8106,
1433
+ "step": 2000
1434
+ },
1435
+ {
1436
+ "epoch": 2.9940119760479043,
1437
+ "eval_loss": 0.6911507844924927,
1438
+ "eval_runtime": 109.6711,
1439
+ "eval_samples_per_second": 9.118,
1440
+ "eval_steps_per_second": 1.14,
1441
+ "step": 2000
1442
+ },
1443
+ {
1444
+ "epoch": 3.0089820359281436,
1445
+ "grad_norm": 0.37890625,
1446
+ "learning_rate": 0.00017690377546927133,
1447
+ "loss": 0.7276,
1448
+ "step": 2010
1449
+ },
1450
+ {
1451
+ "epoch": 3.0239520958083834,
1452
+ "grad_norm": 0.27734375,
1453
+ "learning_rate": 0.00017667943396586848,
1454
+ "loss": 0.6882,
1455
+ "step": 2020
1456
+ },
1457
+ {
1458
+ "epoch": 3.038922155688623,
1459
+ "grad_norm": 0.291015625,
1460
+ "learning_rate": 0.0001764541518877202,
1461
+ "loss": 0.6786,
1462
+ "step": 2030
1463
+ },
1464
+ {
1465
+ "epoch": 3.053892215568862,
1466
+ "grad_norm": 0.31640625,
1467
+ "learning_rate": 0.00017622793199820934,
1468
+ "loss": 0.6352,
1469
+ "step": 2040
1470
+ },
1471
+ {
1472
+ "epoch": 3.068862275449102,
1473
+ "grad_norm": 0.294921875,
1474
+ "learning_rate": 0.00017600077707222224,
1475
+ "loss": 0.6648,
1476
+ "step": 2050
1477
+ },
1478
+ {
1479
+ "epoch": 3.0838323353293413,
1480
+ "grad_norm": 0.294921875,
1481
+ "learning_rate": 0.00017577268989611472,
1482
+ "loss": 0.6601,
1483
+ "step": 2060
1484
+ },
1485
+ {
1486
+ "epoch": 3.0988023952095807,
1487
+ "grad_norm": 0.361328125,
1488
+ "learning_rate": 0.00017554367326767792,
1489
+ "loss": 0.6645,
1490
+ "step": 2070
1491
+ },
1492
+ {
1493
+ "epoch": 3.1137724550898205,
1494
+ "grad_norm": 0.251953125,
1495
+ "learning_rate": 0.00017531372999610384,
1496
+ "loss": 0.6631,
1497
+ "step": 2080
1498
+ },
1499
+ {
1500
+ "epoch": 3.12874251497006,
1501
+ "grad_norm": 0.30078125,
1502
+ "learning_rate": 0.0001750828629019511,
1503
+ "loss": 0.6451,
1504
+ "step": 2090
1505
+ },
1506
+ {
1507
+ "epoch": 3.143712574850299,
1508
+ "grad_norm": 0.365234375,
1509
+ "learning_rate": 0.00017485107481711012,
1510
+ "loss": 0.6703,
1511
+ "step": 2100
1512
+ },
1513
+ {
1514
+ "epoch": 3.158682634730539,
1515
+ "grad_norm": 0.341796875,
1516
+ "learning_rate": 0.00017461836858476856,
1517
+ "loss": 0.6598,
1518
+ "step": 2110
1519
+ },
1520
+ {
1521
+ "epoch": 3.1736526946107784,
1522
+ "grad_norm": 0.287109375,
1523
+ "learning_rate": 0.00017438474705937639,
1524
+ "loss": 0.6689,
1525
+ "step": 2120
1526
+ },
1527
+ {
1528
+ "epoch": 3.1886227544910177,
1529
+ "grad_norm": 0.3125,
1530
+ "learning_rate": 0.0001741502131066107,
1531
+ "loss": 0.6741,
1532
+ "step": 2130
1533
+ },
1534
+ {
1535
+ "epoch": 3.2035928143712575,
1536
+ "grad_norm": 0.31640625,
1537
+ "learning_rate": 0.000173914769603341,
1538
+ "loss": 0.6691,
1539
+ "step": 2140
1540
+ },
1541
+ {
1542
+ "epoch": 3.218562874251497,
1543
+ "grad_norm": 0.322265625,
1544
+ "learning_rate": 0.00017367841943759338,
1545
+ "loss": 0.6702,
1546
+ "step": 2150
1547
+ },
1548
+ {
1549
+ "epoch": 3.2335329341317367,
1550
+ "grad_norm": 0.265625,
1551
+ "learning_rate": 0.00017344116550851543,
1552
+ "loss": 0.6451,
1553
+ "step": 2160
1554
+ },
1555
+ {
1556
+ "epoch": 3.248502994011976,
1557
+ "grad_norm": 0.328125,
1558
+ "learning_rate": 0.00017320301072634066,
1559
+ "loss": 0.6507,
1560
+ "step": 2170
1561
+ },
1562
+ {
1563
+ "epoch": 3.2634730538922154,
1564
+ "grad_norm": 0.32421875,
1565
+ "learning_rate": 0.00017296395801235265,
1566
+ "loss": 0.695,
1567
+ "step": 2180
1568
+ },
1569
+ {
1570
+ "epoch": 3.2784431137724552,
1571
+ "grad_norm": 0.31640625,
1572
+ "learning_rate": 0.00017272401029884933,
1573
+ "loss": 0.6798,
1574
+ "step": 2190
1575
+ },
1576
+ {
1577
+ "epoch": 3.2934131736526946,
1578
+ "grad_norm": 0.330078125,
1579
+ "learning_rate": 0.000172483170529107,
1580
+ "loss": 0.6526,
1581
+ "step": 2200
1582
+ },
1583
+ {
1584
+ "epoch": 3.308383233532934,
1585
+ "grad_norm": 0.35546875,
1586
+ "learning_rate": 0.00017224144165734417,
1587
+ "loss": 0.6538,
1588
+ "step": 2210
1589
+ },
1590
+ {
1591
+ "epoch": 3.3233532934131738,
1592
+ "grad_norm": 0.291015625,
1593
+ "learning_rate": 0.00017199882664868538,
1594
+ "loss": 0.6777,
1595
+ "step": 2220
1596
+ },
1597
+ {
1598
+ "epoch": 3.338323353293413,
1599
+ "grad_norm": 0.28515625,
1600
+ "learning_rate": 0.00017175532847912487,
1601
+ "loss": 0.6762,
1602
+ "step": 2230
1603
+ },
1604
+ {
1605
+ "epoch": 3.3532934131736525,
1606
+ "grad_norm": 0.3203125,
1607
+ "learning_rate": 0.00017151095013548994,
1608
+ "loss": 0.673,
1609
+ "step": 2240
1610
+ },
1611
+ {
1612
+ "epoch": 3.3682634730538923,
1613
+ "grad_norm": 0.318359375,
1614
+ "learning_rate": 0.00017126569461540443,
1615
+ "loss": 0.6757,
1616
+ "step": 2250
1617
+ },
1618
+ {
1619
+ "epoch": 3.3832335329341316,
1620
+ "grad_norm": 0.3046875,
1621
+ "learning_rate": 0.00017101956492725185,
1622
+ "loss": 0.6563,
1623
+ "step": 2260
1624
+ },
1625
+ {
1626
+ "epoch": 3.3982035928143715,
1627
+ "grad_norm": 0.267578125,
1628
+ "learning_rate": 0.00017077256409013866,
1629
+ "loss": 0.6877,
1630
+ "step": 2270
1631
+ },
1632
+ {
1633
+ "epoch": 3.413173652694611,
1634
+ "grad_norm": 0.328125,
1635
+ "learning_rate": 0.000170524695133857,
1636
+ "loss": 0.67,
1637
+ "step": 2280
1638
+ },
1639
+ {
1640
+ "epoch": 3.42814371257485,
1641
+ "grad_norm": 0.326171875,
1642
+ "learning_rate": 0.00017027596109884768,
1643
+ "loss": 0.6808,
1644
+ "step": 2290
1645
+ },
1646
+ {
1647
+ "epoch": 3.44311377245509,
1648
+ "grad_norm": 0.318359375,
1649
+ "learning_rate": 0.00017002636503616282,
1650
+ "loss": 0.6941,
1651
+ "step": 2300
1652
+ },
1653
+ {
1654
+ "epoch": 3.4580838323353293,
1655
+ "grad_norm": 0.279296875,
1656
+ "learning_rate": 0.00016977591000742854,
1657
+ "loss": 0.6798,
1658
+ "step": 2310
1659
+ },
1660
+ {
1661
+ "epoch": 3.4730538922155687,
1662
+ "grad_norm": 0.287109375,
1663
+ "learning_rate": 0.0001695245990848072,
1664
+ "loss": 0.6718,
1665
+ "step": 2320
1666
+ },
1667
+ {
1668
+ "epoch": 3.4880239520958085,
1669
+ "grad_norm": 0.30859375,
1670
+ "learning_rate": 0.00016927243535095997,
1671
+ "loss": 0.6483,
1672
+ "step": 2330
1673
+ },
1674
+ {
1675
+ "epoch": 3.502994011976048,
1676
+ "grad_norm": 0.333984375,
1677
+ "learning_rate": 0.00016901942189900867,
1678
+ "loss": 0.7177,
1679
+ "step": 2340
1680
+ },
1681
+ {
1682
+ "epoch": 3.5179640718562872,
1683
+ "grad_norm": 0.34765625,
1684
+ "learning_rate": 0.00016876556183249822,
1685
+ "loss": 0.6833,
1686
+ "step": 2350
1687
+ },
1688
+ {
1689
+ "epoch": 3.532934131736527,
1690
+ "grad_norm": 0.365234375,
1691
+ "learning_rate": 0.00016851085826535838,
1692
+ "loss": 0.6826,
1693
+ "step": 2360
1694
+ },
1695
+ {
1696
+ "epoch": 3.5479041916167664,
1697
+ "grad_norm": 0.396484375,
1698
+ "learning_rate": 0.00016825531432186543,
1699
+ "loss": 0.6835,
1700
+ "step": 2370
1701
+ },
1702
+ {
1703
+ "epoch": 3.562874251497006,
1704
+ "grad_norm": 0.359375,
1705
+ "learning_rate": 0.00016799893313660408,
1706
+ "loss": 0.6791,
1707
+ "step": 2380
1708
+ },
1709
+ {
1710
+ "epoch": 3.5778443113772456,
1711
+ "grad_norm": 0.291015625,
1712
+ "learning_rate": 0.0001677417178544289,
1713
+ "loss": 0.6787,
1714
+ "step": 2390
1715
+ },
1716
+ {
1717
+ "epoch": 3.592814371257485,
1718
+ "grad_norm": 0.296875,
1719
+ "learning_rate": 0.00016748367163042576,
1720
+ "loss": 0.6542,
1721
+ "step": 2400
1722
+ },
1723
+ {
1724
+ "epoch": 3.6077844311377243,
1725
+ "grad_norm": 0.30078125,
1726
+ "learning_rate": 0.00016722479762987317,
1727
+ "loss": 0.6805,
1728
+ "step": 2410
1729
+ },
1730
+ {
1731
+ "epoch": 3.622754491017964,
1732
+ "grad_norm": 0.314453125,
1733
+ "learning_rate": 0.0001669650990282033,
1734
+ "loss": 0.6859,
1735
+ "step": 2420
1736
+ },
1737
+ {
1738
+ "epoch": 3.6377245508982035,
1739
+ "grad_norm": 0.32421875,
1740
+ "learning_rate": 0.00016670457901096328,
1741
+ "loss": 0.6633,
1742
+ "step": 2430
1743
+ },
1744
+ {
1745
+ "epoch": 3.6526946107784433,
1746
+ "grad_norm": 0.32421875,
1747
+ "learning_rate": 0.00016644324077377592,
1748
+ "loss": 0.6958,
1749
+ "step": 2440
1750
+ },
1751
+ {
1752
+ "epoch": 3.6676646706586826,
1753
+ "grad_norm": 0.3125,
1754
+ "learning_rate": 0.00016618108752230052,
1755
+ "loss": 0.6965,
1756
+ "step": 2450
1757
+ },
1758
+ {
1759
+ "epoch": 3.682634730538922,
1760
+ "grad_norm": 0.3046875,
1761
+ "learning_rate": 0.00016591812247219377,
1762
+ "loss": 0.6851,
1763
+ "step": 2460
1764
+ },
1765
+ {
1766
+ "epoch": 3.697604790419162,
1767
+ "grad_norm": 0.400390625,
1768
+ "learning_rate": 0.00016565434884907002,
1769
+ "loss": 0.6669,
1770
+ "step": 2470
1771
+ },
1772
+ {
1773
+ "epoch": 3.712574850299401,
1774
+ "grad_norm": 0.306640625,
1775
+ "learning_rate": 0.0001653897698884619,
1776
+ "loss": 0.6672,
1777
+ "step": 2480
1778
+ },
1779
+ {
1780
+ "epoch": 3.727544910179641,
1781
+ "grad_norm": 0.30078125,
1782
+ "learning_rate": 0.00016512438883578044,
1783
+ "loss": 0.7049,
1784
+ "step": 2490
1785
+ },
1786
+ {
1787
+ "epoch": 3.7425149700598803,
1788
+ "grad_norm": 0.30859375,
1789
+ "learning_rate": 0.0001648582089462756,
1790
+ "loss": 0.6821,
1791
+ "step": 2500
1792
+ },
1793
+ {
1794
+ "epoch": 3.7425149700598803,
1795
+ "eval_loss": 0.5838193893432617,
1796
+ "eval_runtime": 109.8377,
1797
+ "eval_samples_per_second": 9.104,
1798
+ "eval_steps_per_second": 1.138,
1799
+ "step": 2500
1800
+ },
1801
+ {
1802
+ "epoch": 3.7574850299401197,
1803
+ "grad_norm": 0.302734375,
1804
+ "learning_rate": 0.000164591233484996,
1805
+ "loss": 0.6775,
1806
+ "step": 2510
1807
+ },
1808
+ {
1809
+ "epoch": 3.772455089820359,
1810
+ "grad_norm": 0.328125,
1811
+ "learning_rate": 0.00016432346572674896,
1812
+ "loss": 0.6718,
1813
+ "step": 2520
1814
+ },
1815
+ {
1816
+ "epoch": 3.787425149700599,
1817
+ "grad_norm": 0.337890625,
1818
+ "learning_rate": 0.00016405490895606052,
1819
+ "loss": 0.6989,
1820
+ "step": 2530
1821
+ },
1822
+ {
1823
+ "epoch": 3.802395209580838,
1824
+ "grad_norm": 0.2890625,
1825
+ "learning_rate": 0.00016378556646713484,
1826
+ "loss": 0.6681,
1827
+ "step": 2540
1828
+ },
1829
+ {
1830
+ "epoch": 3.817365269461078,
1831
+ "grad_norm": 0.298828125,
1832
+ "learning_rate": 0.00016351544156381414,
1833
+ "loss": 0.6687,
1834
+ "step": 2550
1835
+ },
1836
+ {
1837
+ "epoch": 3.8323353293413174,
1838
+ "grad_norm": 0.318359375,
1839
+ "learning_rate": 0.00016324453755953773,
1840
+ "loss": 0.6771,
1841
+ "step": 2560
1842
+ },
1843
+ {
1844
+ "epoch": 3.8473053892215567,
1845
+ "grad_norm": 0.3359375,
1846
+ "learning_rate": 0.0001629728577773019,
1847
+ "loss": 0.668,
1848
+ "step": 2570
1849
+ },
1850
+ {
1851
+ "epoch": 3.8622754491017965,
1852
+ "grad_norm": 0.345703125,
1853
+ "learning_rate": 0.00016270040554961868,
1854
+ "loss": 0.7025,
1855
+ "step": 2580
1856
+ },
1857
+ {
1858
+ "epoch": 3.877245508982036,
1859
+ "grad_norm": 0.3046875,
1860
+ "learning_rate": 0.00016242718421847528,
1861
+ "loss": 0.6885,
1862
+ "step": 2590
1863
+ },
1864
+ {
1865
+ "epoch": 3.8922155688622757,
1866
+ "grad_norm": 0.349609375,
1867
+ "learning_rate": 0.00016215319713529293,
1868
+ "loss": 0.6989,
1869
+ "step": 2600
1870
+ },
1871
+ {
1872
+ "epoch": 3.907185628742515,
1873
+ "grad_norm": 0.314453125,
1874
+ "learning_rate": 0.00016187844766088586,
1875
+ "loss": 0.6555,
1876
+ "step": 2610
1877
+ },
1878
+ {
1879
+ "epoch": 3.9221556886227544,
1880
+ "grad_norm": 0.349609375,
1881
+ "learning_rate": 0.00016160293916541997,
1882
+ "loss": 0.6777,
1883
+ "step": 2620
1884
+ },
1885
+ {
1886
+ "epoch": 3.937125748502994,
1887
+ "grad_norm": 0.3046875,
1888
+ "learning_rate": 0.00016132667502837165,
1889
+ "loss": 0.6907,
1890
+ "step": 2630
1891
+ },
1892
+ {
1893
+ "epoch": 3.9520958083832336,
1894
+ "grad_norm": 0.3515625,
1895
+ "learning_rate": 0.00016104965863848617,
1896
+ "loss": 0.6832,
1897
+ "step": 2640
1898
+ },
1899
+ {
1900
+ "epoch": 3.967065868263473,
1901
+ "grad_norm": 0.3828125,
1902
+ "learning_rate": 0.00016077189339373614,
1903
+ "loss": 0.6816,
1904
+ "step": 2650
1905
+ },
1906
+ {
1907
+ "epoch": 3.9820359281437128,
1908
+ "grad_norm": 0.287109375,
1909
+ "learning_rate": 0.00016049338270127998,
1910
+ "loss": 0.6616,
1911
+ "step": 2660
1912
+ },
1913
+ {
1914
+ "epoch": 3.997005988023952,
1915
+ "grad_norm": 0.31640625,
1916
+ "learning_rate": 0.00016021412997741993,
1917
+ "loss": 0.6717,
1918
+ "step": 2670
1919
+ },
1920
+ {
1921
+ "epoch": 4.0119760479041915,
1922
+ "grad_norm": 0.34375,
1923
+ "learning_rate": 0.0001599341386475603,
1924
+ "loss": 0.5491,
1925
+ "step": 2680
1926
+ },
1927
+ {
1928
+ "epoch": 4.026946107784431,
1929
+ "grad_norm": 0.33984375,
1930
+ "learning_rate": 0.00015965341214616523,
1931
+ "loss": 0.539,
1932
+ "step": 2690
1933
+ },
1934
+ {
1935
+ "epoch": 4.04191616766467,
1936
+ "grad_norm": 0.373046875,
1937
+ "learning_rate": 0.0001593719539167169,
1938
+ "loss": 0.5273,
1939
+ "step": 2700
1940
+ },
1941
+ {
1942
+ "epoch": 4.0568862275449105,
1943
+ "grad_norm": 0.349609375,
1944
+ "learning_rate": 0.00015908976741167295,
1945
+ "loss": 0.5344,
1946
+ "step": 2710
1947
+ },
1948
+ {
1949
+ "epoch": 4.07185628742515,
1950
+ "grad_norm": 0.275390625,
1951
+ "learning_rate": 0.00015880685609242438,
1952
+ "loss": 0.5228,
1953
+ "step": 2720
1954
+ },
1955
+ {
1956
+ "epoch": 4.086826347305389,
1957
+ "grad_norm": 0.330078125,
1958
+ "learning_rate": 0.00015852322342925295,
1959
+ "loss": 0.4878,
1960
+ "step": 2730
1961
+ },
1962
+ {
1963
+ "epoch": 4.1017964071856285,
1964
+ "grad_norm": 0.361328125,
1965
+ "learning_rate": 0.00015823887290128868,
1966
+ "loss": 0.5228,
1967
+ "step": 2740
1968
+ },
1969
+ {
1970
+ "epoch": 4.116766467065868,
1971
+ "grad_norm": 0.3984375,
1972
+ "learning_rate": 0.00015795380799646716,
1973
+ "loss": 0.5453,
1974
+ "step": 2750
1975
+ },
1976
+ {
1977
+ "epoch": 4.131736526946108,
1978
+ "grad_norm": 0.326171875,
1979
+ "learning_rate": 0.00015766803221148673,
1980
+ "loss": 0.533,
1981
+ "step": 2760
1982
+ },
1983
+ {
1984
+ "epoch": 4.1467065868263475,
1985
+ "grad_norm": 0.296875,
1986
+ "learning_rate": 0.00015738154905176562,
1987
+ "loss": 0.5336,
1988
+ "step": 2770
1989
+ },
1990
+ {
1991
+ "epoch": 4.161676646706587,
1992
+ "grad_norm": 0.390625,
1993
+ "learning_rate": 0.00015709436203139893,
1994
+ "loss": 0.5573,
1995
+ "step": 2780
1996
+ },
1997
+ {
1998
+ "epoch": 4.176646706586826,
1999
+ "grad_norm": 0.31640625,
2000
+ "learning_rate": 0.00015680647467311557,
2001
+ "loss": 0.5451,
2002
+ "step": 2790
2003
+ },
2004
+ {
2005
+ "epoch": 4.191616766467066,
2006
+ "grad_norm": 0.318359375,
2007
+ "learning_rate": 0.00015651789050823508,
2008
+ "loss": 0.5308,
2009
+ "step": 2800
2010
+ },
2011
+ {
2012
+ "epoch": 4.206586826347305,
2013
+ "grad_norm": 0.375,
2014
+ "learning_rate": 0.00015622861307662414,
2015
+ "loss": 0.5224,
2016
+ "step": 2810
2017
+ },
2018
+ {
2019
+ "epoch": 4.221556886227545,
2020
+ "grad_norm": 0.376953125,
2021
+ "learning_rate": 0.00015593864592665333,
2022
+ "loss": 0.555,
2023
+ "step": 2820
2024
+ },
2025
+ {
2026
+ "epoch": 4.236526946107785,
2027
+ "grad_norm": 0.3125,
2028
+ "learning_rate": 0.00015564799261515356,
2029
+ "loss": 0.5702,
2030
+ "step": 2830
2031
+ },
2032
+ {
2033
+ "epoch": 4.251497005988024,
2034
+ "grad_norm": 0.34375,
2035
+ "learning_rate": 0.0001553566567073723,
2036
+ "loss": 0.5246,
2037
+ "step": 2840
2038
+ },
2039
+ {
2040
+ "epoch": 4.266467065868263,
2041
+ "grad_norm": 0.337890625,
2042
+ "learning_rate": 0.0001550646417769301,
2043
+ "loss": 0.5411,
2044
+ "step": 2850
2045
+ },
2046
+ {
2047
+ "epoch": 4.281437125748503,
2048
+ "grad_norm": 0.474609375,
2049
+ "learning_rate": 0.00015477195140577662,
2050
+ "loss": 0.5637,
2051
+ "step": 2860
2052
+ },
2053
+ {
2054
+ "epoch": 4.296407185628743,
2055
+ "grad_norm": 0.318359375,
2056
+ "learning_rate": 0.00015447858918414656,
2057
+ "loss": 0.5507,
2058
+ "step": 2870
2059
+ },
2060
+ {
2061
+ "epoch": 4.311377245508982,
2062
+ "grad_norm": 0.341796875,
2063
+ "learning_rate": 0.00015418455871051592,
2064
+ "loss": 0.5451,
2065
+ "step": 2880
2066
+ },
2067
+ {
2068
+ "epoch": 4.326347305389222,
2069
+ "grad_norm": 0.328125,
2070
+ "learning_rate": 0.00015388986359155758,
2071
+ "loss": 0.5485,
2072
+ "step": 2890
2073
+ },
2074
+ {
2075
+ "epoch": 4.341317365269461,
2076
+ "grad_norm": 0.3203125,
2077
+ "learning_rate": 0.0001535945074420972,
2078
+ "loss": 0.5347,
2079
+ "step": 2900
2080
+ },
2081
+ {
2082
+ "epoch": 4.3562874251497,
2083
+ "grad_norm": 0.3203125,
2084
+ "learning_rate": 0.00015329849388506886,
2085
+ "loss": 0.5242,
2086
+ "step": 2910
2087
+ },
2088
+ {
2089
+ "epoch": 4.37125748502994,
2090
+ "grad_norm": 0.3359375,
2091
+ "learning_rate": 0.00015300182655147068,
2092
+ "loss": 0.5513,
2093
+ "step": 2920
2094
+ },
2095
+ {
2096
+ "epoch": 4.38622754491018,
2097
+ "grad_norm": 0.345703125,
2098
+ "learning_rate": 0.00015270450908032012,
2099
+ "loss": 0.5713,
2100
+ "step": 2930
2101
+ },
2102
+ {
2103
+ "epoch": 4.401197604790419,
2104
+ "grad_norm": 0.37109375,
2105
+ "learning_rate": 0.0001524065451186095,
2106
+ "loss": 0.5441,
2107
+ "step": 2940
2108
+ },
2109
+ {
2110
+ "epoch": 4.416167664670659,
2111
+ "grad_norm": 0.3984375,
2112
+ "learning_rate": 0.00015210793832126112,
2113
+ "loss": 0.548,
2114
+ "step": 2950
2115
+ },
2116
+ {
2117
+ "epoch": 4.431137724550898,
2118
+ "grad_norm": 0.419921875,
2119
+ "learning_rate": 0.00015180869235108263,
2120
+ "loss": 0.5245,
2121
+ "step": 2960
2122
+ },
2123
+ {
2124
+ "epoch": 4.446107784431137,
2125
+ "grad_norm": 0.341796875,
2126
+ "learning_rate": 0.00015150881087872185,
2127
+ "loss": 0.5238,
2128
+ "step": 2970
2129
+ },
2130
+ {
2131
+ "epoch": 4.461077844311378,
2132
+ "grad_norm": 0.37109375,
2133
+ "learning_rate": 0.000151208297582622,
2134
+ "loss": 0.5681,
2135
+ "step": 2980
2136
+ },
2137
+ {
2138
+ "epoch": 4.476047904191617,
2139
+ "grad_norm": 0.455078125,
2140
+ "learning_rate": 0.00015090715614897633,
2141
+ "loss": 0.5413,
2142
+ "step": 2990
2143
+ },
2144
+ {
2145
+ "epoch": 4.491017964071856,
2146
+ "grad_norm": 0.361328125,
2147
+ "learning_rate": 0.00015060539027168316,
2148
+ "loss": 0.563,
2149
+ "step": 3000
2150
+ },
2151
+ {
2152
+ "epoch": 4.491017964071856,
2153
+ "eval_loss": 0.49006855487823486,
2154
+ "eval_runtime": 109.8413,
2155
+ "eval_samples_per_second": 9.104,
2156
+ "eval_steps_per_second": 1.138,
2157
+ "step": 3000
2158
+ }
2159
+ ],
2160
+ "logging_steps": 10,
2161
+ "max_steps": 9000,
2162
+ "num_input_tokens_seen": 0,
2163
+ "num_train_epochs": 14,
2164
+ "save_steps": 500,
2165
+ "stateful_callbacks": {
2166
+ "TrainerControl": {
2167
+ "args": {
2168
+ "should_epoch_stop": false,
2169
+ "should_evaluate": false,
2170
+ "should_log": false,
2171
+ "should_save": true,
2172
+ "should_training_stop": false
2173
+ },
2174
+ "attributes": {}
2175
+ }
2176
+ },
2177
+ "total_flos": 2.0856796870606848e+17,
2178
+ "train_batch_size": 16,
2179
+ "trial_name": null,
2180
+ "trial_params": null
2181
+ }
checkpoint-3000/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-3500/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-3500/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }
checkpoint-3500/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c1d5eea19448e090a91fa5ea43f0d79c56a781afe864b7a227e38c8b8936b467
3
+ size 802337544
checkpoint-3500/optimizer.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1dd3b6b44333371c1d0e5b24eec6db7e92701c0082af10e811e0b982578dc15f
3
+ size 2407189434
checkpoint-3500/rng_state.pth ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:268a65589b6324f2064569fb6037f45292809742fb53ca02cbf8e73f85c160a6
3
+ size 13990
checkpoint-3500/scheduler.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1eaeae1e6a0e5c875209bd9dbb27fd931b7036ff6fd03d97a0a34099962ef5f0
3
+ size 1000
checkpoint-3500/trainer_state.json ADDED
@@ -0,0 +1,2539 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": 0.4033271074295044,
3
+ "best_model_checkpoint": "./0.4b_finetuned_results/checkpoint-3500",
4
+ "epoch": 5.2395209580838324,
5
+ "eval_steps": 500,
6
+ "global_step": 3500,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "epoch": 0.014970059880239521,
13
+ "grad_norm": 4.375,
14
+ "learning_rate": 6.666666666666667e-05,
15
+ "loss": 3.8882,
16
+ "step": 10
17
+ },
18
+ {
19
+ "epoch": 0.029940119760479042,
20
+ "grad_norm": 6.0625,
21
+ "learning_rate": 0.00013333333333333334,
22
+ "loss": 3.2257,
23
+ "step": 20
24
+ },
25
+ {
26
+ "epoch": 0.04491017964071856,
27
+ "grad_norm": 6.28125,
28
+ "learning_rate": 0.0002,
29
+ "loss": 2.92,
30
+ "step": 30
31
+ },
32
+ {
33
+ "epoch": 0.059880239520958084,
34
+ "grad_norm": 1.9765625,
35
+ "learning_rate": 0.00019999938668382333,
36
+ "loss": 2.3984,
37
+ "step": 40
38
+ },
39
+ {
40
+ "epoch": 0.0748502994011976,
41
+ "grad_norm": 1.484375,
42
+ "learning_rate": 0.00019999754674281632,
43
+ "loss": 2.1626,
44
+ "step": 50
45
+ },
46
+ {
47
+ "epoch": 0.08982035928143713,
48
+ "grad_norm": 3.375,
49
+ "learning_rate": 0.0001999944801995484,
50
+ "loss": 2.0388,
51
+ "step": 60
52
+ },
53
+ {
54
+ "epoch": 0.10479041916167664,
55
+ "grad_norm": 1.7890625,
56
+ "learning_rate": 0.0001999901870916347,
57
+ "loss": 2.0121,
58
+ "step": 70
59
+ },
60
+ {
61
+ "epoch": 0.11976047904191617,
62
+ "grad_norm": 1.609375,
63
+ "learning_rate": 0.00019998466747173592,
64
+ "loss": 1.8579,
65
+ "step": 80
66
+ },
67
+ {
68
+ "epoch": 0.1347305389221557,
69
+ "grad_norm": 0.81640625,
70
+ "learning_rate": 0.00019997792140755746,
71
+ "loss": 1.8254,
72
+ "step": 90
73
+ },
74
+ {
75
+ "epoch": 0.1497005988023952,
76
+ "grad_norm": 1.515625,
77
+ "learning_rate": 0.0001999699489818488,
78
+ "loss": 1.7037,
79
+ "step": 100
80
+ },
81
+ {
82
+ "epoch": 0.16467065868263472,
83
+ "grad_norm": 0.94140625,
84
+ "learning_rate": 0.00019996075029240219,
85
+ "loss": 1.6647,
86
+ "step": 110
87
+ },
88
+ {
89
+ "epoch": 0.17964071856287425,
90
+ "grad_norm": 0.61328125,
91
+ "learning_rate": 0.0001999503254520518,
92
+ "loss": 1.5988,
93
+ "step": 120
94
+ },
95
+ {
96
+ "epoch": 0.19461077844311378,
97
+ "grad_norm": 0.337890625,
98
+ "learning_rate": 0.00019993867458867207,
99
+ "loss": 1.6197,
100
+ "step": 130
101
+ },
102
+ {
103
+ "epoch": 0.20958083832335328,
104
+ "grad_norm": 0.47265625,
105
+ "learning_rate": 0.00019992579784517626,
106
+ "loss": 1.5954,
107
+ "step": 140
108
+ },
109
+ {
110
+ "epoch": 0.2245508982035928,
111
+ "grad_norm": 0.33203125,
112
+ "learning_rate": 0.00019991169537951468,
113
+ "loss": 1.5666,
114
+ "step": 150
115
+ },
116
+ {
117
+ "epoch": 0.23952095808383234,
118
+ "grad_norm": 0.52734375,
119
+ "learning_rate": 0.00019989636736467278,
120
+ "loss": 1.5227,
121
+ "step": 160
122
+ },
123
+ {
124
+ "epoch": 0.25449101796407186,
125
+ "grad_norm": 0.34375,
126
+ "learning_rate": 0.00019987981398866887,
127
+ "loss": 1.5048,
128
+ "step": 170
129
+ },
130
+ {
131
+ "epoch": 0.2694610778443114,
132
+ "grad_norm": 0.46875,
133
+ "learning_rate": 0.00019986203545455203,
134
+ "loss": 1.4755,
135
+ "step": 180
136
+ },
137
+ {
138
+ "epoch": 0.2844311377245509,
139
+ "grad_norm": 0.51953125,
140
+ "learning_rate": 0.0001998430319803996,
141
+ "loss": 1.4505,
142
+ "step": 190
143
+ },
144
+ {
145
+ "epoch": 0.2994011976047904,
146
+ "grad_norm": 0.38671875,
147
+ "learning_rate": 0.00019982280379931422,
148
+ "loss": 1.4295,
149
+ "step": 200
150
+ },
151
+ {
152
+ "epoch": 0.3143712574850299,
153
+ "grad_norm": 0.34765625,
154
+ "learning_rate": 0.00019980135115942136,
155
+ "loss": 1.4683,
156
+ "step": 210
157
+ },
158
+ {
159
+ "epoch": 0.32934131736526945,
160
+ "grad_norm": 0.306640625,
161
+ "learning_rate": 0.00019977867432386604,
162
+ "loss": 1.4427,
163
+ "step": 220
164
+ },
165
+ {
166
+ "epoch": 0.344311377245509,
167
+ "grad_norm": 0.357421875,
168
+ "learning_rate": 0.00019975477357080966,
169
+ "loss": 1.3852,
170
+ "step": 230
171
+ },
172
+ {
173
+ "epoch": 0.3592814371257485,
174
+ "grad_norm": 0.361328125,
175
+ "learning_rate": 0.00019972964919342663,
176
+ "loss": 1.427,
177
+ "step": 240
178
+ },
179
+ {
180
+ "epoch": 0.37425149700598803,
181
+ "grad_norm": 0.306640625,
182
+ "learning_rate": 0.00019970330149990062,
183
+ "loss": 1.3759,
184
+ "step": 250
185
+ },
186
+ {
187
+ "epoch": 0.38922155688622756,
188
+ "grad_norm": 0.3515625,
189
+ "learning_rate": 0.00019967573081342103,
190
+ "loss": 1.3559,
191
+ "step": 260
192
+ },
193
+ {
194
+ "epoch": 0.4041916167664671,
195
+ "grad_norm": 0.28515625,
196
+ "learning_rate": 0.00019964693747217874,
197
+ "loss": 1.3715,
198
+ "step": 270
199
+ },
200
+ {
201
+ "epoch": 0.41916167664670656,
202
+ "grad_norm": 0.30859375,
203
+ "learning_rate": 0.00019961692182936225,
204
+ "loss": 1.2932,
205
+ "step": 280
206
+ },
207
+ {
208
+ "epoch": 0.4341317365269461,
209
+ "grad_norm": 0.306640625,
210
+ "learning_rate": 0.00019958568425315314,
211
+ "loss": 1.3086,
212
+ "step": 290
213
+ },
214
+ {
215
+ "epoch": 0.4491017964071856,
216
+ "grad_norm": 0.291015625,
217
+ "learning_rate": 0.00019955322512672162,
218
+ "loss": 1.3091,
219
+ "step": 300
220
+ },
221
+ {
222
+ "epoch": 0.46407185628742514,
223
+ "grad_norm": 0.248046875,
224
+ "learning_rate": 0.00019951954484822182,
225
+ "loss": 1.3196,
226
+ "step": 310
227
+ },
228
+ {
229
+ "epoch": 0.47904191616766467,
230
+ "grad_norm": 0.267578125,
231
+ "learning_rate": 0.00019948464383078696,
232
+ "loss": 1.2944,
233
+ "step": 320
234
+ },
235
+ {
236
+ "epoch": 0.4940119760479042,
237
+ "grad_norm": 0.375,
238
+ "learning_rate": 0.00019944852250252418,
239
+ "loss": 1.3461,
240
+ "step": 330
241
+ },
242
+ {
243
+ "epoch": 0.5089820359281437,
244
+ "grad_norm": 0.275390625,
245
+ "learning_rate": 0.00019941118130650942,
246
+ "loss": 1.3221,
247
+ "step": 340
248
+ },
249
+ {
250
+ "epoch": 0.5239520958083832,
251
+ "grad_norm": 0.23828125,
252
+ "learning_rate": 0.00019937262070078183,
253
+ "loss": 1.3111,
254
+ "step": 350
255
+ },
256
+ {
257
+ "epoch": 0.5389221556886228,
258
+ "grad_norm": 0.2578125,
259
+ "learning_rate": 0.0001993328411583383,
260
+ "loss": 1.3128,
261
+ "step": 360
262
+ },
263
+ {
264
+ "epoch": 0.5538922155688623,
265
+ "grad_norm": 0.2578125,
266
+ "learning_rate": 0.00019929184316712758,
267
+ "loss": 1.2618,
268
+ "step": 370
269
+ },
270
+ {
271
+ "epoch": 0.5688622754491018,
272
+ "grad_norm": 0.29296875,
273
+ "learning_rate": 0.00019924962723004425,
274
+ "loss": 1.2893,
275
+ "step": 380
276
+ },
277
+ {
278
+ "epoch": 0.5838323353293413,
279
+ "grad_norm": 0.30859375,
280
+ "learning_rate": 0.0001992061938649227,
281
+ "loss": 1.2727,
282
+ "step": 390
283
+ },
284
+ {
285
+ "epoch": 0.5988023952095808,
286
+ "grad_norm": 0.3359375,
287
+ "learning_rate": 0.0001991615436045306,
288
+ "loss": 1.293,
289
+ "step": 400
290
+ },
291
+ {
292
+ "epoch": 0.6137724550898204,
293
+ "grad_norm": 0.314453125,
294
+ "learning_rate": 0.0001991156769965625,
295
+ "loss": 1.2692,
296
+ "step": 410
297
+ },
298
+ {
299
+ "epoch": 0.6287425149700598,
300
+ "grad_norm": 0.326171875,
301
+ "learning_rate": 0.00019906859460363307,
302
+ "loss": 1.2588,
303
+ "step": 420
304
+ },
305
+ {
306
+ "epoch": 0.6437125748502994,
307
+ "grad_norm": 0.26953125,
308
+ "learning_rate": 0.00019902029700327018,
309
+ "loss": 1.2576,
310
+ "step": 430
311
+ },
312
+ {
313
+ "epoch": 0.6586826347305389,
314
+ "grad_norm": 0.2890625,
315
+ "learning_rate": 0.0001989707847879078,
316
+ "loss": 1.2595,
317
+ "step": 440
318
+ },
319
+ {
320
+ "epoch": 0.6736526946107785,
321
+ "grad_norm": 0.337890625,
322
+ "learning_rate": 0.00019892005856487878,
323
+ "loss": 1.2331,
324
+ "step": 450
325
+ },
326
+ {
327
+ "epoch": 0.688622754491018,
328
+ "grad_norm": 0.28515625,
329
+ "learning_rate": 0.0001988681189564074,
330
+ "loss": 1.2161,
331
+ "step": 460
332
+ },
333
+ {
334
+ "epoch": 0.7035928143712575,
335
+ "grad_norm": 0.25390625,
336
+ "learning_rate": 0.0001988149665996017,
337
+ "loss": 1.2675,
338
+ "step": 470
339
+ },
340
+ {
341
+ "epoch": 0.718562874251497,
342
+ "grad_norm": 0.26953125,
343
+ "learning_rate": 0.00019876060214644566,
344
+ "loss": 1.269,
345
+ "step": 480
346
+ },
347
+ {
348
+ "epoch": 0.7335329341317365,
349
+ "grad_norm": 0.40625,
350
+ "learning_rate": 0.00019870502626379127,
351
+ "loss": 1.2342,
352
+ "step": 490
353
+ },
354
+ {
355
+ "epoch": 0.7485029940119761,
356
+ "grad_norm": 0.298828125,
357
+ "learning_rate": 0.00019864823963335033,
358
+ "loss": 1.2351,
359
+ "step": 500
360
+ },
361
+ {
362
+ "epoch": 0.7485029940119761,
363
+ "eval_loss": 1.1021808385849,
364
+ "eval_runtime": 109.4058,
365
+ "eval_samples_per_second": 9.14,
366
+ "eval_steps_per_second": 1.143,
367
+ "step": 500
368
+ },
369
+ {
370
+ "epoch": 0.7634730538922155,
371
+ "grad_norm": 0.3671875,
372
+ "learning_rate": 0.00019859024295168593,
373
+ "loss": 1.2235,
374
+ "step": 510
375
+ },
376
+ {
377
+ "epoch": 0.7784431137724551,
378
+ "grad_norm": 0.267578125,
379
+ "learning_rate": 0.0001985310369302042,
380
+ "loss": 1.2353,
381
+ "step": 520
382
+ },
383
+ {
384
+ "epoch": 0.7934131736526946,
385
+ "grad_norm": 0.24609375,
386
+ "learning_rate": 0.00019847062229514533,
387
+ "loss": 1.2445,
388
+ "step": 530
389
+ },
390
+ {
391
+ "epoch": 0.8083832335329342,
392
+ "grad_norm": 0.265625,
393
+ "learning_rate": 0.00019840899978757485,
394
+ "loss": 1.2687,
395
+ "step": 540
396
+ },
397
+ {
398
+ "epoch": 0.8233532934131736,
399
+ "grad_norm": 0.2138671875,
400
+ "learning_rate": 0.0001983461701633742,
401
+ "loss": 1.2026,
402
+ "step": 550
403
+ },
404
+ {
405
+ "epoch": 0.8383233532934131,
406
+ "grad_norm": 0.283203125,
407
+ "learning_rate": 0.00019828213419323208,
408
+ "loss": 1.2304,
409
+ "step": 560
410
+ },
411
+ {
412
+ "epoch": 0.8532934131736527,
413
+ "grad_norm": 0.26953125,
414
+ "learning_rate": 0.00019821689266263427,
415
+ "loss": 1.1961,
416
+ "step": 570
417
+ },
418
+ {
419
+ "epoch": 0.8682634730538922,
420
+ "grad_norm": 0.251953125,
421
+ "learning_rate": 0.00019815044637185456,
422
+ "loss": 1.158,
423
+ "step": 580
424
+ },
425
+ {
426
+ "epoch": 0.8832335329341318,
427
+ "grad_norm": 0.251953125,
428
+ "learning_rate": 0.00019808279613594464,
429
+ "loss": 1.1804,
430
+ "step": 590
431
+ },
432
+ {
433
+ "epoch": 0.8982035928143712,
434
+ "grad_norm": 0.25390625,
435
+ "learning_rate": 0.00019801394278472418,
436
+ "loss": 1.1705,
437
+ "step": 600
438
+ },
439
+ {
440
+ "epoch": 0.9131736526946108,
441
+ "grad_norm": 0.259765625,
442
+ "learning_rate": 0.0001979438871627707,
443
+ "loss": 1.1816,
444
+ "step": 610
445
+ },
446
+ {
447
+ "epoch": 0.9281437125748503,
448
+ "grad_norm": 0.3515625,
449
+ "learning_rate": 0.00019787263012940905,
450
+ "loss": 1.2516,
451
+ "step": 620
452
+ },
453
+ {
454
+ "epoch": 0.9431137724550899,
455
+ "grad_norm": 0.2734375,
456
+ "learning_rate": 0.00019780017255870114,
457
+ "loss": 1.2214,
458
+ "step": 630
459
+ },
460
+ {
461
+ "epoch": 0.9580838323353293,
462
+ "grad_norm": 0.27734375,
463
+ "learning_rate": 0.00019772651533943493,
464
+ "loss": 1.1855,
465
+ "step": 640
466
+ },
467
+ {
468
+ "epoch": 0.9730538922155688,
469
+ "grad_norm": 0.345703125,
470
+ "learning_rate": 0.0001976516593751137,
471
+ "loss": 1.1784,
472
+ "step": 650
473
+ },
474
+ {
475
+ "epoch": 0.9880239520958084,
476
+ "grad_norm": 0.32421875,
477
+ "learning_rate": 0.00019757560558394493,
478
+ "loss": 1.194,
479
+ "step": 660
480
+ },
481
+ {
482
+ "epoch": 1.0029940119760479,
483
+ "grad_norm": 0.283203125,
484
+ "learning_rate": 0.00019749835489882905,
485
+ "loss": 1.198,
486
+ "step": 670
487
+ },
488
+ {
489
+ "epoch": 1.0179640718562875,
490
+ "grad_norm": 0.255859375,
491
+ "learning_rate": 0.00019741990826734794,
492
+ "loss": 1.0274,
493
+ "step": 680
494
+ },
495
+ {
496
+ "epoch": 1.032934131736527,
497
+ "grad_norm": 0.30078125,
498
+ "learning_rate": 0.00019734026665175334,
499
+ "loss": 0.9878,
500
+ "step": 690
501
+ },
502
+ {
503
+ "epoch": 1.0479041916167664,
504
+ "grad_norm": 0.32421875,
505
+ "learning_rate": 0.0001972594310289551,
506
+ "loss": 1.0292,
507
+ "step": 700
508
+ },
509
+ {
510
+ "epoch": 1.062874251497006,
511
+ "grad_norm": 0.2890625,
512
+ "learning_rate": 0.00019717740239050914,
513
+ "loss": 1.0265,
514
+ "step": 710
515
+ },
516
+ {
517
+ "epoch": 1.0778443113772456,
518
+ "grad_norm": 0.314453125,
519
+ "learning_rate": 0.0001970941817426052,
520
+ "loss": 0.9696,
521
+ "step": 720
522
+ },
523
+ {
524
+ "epoch": 1.092814371257485,
525
+ "grad_norm": 0.4453125,
526
+ "learning_rate": 0.0001970097701060548,
527
+ "loss": 0.9735,
528
+ "step": 730
529
+ },
530
+ {
531
+ "epoch": 1.1077844311377245,
532
+ "grad_norm": 0.33984375,
533
+ "learning_rate": 0.00019692416851627826,
534
+ "loss": 1.0029,
535
+ "step": 740
536
+ },
537
+ {
538
+ "epoch": 1.122754491017964,
539
+ "grad_norm": 0.26171875,
540
+ "learning_rate": 0.00019683737802329244,
541
+ "loss": 1.0072,
542
+ "step": 750
543
+ },
544
+ {
545
+ "epoch": 1.1377245508982037,
546
+ "grad_norm": 0.271484375,
547
+ "learning_rate": 0.0001967493996916976,
548
+ "loss": 1.0173,
549
+ "step": 760
550
+ },
551
+ {
552
+ "epoch": 1.152694610778443,
553
+ "grad_norm": 0.267578125,
554
+ "learning_rate": 0.00019666023460066442,
555
+ "loss": 0.9945,
556
+ "step": 770
557
+ },
558
+ {
559
+ "epoch": 1.1676646706586826,
560
+ "grad_norm": 0.287109375,
561
+ "learning_rate": 0.00019656988384392075,
562
+ "loss": 0.9927,
563
+ "step": 780
564
+ },
565
+ {
566
+ "epoch": 1.1826347305389222,
567
+ "grad_norm": 0.283203125,
568
+ "learning_rate": 0.00019647834852973818,
569
+ "loss": 0.9995,
570
+ "step": 790
571
+ },
572
+ {
573
+ "epoch": 1.1976047904191618,
574
+ "grad_norm": 0.29296875,
575
+ "learning_rate": 0.00019638562978091853,
576
+ "loss": 0.9957,
577
+ "step": 800
578
+ },
579
+ {
580
+ "epoch": 1.2125748502994012,
581
+ "grad_norm": 0.2490234375,
582
+ "learning_rate": 0.00019629172873477995,
583
+ "loss": 0.9653,
584
+ "step": 810
585
+ },
586
+ {
587
+ "epoch": 1.2275449101796407,
588
+ "grad_norm": 0.2451171875,
589
+ "learning_rate": 0.00019619664654314302,
590
+ "loss": 0.9714,
591
+ "step": 820
592
+ },
593
+ {
594
+ "epoch": 1.2425149700598803,
595
+ "grad_norm": 0.302734375,
596
+ "learning_rate": 0.0001961003843723167,
597
+ "loss": 1.0226,
598
+ "step": 830
599
+ },
600
+ {
601
+ "epoch": 1.2574850299401197,
602
+ "grad_norm": 0.283203125,
603
+ "learning_rate": 0.00019600294340308398,
604
+ "loss": 1.0417,
605
+ "step": 840
606
+ },
607
+ {
608
+ "epoch": 1.2724550898203593,
609
+ "grad_norm": 0.287109375,
610
+ "learning_rate": 0.00019590432483068722,
611
+ "loss": 0.9593,
612
+ "step": 850
613
+ },
614
+ {
615
+ "epoch": 1.2874251497005988,
616
+ "grad_norm": 0.27734375,
617
+ "learning_rate": 0.00019580452986481378,
618
+ "loss": 1.0255,
619
+ "step": 860
620
+ },
621
+ {
622
+ "epoch": 1.3023952095808382,
623
+ "grad_norm": 0.341796875,
624
+ "learning_rate": 0.00019570355972958097,
625
+ "loss": 0.9971,
626
+ "step": 870
627
+ },
628
+ {
629
+ "epoch": 1.3173652694610778,
630
+ "grad_norm": 0.322265625,
631
+ "learning_rate": 0.00019560141566352115,
632
+ "loss": 0.9914,
633
+ "step": 880
634
+ },
635
+ {
636
+ "epoch": 1.3323353293413174,
637
+ "grad_norm": 0.2255859375,
638
+ "learning_rate": 0.0001954980989195665,
639
+ "loss": 0.9699,
640
+ "step": 890
641
+ },
642
+ {
643
+ "epoch": 1.347305389221557,
644
+ "grad_norm": 0.25390625,
645
+ "learning_rate": 0.0001953936107650336,
646
+ "loss": 0.9667,
647
+ "step": 900
648
+ },
649
+ {
650
+ "epoch": 1.3622754491017965,
651
+ "grad_norm": 0.28515625,
652
+ "learning_rate": 0.00019528795248160795,
653
+ "loss": 0.9813,
654
+ "step": 910
655
+ },
656
+ {
657
+ "epoch": 1.377245508982036,
658
+ "grad_norm": 0.23828125,
659
+ "learning_rate": 0.0001951811253653283,
660
+ "loss": 0.9861,
661
+ "step": 920
662
+ },
663
+ {
664
+ "epoch": 1.3922155688622755,
665
+ "grad_norm": 0.26953125,
666
+ "learning_rate": 0.00019507313072657055,
667
+ "loss": 0.9772,
668
+ "step": 930
669
+ },
670
+ {
671
+ "epoch": 1.407185628742515,
672
+ "grad_norm": 0.259765625,
673
+ "learning_rate": 0.00019496396989003193,
674
+ "loss": 1.0045,
675
+ "step": 940
676
+ },
677
+ {
678
+ "epoch": 1.4221556886227544,
679
+ "grad_norm": 0.302734375,
680
+ "learning_rate": 0.00019485364419471454,
681
+ "loss": 0.9919,
682
+ "step": 950
683
+ },
684
+ {
685
+ "epoch": 1.437125748502994,
686
+ "grad_norm": 0.279296875,
687
+ "learning_rate": 0.00019474215499390912,
688
+ "loss": 0.9796,
689
+ "step": 960
690
+ },
691
+ {
692
+ "epoch": 1.4520958083832336,
693
+ "grad_norm": 0.26171875,
694
+ "learning_rate": 0.00019462950365517817,
695
+ "loss": 0.9821,
696
+ "step": 970
697
+ },
698
+ {
699
+ "epoch": 1.467065868263473,
700
+ "grad_norm": 0.3203125,
701
+ "learning_rate": 0.00019451569156033954,
702
+ "loss": 1.0337,
703
+ "step": 980
704
+ },
705
+ {
706
+ "epoch": 1.4820359281437125,
707
+ "grad_norm": 0.310546875,
708
+ "learning_rate": 0.00019440072010544918,
709
+ "loss": 0.9987,
710
+ "step": 990
711
+ },
712
+ {
713
+ "epoch": 1.4970059880239521,
714
+ "grad_norm": 0.27734375,
715
+ "learning_rate": 0.00019428459070078416,
716
+ "loss": 1.004,
717
+ "step": 1000
718
+ },
719
+ {
720
+ "epoch": 1.4970059880239521,
721
+ "eval_loss": 0.9072233438491821,
722
+ "eval_runtime": 109.5318,
723
+ "eval_samples_per_second": 9.13,
724
+ "eval_steps_per_second": 1.141,
725
+ "step": 1000
726
+ },
727
+ {
728
+ "epoch": 1.5119760479041915,
729
+ "grad_norm": 0.2890625,
730
+ "learning_rate": 0.00019416730477082533,
731
+ "loss": 0.9444,
732
+ "step": 1010
733
+ },
734
+ {
735
+ "epoch": 1.5269461077844313,
736
+ "grad_norm": 0.296875,
737
+ "learning_rate": 0.00019404886375423984,
738
+ "loss": 0.9829,
739
+ "step": 1020
740
+ },
741
+ {
742
+ "epoch": 1.5419161676646707,
743
+ "grad_norm": 0.27734375,
744
+ "learning_rate": 0.00019392926910386353,
745
+ "loss": 0.9532,
746
+ "step": 1030
747
+ },
748
+ {
749
+ "epoch": 1.55688622754491,
750
+ "grad_norm": 0.265625,
751
+ "learning_rate": 0.00019380852228668304,
752
+ "loss": 0.9769,
753
+ "step": 1040
754
+ },
755
+ {
756
+ "epoch": 1.5718562874251498,
757
+ "grad_norm": 0.275390625,
758
+ "learning_rate": 0.00019368662478381799,
759
+ "loss": 0.9783,
760
+ "step": 1050
761
+ },
762
+ {
763
+ "epoch": 1.5868263473053892,
764
+ "grad_norm": 0.2890625,
765
+ "learning_rate": 0.00019356357809050247,
766
+ "loss": 0.9881,
767
+ "step": 1060
768
+ },
769
+ {
770
+ "epoch": 1.6017964071856288,
771
+ "grad_norm": 0.232421875,
772
+ "learning_rate": 0.00019343938371606712,
773
+ "loss": 0.9883,
774
+ "step": 1070
775
+ },
776
+ {
777
+ "epoch": 1.6167664670658684,
778
+ "grad_norm": 0.2734375,
779
+ "learning_rate": 0.00019331404318392027,
780
+ "loss": 0.9893,
781
+ "step": 1080
782
+ },
783
+ {
784
+ "epoch": 1.6317365269461077,
785
+ "grad_norm": 0.248046875,
786
+ "learning_rate": 0.00019318755803152945,
787
+ "loss": 0.9851,
788
+ "step": 1090
789
+ },
790
+ {
791
+ "epoch": 1.6467065868263473,
792
+ "grad_norm": 0.28515625,
793
+ "learning_rate": 0.00019305992981040246,
794
+ "loss": 0.9531,
795
+ "step": 1100
796
+ },
797
+ {
798
+ "epoch": 1.6616766467065869,
799
+ "grad_norm": 0.275390625,
800
+ "learning_rate": 0.00019293116008606837,
801
+ "loss": 0.9717,
802
+ "step": 1110
803
+ },
804
+ {
805
+ "epoch": 1.6766467065868262,
806
+ "grad_norm": 0.232421875,
807
+ "learning_rate": 0.00019280125043805824,
808
+ "loss": 0.9699,
809
+ "step": 1120
810
+ },
811
+ {
812
+ "epoch": 1.6916167664670658,
813
+ "grad_norm": 0.2294921875,
814
+ "learning_rate": 0.00019267020245988592,
815
+ "loss": 0.9407,
816
+ "step": 1130
817
+ },
818
+ {
819
+ "epoch": 1.7065868263473054,
820
+ "grad_norm": 0.251953125,
821
+ "learning_rate": 0.00019253801775902824,
822
+ "loss": 0.977,
823
+ "step": 1140
824
+ },
825
+ {
826
+ "epoch": 1.7215568862275448,
827
+ "grad_norm": 0.287109375,
828
+ "learning_rate": 0.0001924046979569055,
829
+ "loss": 0.9549,
830
+ "step": 1150
831
+ },
832
+ {
833
+ "epoch": 1.7365269461077846,
834
+ "grad_norm": 0.248046875,
835
+ "learning_rate": 0.00019227024468886157,
836
+ "loss": 0.9824,
837
+ "step": 1160
838
+ },
839
+ {
840
+ "epoch": 1.751497005988024,
841
+ "grad_norm": 0.259765625,
842
+ "learning_rate": 0.00019213465960414368,
843
+ "loss": 0.9936,
844
+ "step": 1170
845
+ },
846
+ {
847
+ "epoch": 1.7664670658682635,
848
+ "grad_norm": 0.29296875,
849
+ "learning_rate": 0.00019199794436588243,
850
+ "loss": 1.0042,
851
+ "step": 1180
852
+ },
853
+ {
854
+ "epoch": 1.781437125748503,
855
+ "grad_norm": 0.23828125,
856
+ "learning_rate": 0.0001918601006510711,
857
+ "loss": 0.9629,
858
+ "step": 1190
859
+ },
860
+ {
861
+ "epoch": 1.7964071856287425,
862
+ "grad_norm": 0.2890625,
863
+ "learning_rate": 0.00019172113015054532,
864
+ "loss": 0.9522,
865
+ "step": 1200
866
+ },
867
+ {
868
+ "epoch": 1.811377245508982,
869
+ "grad_norm": 0.26953125,
870
+ "learning_rate": 0.0001915810345689622,
871
+ "loss": 0.9806,
872
+ "step": 1210
873
+ },
874
+ {
875
+ "epoch": 1.8263473053892216,
876
+ "grad_norm": 0.263671875,
877
+ "learning_rate": 0.00019143981562477947,
878
+ "loss": 0.9736,
879
+ "step": 1220
880
+ },
881
+ {
882
+ "epoch": 1.841317365269461,
883
+ "grad_norm": 0.2578125,
884
+ "learning_rate": 0.00019129747505023436,
885
+ "loss": 0.9701,
886
+ "step": 1230
887
+ },
888
+ {
889
+ "epoch": 1.8562874251497006,
890
+ "grad_norm": 0.22265625,
891
+ "learning_rate": 0.00019115401459132247,
892
+ "loss": 0.9494,
893
+ "step": 1240
894
+ },
895
+ {
896
+ "epoch": 1.8712574850299402,
897
+ "grad_norm": 0.251953125,
898
+ "learning_rate": 0.00019100943600777615,
899
+ "loss": 0.9922,
900
+ "step": 1250
901
+ },
902
+ {
903
+ "epoch": 1.8862275449101795,
904
+ "grad_norm": 0.2490234375,
905
+ "learning_rate": 0.00019086374107304312,
906
+ "loss": 0.9711,
907
+ "step": 1260
908
+ },
909
+ {
910
+ "epoch": 1.9011976047904193,
911
+ "grad_norm": 0.25390625,
912
+ "learning_rate": 0.00019071693157426457,
913
+ "loss": 0.9664,
914
+ "step": 1270
915
+ },
916
+ {
917
+ "epoch": 1.9161676646706587,
918
+ "grad_norm": 0.26171875,
919
+ "learning_rate": 0.00019056900931225333,
920
+ "loss": 0.9591,
921
+ "step": 1280
922
+ },
923
+ {
924
+ "epoch": 1.931137724550898,
925
+ "grad_norm": 0.380859375,
926
+ "learning_rate": 0.00019041997610147167,
927
+ "loss": 0.9942,
928
+ "step": 1290
929
+ },
930
+ {
931
+ "epoch": 1.9461077844311379,
932
+ "grad_norm": 0.2294921875,
933
+ "learning_rate": 0.0001902698337700092,
934
+ "loss": 0.9391,
935
+ "step": 1300
936
+ },
937
+ {
938
+ "epoch": 1.9610778443113772,
939
+ "grad_norm": 0.298828125,
940
+ "learning_rate": 0.00019011858415956038,
941
+ "loss": 0.9993,
942
+ "step": 1310
943
+ },
944
+ {
945
+ "epoch": 1.9760479041916168,
946
+ "grad_norm": 0.29296875,
947
+ "learning_rate": 0.0001899662291254018,
948
+ "loss": 0.9571,
949
+ "step": 1320
950
+ },
951
+ {
952
+ "epoch": 1.9910179640718564,
953
+ "grad_norm": 0.283203125,
954
+ "learning_rate": 0.0001898127705363696,
955
+ "loss": 0.9835,
956
+ "step": 1330
957
+ },
958
+ {
959
+ "epoch": 2.0059880239520957,
960
+ "grad_norm": 0.27734375,
961
+ "learning_rate": 0.00018965821027483654,
962
+ "loss": 0.9305,
963
+ "step": 1340
964
+ },
965
+ {
966
+ "epoch": 2.020958083832335,
967
+ "grad_norm": 0.26171875,
968
+ "learning_rate": 0.00018950255023668876,
969
+ "loss": 0.8295,
970
+ "step": 1350
971
+ },
972
+ {
973
+ "epoch": 2.035928143712575,
974
+ "grad_norm": 0.240234375,
975
+ "learning_rate": 0.00018934579233130267,
976
+ "loss": 0.7653,
977
+ "step": 1360
978
+ },
979
+ {
980
+ "epoch": 2.0508982035928143,
981
+ "grad_norm": 0.29296875,
982
+ "learning_rate": 0.00018918793848152142,
983
+ "loss": 0.7581,
984
+ "step": 1370
985
+ },
986
+ {
987
+ "epoch": 2.065868263473054,
988
+ "grad_norm": 0.29296875,
989
+ "learning_rate": 0.00018902899062363143,
990
+ "loss": 0.7983,
991
+ "step": 1380
992
+ },
993
+ {
994
+ "epoch": 2.0808383233532934,
995
+ "grad_norm": 0.279296875,
996
+ "learning_rate": 0.0001888689507073385,
997
+ "loss": 0.8187,
998
+ "step": 1390
999
+ },
1000
+ {
1001
+ "epoch": 2.095808383233533,
1002
+ "grad_norm": 0.29296875,
1003
+ "learning_rate": 0.0001887078206957441,
1004
+ "loss": 0.7879,
1005
+ "step": 1400
1006
+ },
1007
+ {
1008
+ "epoch": 2.1107784431137726,
1009
+ "grad_norm": 0.2734375,
1010
+ "learning_rate": 0.000188545602565321,
1011
+ "loss": 0.8083,
1012
+ "step": 1410
1013
+ },
1014
+ {
1015
+ "epoch": 2.125748502994012,
1016
+ "grad_norm": 0.263671875,
1017
+ "learning_rate": 0.00018838229830588934,
1018
+ "loss": 0.8057,
1019
+ "step": 1420
1020
+ },
1021
+ {
1022
+ "epoch": 2.1407185628742513,
1023
+ "grad_norm": 0.244140625,
1024
+ "learning_rate": 0.00018821790992059196,
1025
+ "loss": 0.8194,
1026
+ "step": 1430
1027
+ },
1028
+ {
1029
+ "epoch": 2.155688622754491,
1030
+ "grad_norm": 0.2451171875,
1031
+ "learning_rate": 0.00018805243942587,
1032
+ "loss": 0.7958,
1033
+ "step": 1440
1034
+ },
1035
+ {
1036
+ "epoch": 2.1706586826347305,
1037
+ "grad_norm": 0.248046875,
1038
+ "learning_rate": 0.00018788588885143808,
1039
+ "loss": 0.8169,
1040
+ "step": 1450
1041
+ },
1042
+ {
1043
+ "epoch": 2.18562874251497,
1044
+ "grad_norm": 0.265625,
1045
+ "learning_rate": 0.00018771826024025946,
1046
+ "loss": 0.7722,
1047
+ "step": 1460
1048
+ },
1049
+ {
1050
+ "epoch": 2.2005988023952097,
1051
+ "grad_norm": 0.265625,
1052
+ "learning_rate": 0.0001875495556485208,
1053
+ "loss": 0.7934,
1054
+ "step": 1470
1055
+ },
1056
+ {
1057
+ "epoch": 2.215568862275449,
1058
+ "grad_norm": 0.244140625,
1059
+ "learning_rate": 0.00018737977714560738,
1060
+ "loss": 0.7915,
1061
+ "step": 1480
1062
+ },
1063
+ {
1064
+ "epoch": 2.230538922155689,
1065
+ "grad_norm": 0.310546875,
1066
+ "learning_rate": 0.00018720892681407708,
1067
+ "loss": 0.8021,
1068
+ "step": 1490
1069
+ },
1070
+ {
1071
+ "epoch": 2.245508982035928,
1072
+ "grad_norm": 0.298828125,
1073
+ "learning_rate": 0.00018703700674963547,
1074
+ "loss": 0.7987,
1075
+ "step": 1500
1076
+ },
1077
+ {
1078
+ "epoch": 2.245508982035928,
1079
+ "eval_loss": 0.7798940539360046,
1080
+ "eval_runtime": 109.62,
1081
+ "eval_samples_per_second": 9.122,
1082
+ "eval_steps_per_second": 1.14,
1083
+ "step": 1500
1084
+ },
1085
+ {
1086
+ "epoch": 2.2604790419161676,
1087
+ "grad_norm": 0.259765625,
1088
+ "learning_rate": 0.00018686401906110964,
1089
+ "loss": 0.7979,
1090
+ "step": 1510
1091
+ },
1092
+ {
1093
+ "epoch": 2.2754491017964074,
1094
+ "grad_norm": 0.265625,
1095
+ "learning_rate": 0.00018668996587042252,
1096
+ "loss": 0.8255,
1097
+ "step": 1520
1098
+ },
1099
+ {
1100
+ "epoch": 2.2904191616766467,
1101
+ "grad_norm": 0.2470703125,
1102
+ "learning_rate": 0.00018651484931256685,
1103
+ "loss": 0.8252,
1104
+ "step": 1530
1105
+ },
1106
+ {
1107
+ "epoch": 2.305389221556886,
1108
+ "grad_norm": 0.25,
1109
+ "learning_rate": 0.00018633867153557905,
1110
+ "loss": 0.8455,
1111
+ "step": 1540
1112
+ },
1113
+ {
1114
+ "epoch": 2.320359281437126,
1115
+ "grad_norm": 0.302734375,
1116
+ "learning_rate": 0.00018616143470051263,
1117
+ "loss": 0.8118,
1118
+ "step": 1550
1119
+ },
1120
+ {
1121
+ "epoch": 2.3353293413173652,
1122
+ "grad_norm": 0.283203125,
1123
+ "learning_rate": 0.00018598314098141206,
1124
+ "loss": 0.8122,
1125
+ "step": 1560
1126
+ },
1127
+ {
1128
+ "epoch": 2.3502994011976046,
1129
+ "grad_norm": 0.24609375,
1130
+ "learning_rate": 0.00018580379256528576,
1131
+ "loss": 0.7965,
1132
+ "step": 1570
1133
+ },
1134
+ {
1135
+ "epoch": 2.3652694610778444,
1136
+ "grad_norm": 0.279296875,
1137
+ "learning_rate": 0.00018562339165207936,
1138
+ "loss": 0.8309,
1139
+ "step": 1580
1140
+ },
1141
+ {
1142
+ "epoch": 2.3802395209580838,
1143
+ "grad_norm": 0.31640625,
1144
+ "learning_rate": 0.00018544194045464886,
1145
+ "loss": 0.8046,
1146
+ "step": 1590
1147
+ },
1148
+ {
1149
+ "epoch": 2.3952095808383236,
1150
+ "grad_norm": 0.2734375,
1151
+ "learning_rate": 0.0001852594411987334,
1152
+ "loss": 0.8467,
1153
+ "step": 1600
1154
+ },
1155
+ {
1156
+ "epoch": 2.410179640718563,
1157
+ "grad_norm": 0.275390625,
1158
+ "learning_rate": 0.00018507589612292783,
1159
+ "loss": 0.8566,
1160
+ "step": 1610
1161
+ },
1162
+ {
1163
+ "epoch": 2.4251497005988023,
1164
+ "grad_norm": 0.30859375,
1165
+ "learning_rate": 0.00018489130747865548,
1166
+ "loss": 0.8297,
1167
+ "step": 1620
1168
+ },
1169
+ {
1170
+ "epoch": 2.440119760479042,
1171
+ "grad_norm": 0.359375,
1172
+ "learning_rate": 0.00018470567753014035,
1173
+ "loss": 0.7823,
1174
+ "step": 1630
1175
+ },
1176
+ {
1177
+ "epoch": 2.4550898203592815,
1178
+ "grad_norm": 0.29296875,
1179
+ "learning_rate": 0.0001845190085543795,
1180
+ "loss": 0.8558,
1181
+ "step": 1640
1182
+ },
1183
+ {
1184
+ "epoch": 2.470059880239521,
1185
+ "grad_norm": 0.3046875,
1186
+ "learning_rate": 0.0001843313028411149,
1187
+ "loss": 0.8262,
1188
+ "step": 1650
1189
+ },
1190
+ {
1191
+ "epoch": 2.4850299401197606,
1192
+ "grad_norm": 0.2578125,
1193
+ "learning_rate": 0.00018414256269280564,
1194
+ "loss": 0.7982,
1195
+ "step": 1660
1196
+ },
1197
+ {
1198
+ "epoch": 2.5,
1199
+ "grad_norm": 0.314453125,
1200
+ "learning_rate": 0.00018395279042459937,
1201
+ "loss": 0.8182,
1202
+ "step": 1670
1203
+ },
1204
+ {
1205
+ "epoch": 2.5149700598802394,
1206
+ "grad_norm": 0.328125,
1207
+ "learning_rate": 0.00018376198836430417,
1208
+ "loss": 0.8275,
1209
+ "step": 1680
1210
+ },
1211
+ {
1212
+ "epoch": 2.529940119760479,
1213
+ "grad_norm": 0.271484375,
1214
+ "learning_rate": 0.00018357015885235982,
1215
+ "loss": 0.8102,
1216
+ "step": 1690
1217
+ },
1218
+ {
1219
+ "epoch": 2.5449101796407185,
1220
+ "grad_norm": 0.3125,
1221
+ "learning_rate": 0.0001833773042418092,
1222
+ "loss": 0.8145,
1223
+ "step": 1700
1224
+ },
1225
+ {
1226
+ "epoch": 2.5598802395209583,
1227
+ "grad_norm": 1.40625,
1228
+ "learning_rate": 0.00018318342689826938,
1229
+ "loss": 0.8279,
1230
+ "step": 1710
1231
+ },
1232
+ {
1233
+ "epoch": 2.5748502994011977,
1234
+ "grad_norm": 0.275390625,
1235
+ "learning_rate": 0.00018298852919990252,
1236
+ "loss": 0.8354,
1237
+ "step": 1720
1238
+ },
1239
+ {
1240
+ "epoch": 2.589820359281437,
1241
+ "grad_norm": 0.3125,
1242
+ "learning_rate": 0.0001827926135373869,
1243
+ "loss": 0.8106,
1244
+ "step": 1730
1245
+ },
1246
+ {
1247
+ "epoch": 2.6047904191616764,
1248
+ "grad_norm": 0.267578125,
1249
+ "learning_rate": 0.00018259568231388738,
1250
+ "loss": 0.7983,
1251
+ "step": 1740
1252
+ },
1253
+ {
1254
+ "epoch": 2.6197604790419162,
1255
+ "grad_norm": 0.251953125,
1256
+ "learning_rate": 0.00018239773794502607,
1257
+ "loss": 0.8183,
1258
+ "step": 1750
1259
+ },
1260
+ {
1261
+ "epoch": 2.6347305389221556,
1262
+ "grad_norm": 0.294921875,
1263
+ "learning_rate": 0.00018219878285885267,
1264
+ "loss": 0.8462,
1265
+ "step": 1760
1266
+ },
1267
+ {
1268
+ "epoch": 2.6497005988023954,
1269
+ "grad_norm": 0.283203125,
1270
+ "learning_rate": 0.0001819988194958146,
1271
+ "loss": 0.8227,
1272
+ "step": 1770
1273
+ },
1274
+ {
1275
+ "epoch": 2.6646706586826348,
1276
+ "grad_norm": 0.2431640625,
1277
+ "learning_rate": 0.0001817978503087272,
1278
+ "loss": 0.8104,
1279
+ "step": 1780
1280
+ },
1281
+ {
1282
+ "epoch": 2.679640718562874,
1283
+ "grad_norm": 0.271484375,
1284
+ "learning_rate": 0.0001815958777627435,
1285
+ "loss": 0.7972,
1286
+ "step": 1790
1287
+ },
1288
+ {
1289
+ "epoch": 2.694610778443114,
1290
+ "grad_norm": 0.283203125,
1291
+ "learning_rate": 0.00018139290433532416,
1292
+ "loss": 0.8339,
1293
+ "step": 1800
1294
+ },
1295
+ {
1296
+ "epoch": 2.7095808383233533,
1297
+ "grad_norm": 0.298828125,
1298
+ "learning_rate": 0.00018118893251620682,
1299
+ "loss": 0.8723,
1300
+ "step": 1810
1301
+ },
1302
+ {
1303
+ "epoch": 2.724550898203593,
1304
+ "grad_norm": 0.3203125,
1305
+ "learning_rate": 0.00018098396480737585,
1306
+ "loss": 0.8544,
1307
+ "step": 1820
1308
+ },
1309
+ {
1310
+ "epoch": 2.7395209580838324,
1311
+ "grad_norm": 0.265625,
1312
+ "learning_rate": 0.0001807780037230315,
1313
+ "loss": 0.8557,
1314
+ "step": 1830
1315
+ },
1316
+ {
1317
+ "epoch": 2.754491017964072,
1318
+ "grad_norm": 0.279296875,
1319
+ "learning_rate": 0.00018057105178955905,
1320
+ "loss": 0.8283,
1321
+ "step": 1840
1322
+ },
1323
+ {
1324
+ "epoch": 2.769461077844311,
1325
+ "grad_norm": 0.3828125,
1326
+ "learning_rate": 0.00018036311154549784,
1327
+ "loss": 0.7906,
1328
+ "step": 1850
1329
+ },
1330
+ {
1331
+ "epoch": 2.784431137724551,
1332
+ "grad_norm": 0.2490234375,
1333
+ "learning_rate": 0.0001801541855415102,
1334
+ "loss": 0.8036,
1335
+ "step": 1860
1336
+ },
1337
+ {
1338
+ "epoch": 2.7994011976047903,
1339
+ "grad_norm": 0.298828125,
1340
+ "learning_rate": 0.00017994427634035015,
1341
+ "loss": 0.7828,
1342
+ "step": 1870
1343
+ },
1344
+ {
1345
+ "epoch": 2.81437125748503,
1346
+ "grad_norm": 0.306640625,
1347
+ "learning_rate": 0.00017973338651683176,
1348
+ "loss": 0.7915,
1349
+ "step": 1880
1350
+ },
1351
+ {
1352
+ "epoch": 2.8293413173652695,
1353
+ "grad_norm": 0.228515625,
1354
+ "learning_rate": 0.00017952151865779792,
1355
+ "loss": 0.8141,
1356
+ "step": 1890
1357
+ },
1358
+ {
1359
+ "epoch": 2.844311377245509,
1360
+ "grad_norm": 0.349609375,
1361
+ "learning_rate": 0.00017930867536208826,
1362
+ "loss": 0.8155,
1363
+ "step": 1900
1364
+ },
1365
+ {
1366
+ "epoch": 2.8592814371257482,
1367
+ "grad_norm": 0.310546875,
1368
+ "learning_rate": 0.00017909485924050758,
1369
+ "loss": 0.8004,
1370
+ "step": 1910
1371
+ },
1372
+ {
1373
+ "epoch": 2.874251497005988,
1374
+ "grad_norm": 0.2470703125,
1375
+ "learning_rate": 0.00017888007291579357,
1376
+ "loss": 0.803,
1377
+ "step": 1920
1378
+ },
1379
+ {
1380
+ "epoch": 2.8892215568862274,
1381
+ "grad_norm": 0.275390625,
1382
+ "learning_rate": 0.00017866431902258478,
1383
+ "loss": 0.804,
1384
+ "step": 1930
1385
+ },
1386
+ {
1387
+ "epoch": 2.904191616766467,
1388
+ "grad_norm": 0.310546875,
1389
+ "learning_rate": 0.00017844760020738827,
1390
+ "loss": 0.8154,
1391
+ "step": 1940
1392
+ },
1393
+ {
1394
+ "epoch": 2.9191616766467066,
1395
+ "grad_norm": 0.251953125,
1396
+ "learning_rate": 0.00017822991912854713,
1397
+ "loss": 0.8257,
1398
+ "step": 1950
1399
+ },
1400
+ {
1401
+ "epoch": 2.934131736526946,
1402
+ "grad_norm": 0.3125,
1403
+ "learning_rate": 0.00017801127845620793,
1404
+ "loss": 0.8386,
1405
+ "step": 1960
1406
+ },
1407
+ {
1408
+ "epoch": 2.9491017964071857,
1409
+ "grad_norm": 0.275390625,
1410
+ "learning_rate": 0.0001777916808722879,
1411
+ "loss": 0.8003,
1412
+ "step": 1970
1413
+ },
1414
+ {
1415
+ "epoch": 2.964071856287425,
1416
+ "grad_norm": 0.28515625,
1417
+ "learning_rate": 0.000177571129070442,
1418
+ "loss": 0.8055,
1419
+ "step": 1980
1420
+ },
1421
+ {
1422
+ "epoch": 2.979041916167665,
1423
+ "grad_norm": 0.271484375,
1424
+ "learning_rate": 0.00017734962575603,
1425
+ "loss": 0.8233,
1426
+ "step": 1990
1427
+ },
1428
+ {
1429
+ "epoch": 2.9940119760479043,
1430
+ "grad_norm": 0.255859375,
1431
+ "learning_rate": 0.00017712717364608328,
1432
+ "loss": 0.8106,
1433
+ "step": 2000
1434
+ },
1435
+ {
1436
+ "epoch": 2.9940119760479043,
1437
+ "eval_loss": 0.6911507844924927,
1438
+ "eval_runtime": 109.6711,
1439
+ "eval_samples_per_second": 9.118,
1440
+ "eval_steps_per_second": 1.14,
1441
+ "step": 2000
1442
+ },
1443
+ {
1444
+ "epoch": 3.0089820359281436,
1445
+ "grad_norm": 0.37890625,
1446
+ "learning_rate": 0.00017690377546927133,
1447
+ "loss": 0.7276,
1448
+ "step": 2010
1449
+ },
1450
+ {
1451
+ "epoch": 3.0239520958083834,
1452
+ "grad_norm": 0.27734375,
1453
+ "learning_rate": 0.00017667943396586848,
1454
+ "loss": 0.6882,
1455
+ "step": 2020
1456
+ },
1457
+ {
1458
+ "epoch": 3.038922155688623,
1459
+ "grad_norm": 0.291015625,
1460
+ "learning_rate": 0.0001764541518877202,
1461
+ "loss": 0.6786,
1462
+ "step": 2030
1463
+ },
1464
+ {
1465
+ "epoch": 3.053892215568862,
1466
+ "grad_norm": 0.31640625,
1467
+ "learning_rate": 0.00017622793199820934,
1468
+ "loss": 0.6352,
1469
+ "step": 2040
1470
+ },
1471
+ {
1472
+ "epoch": 3.068862275449102,
1473
+ "grad_norm": 0.294921875,
1474
+ "learning_rate": 0.00017600077707222224,
1475
+ "loss": 0.6648,
1476
+ "step": 2050
1477
+ },
1478
+ {
1479
+ "epoch": 3.0838323353293413,
1480
+ "grad_norm": 0.294921875,
1481
+ "learning_rate": 0.00017577268989611472,
1482
+ "loss": 0.6601,
1483
+ "step": 2060
1484
+ },
1485
+ {
1486
+ "epoch": 3.0988023952095807,
1487
+ "grad_norm": 0.361328125,
1488
+ "learning_rate": 0.00017554367326767792,
1489
+ "loss": 0.6645,
1490
+ "step": 2070
1491
+ },
1492
+ {
1493
+ "epoch": 3.1137724550898205,
1494
+ "grad_norm": 0.251953125,
1495
+ "learning_rate": 0.00017531372999610384,
1496
+ "loss": 0.6631,
1497
+ "step": 2080
1498
+ },
1499
+ {
1500
+ "epoch": 3.12874251497006,
1501
+ "grad_norm": 0.30078125,
1502
+ "learning_rate": 0.0001750828629019511,
1503
+ "loss": 0.6451,
1504
+ "step": 2090
1505
+ },
1506
+ {
1507
+ "epoch": 3.143712574850299,
1508
+ "grad_norm": 0.365234375,
1509
+ "learning_rate": 0.00017485107481711012,
1510
+ "loss": 0.6703,
1511
+ "step": 2100
1512
+ },
1513
+ {
1514
+ "epoch": 3.158682634730539,
1515
+ "grad_norm": 0.341796875,
1516
+ "learning_rate": 0.00017461836858476856,
1517
+ "loss": 0.6598,
1518
+ "step": 2110
1519
+ },
1520
+ {
1521
+ "epoch": 3.1736526946107784,
1522
+ "grad_norm": 0.287109375,
1523
+ "learning_rate": 0.00017438474705937639,
1524
+ "loss": 0.6689,
1525
+ "step": 2120
1526
+ },
1527
+ {
1528
+ "epoch": 3.1886227544910177,
1529
+ "grad_norm": 0.3125,
1530
+ "learning_rate": 0.0001741502131066107,
1531
+ "loss": 0.6741,
1532
+ "step": 2130
1533
+ },
1534
+ {
1535
+ "epoch": 3.2035928143712575,
1536
+ "grad_norm": 0.31640625,
1537
+ "learning_rate": 0.000173914769603341,
1538
+ "loss": 0.6691,
1539
+ "step": 2140
1540
+ },
1541
+ {
1542
+ "epoch": 3.218562874251497,
1543
+ "grad_norm": 0.322265625,
1544
+ "learning_rate": 0.00017367841943759338,
1545
+ "loss": 0.6702,
1546
+ "step": 2150
1547
+ },
1548
+ {
1549
+ "epoch": 3.2335329341317367,
1550
+ "grad_norm": 0.265625,
1551
+ "learning_rate": 0.00017344116550851543,
1552
+ "loss": 0.6451,
1553
+ "step": 2160
1554
+ },
1555
+ {
1556
+ "epoch": 3.248502994011976,
1557
+ "grad_norm": 0.328125,
1558
+ "learning_rate": 0.00017320301072634066,
1559
+ "loss": 0.6507,
1560
+ "step": 2170
1561
+ },
1562
+ {
1563
+ "epoch": 3.2634730538922154,
1564
+ "grad_norm": 0.32421875,
1565
+ "learning_rate": 0.00017296395801235265,
1566
+ "loss": 0.695,
1567
+ "step": 2180
1568
+ },
1569
+ {
1570
+ "epoch": 3.2784431137724552,
1571
+ "grad_norm": 0.31640625,
1572
+ "learning_rate": 0.00017272401029884933,
1573
+ "loss": 0.6798,
1574
+ "step": 2190
1575
+ },
1576
+ {
1577
+ "epoch": 3.2934131736526946,
1578
+ "grad_norm": 0.330078125,
1579
+ "learning_rate": 0.000172483170529107,
1580
+ "loss": 0.6526,
1581
+ "step": 2200
1582
+ },
1583
+ {
1584
+ "epoch": 3.308383233532934,
1585
+ "grad_norm": 0.35546875,
1586
+ "learning_rate": 0.00017224144165734417,
1587
+ "loss": 0.6538,
1588
+ "step": 2210
1589
+ },
1590
+ {
1591
+ "epoch": 3.3233532934131738,
1592
+ "grad_norm": 0.291015625,
1593
+ "learning_rate": 0.00017199882664868538,
1594
+ "loss": 0.6777,
1595
+ "step": 2220
1596
+ },
1597
+ {
1598
+ "epoch": 3.338323353293413,
1599
+ "grad_norm": 0.28515625,
1600
+ "learning_rate": 0.00017175532847912487,
1601
+ "loss": 0.6762,
1602
+ "step": 2230
1603
+ },
1604
+ {
1605
+ "epoch": 3.3532934131736525,
1606
+ "grad_norm": 0.3203125,
1607
+ "learning_rate": 0.00017151095013548994,
1608
+ "loss": 0.673,
1609
+ "step": 2240
1610
+ },
1611
+ {
1612
+ "epoch": 3.3682634730538923,
1613
+ "grad_norm": 0.318359375,
1614
+ "learning_rate": 0.00017126569461540443,
1615
+ "loss": 0.6757,
1616
+ "step": 2250
1617
+ },
1618
+ {
1619
+ "epoch": 3.3832335329341316,
1620
+ "grad_norm": 0.3046875,
1621
+ "learning_rate": 0.00017101956492725185,
1622
+ "loss": 0.6563,
1623
+ "step": 2260
1624
+ },
1625
+ {
1626
+ "epoch": 3.3982035928143715,
1627
+ "grad_norm": 0.267578125,
1628
+ "learning_rate": 0.00017077256409013866,
1629
+ "loss": 0.6877,
1630
+ "step": 2270
1631
+ },
1632
+ {
1633
+ "epoch": 3.413173652694611,
1634
+ "grad_norm": 0.328125,
1635
+ "learning_rate": 0.000170524695133857,
1636
+ "loss": 0.67,
1637
+ "step": 2280
1638
+ },
1639
+ {
1640
+ "epoch": 3.42814371257485,
1641
+ "grad_norm": 0.326171875,
1642
+ "learning_rate": 0.00017027596109884768,
1643
+ "loss": 0.6808,
1644
+ "step": 2290
1645
+ },
1646
+ {
1647
+ "epoch": 3.44311377245509,
1648
+ "grad_norm": 0.318359375,
1649
+ "learning_rate": 0.00017002636503616282,
1650
+ "loss": 0.6941,
1651
+ "step": 2300
1652
+ },
1653
+ {
1654
+ "epoch": 3.4580838323353293,
1655
+ "grad_norm": 0.279296875,
1656
+ "learning_rate": 0.00016977591000742854,
1657
+ "loss": 0.6798,
1658
+ "step": 2310
1659
+ },
1660
+ {
1661
+ "epoch": 3.4730538922155687,
1662
+ "grad_norm": 0.287109375,
1663
+ "learning_rate": 0.0001695245990848072,
1664
+ "loss": 0.6718,
1665
+ "step": 2320
1666
+ },
1667
+ {
1668
+ "epoch": 3.4880239520958085,
1669
+ "grad_norm": 0.30859375,
1670
+ "learning_rate": 0.00016927243535095997,
1671
+ "loss": 0.6483,
1672
+ "step": 2330
1673
+ },
1674
+ {
1675
+ "epoch": 3.502994011976048,
1676
+ "grad_norm": 0.333984375,
1677
+ "learning_rate": 0.00016901942189900867,
1678
+ "loss": 0.7177,
1679
+ "step": 2340
1680
+ },
1681
+ {
1682
+ "epoch": 3.5179640718562872,
1683
+ "grad_norm": 0.34765625,
1684
+ "learning_rate": 0.00016876556183249822,
1685
+ "loss": 0.6833,
1686
+ "step": 2350
1687
+ },
1688
+ {
1689
+ "epoch": 3.532934131736527,
1690
+ "grad_norm": 0.365234375,
1691
+ "learning_rate": 0.00016851085826535838,
1692
+ "loss": 0.6826,
1693
+ "step": 2360
1694
+ },
1695
+ {
1696
+ "epoch": 3.5479041916167664,
1697
+ "grad_norm": 0.396484375,
1698
+ "learning_rate": 0.00016825531432186543,
1699
+ "loss": 0.6835,
1700
+ "step": 2370
1701
+ },
1702
+ {
1703
+ "epoch": 3.562874251497006,
1704
+ "grad_norm": 0.359375,
1705
+ "learning_rate": 0.00016799893313660408,
1706
+ "loss": 0.6791,
1707
+ "step": 2380
1708
+ },
1709
+ {
1710
+ "epoch": 3.5778443113772456,
1711
+ "grad_norm": 0.291015625,
1712
+ "learning_rate": 0.0001677417178544289,
1713
+ "loss": 0.6787,
1714
+ "step": 2390
1715
+ },
1716
+ {
1717
+ "epoch": 3.592814371257485,
1718
+ "grad_norm": 0.296875,
1719
+ "learning_rate": 0.00016748367163042576,
1720
+ "loss": 0.6542,
1721
+ "step": 2400
1722
+ },
1723
+ {
1724
+ "epoch": 3.6077844311377243,
1725
+ "grad_norm": 0.30078125,
1726
+ "learning_rate": 0.00016722479762987317,
1727
+ "loss": 0.6805,
1728
+ "step": 2410
1729
+ },
1730
+ {
1731
+ "epoch": 3.622754491017964,
1732
+ "grad_norm": 0.314453125,
1733
+ "learning_rate": 0.0001669650990282033,
1734
+ "loss": 0.6859,
1735
+ "step": 2420
1736
+ },
1737
+ {
1738
+ "epoch": 3.6377245508982035,
1739
+ "grad_norm": 0.32421875,
1740
+ "learning_rate": 0.00016670457901096328,
1741
+ "loss": 0.6633,
1742
+ "step": 2430
1743
+ },
1744
+ {
1745
+ "epoch": 3.6526946107784433,
1746
+ "grad_norm": 0.32421875,
1747
+ "learning_rate": 0.00016644324077377592,
1748
+ "loss": 0.6958,
1749
+ "step": 2440
1750
+ },
1751
+ {
1752
+ "epoch": 3.6676646706586826,
1753
+ "grad_norm": 0.3125,
1754
+ "learning_rate": 0.00016618108752230052,
1755
+ "loss": 0.6965,
1756
+ "step": 2450
1757
+ },
1758
+ {
1759
+ "epoch": 3.682634730538922,
1760
+ "grad_norm": 0.3046875,
1761
+ "learning_rate": 0.00016591812247219377,
1762
+ "loss": 0.6851,
1763
+ "step": 2460
1764
+ },
1765
+ {
1766
+ "epoch": 3.697604790419162,
1767
+ "grad_norm": 0.400390625,
1768
+ "learning_rate": 0.00016565434884907002,
1769
+ "loss": 0.6669,
1770
+ "step": 2470
1771
+ },
1772
+ {
1773
+ "epoch": 3.712574850299401,
1774
+ "grad_norm": 0.306640625,
1775
+ "learning_rate": 0.0001653897698884619,
1776
+ "loss": 0.6672,
1777
+ "step": 2480
1778
+ },
1779
+ {
1780
+ "epoch": 3.727544910179641,
1781
+ "grad_norm": 0.30078125,
1782
+ "learning_rate": 0.00016512438883578044,
1783
+ "loss": 0.7049,
1784
+ "step": 2490
1785
+ },
1786
+ {
1787
+ "epoch": 3.7425149700598803,
1788
+ "grad_norm": 0.30859375,
1789
+ "learning_rate": 0.0001648582089462756,
1790
+ "loss": 0.6821,
1791
+ "step": 2500
1792
+ },
1793
+ {
1794
+ "epoch": 3.7425149700598803,
1795
+ "eval_loss": 0.5838193893432617,
1796
+ "eval_runtime": 109.8377,
1797
+ "eval_samples_per_second": 9.104,
1798
+ "eval_steps_per_second": 1.138,
1799
+ "step": 2500
1800
+ },
1801
+ {
1802
+ "epoch": 3.7574850299401197,
1803
+ "grad_norm": 0.302734375,
1804
+ "learning_rate": 0.000164591233484996,
1805
+ "loss": 0.6775,
1806
+ "step": 2510
1807
+ },
1808
+ {
1809
+ "epoch": 3.772455089820359,
1810
+ "grad_norm": 0.328125,
1811
+ "learning_rate": 0.00016432346572674896,
1812
+ "loss": 0.6718,
1813
+ "step": 2520
1814
+ },
1815
+ {
1816
+ "epoch": 3.787425149700599,
1817
+ "grad_norm": 0.337890625,
1818
+ "learning_rate": 0.00016405490895606052,
1819
+ "loss": 0.6989,
1820
+ "step": 2530
1821
+ },
1822
+ {
1823
+ "epoch": 3.802395209580838,
1824
+ "grad_norm": 0.2890625,
1825
+ "learning_rate": 0.00016378556646713484,
1826
+ "loss": 0.6681,
1827
+ "step": 2540
1828
+ },
1829
+ {
1830
+ "epoch": 3.817365269461078,
1831
+ "grad_norm": 0.298828125,
1832
+ "learning_rate": 0.00016351544156381414,
1833
+ "loss": 0.6687,
1834
+ "step": 2550
1835
+ },
1836
+ {
1837
+ "epoch": 3.8323353293413174,
1838
+ "grad_norm": 0.318359375,
1839
+ "learning_rate": 0.00016324453755953773,
1840
+ "loss": 0.6771,
1841
+ "step": 2560
1842
+ },
1843
+ {
1844
+ "epoch": 3.8473053892215567,
1845
+ "grad_norm": 0.3359375,
1846
+ "learning_rate": 0.0001629728577773019,
1847
+ "loss": 0.668,
1848
+ "step": 2570
1849
+ },
1850
+ {
1851
+ "epoch": 3.8622754491017965,
1852
+ "grad_norm": 0.345703125,
1853
+ "learning_rate": 0.00016270040554961868,
1854
+ "loss": 0.7025,
1855
+ "step": 2580
1856
+ },
1857
+ {
1858
+ "epoch": 3.877245508982036,
1859
+ "grad_norm": 0.3046875,
1860
+ "learning_rate": 0.00016242718421847528,
1861
+ "loss": 0.6885,
1862
+ "step": 2590
1863
+ },
1864
+ {
1865
+ "epoch": 3.8922155688622757,
1866
+ "grad_norm": 0.349609375,
1867
+ "learning_rate": 0.00016215319713529293,
1868
+ "loss": 0.6989,
1869
+ "step": 2600
1870
+ },
1871
+ {
1872
+ "epoch": 3.907185628742515,
1873
+ "grad_norm": 0.314453125,
1874
+ "learning_rate": 0.00016187844766088586,
1875
+ "loss": 0.6555,
1876
+ "step": 2610
1877
+ },
1878
+ {
1879
+ "epoch": 3.9221556886227544,
1880
+ "grad_norm": 0.349609375,
1881
+ "learning_rate": 0.00016160293916541997,
1882
+ "loss": 0.6777,
1883
+ "step": 2620
1884
+ },
1885
+ {
1886
+ "epoch": 3.937125748502994,
1887
+ "grad_norm": 0.3046875,
1888
+ "learning_rate": 0.00016132667502837165,
1889
+ "loss": 0.6907,
1890
+ "step": 2630
1891
+ },
1892
+ {
1893
+ "epoch": 3.9520958083832336,
1894
+ "grad_norm": 0.3515625,
1895
+ "learning_rate": 0.00016104965863848617,
1896
+ "loss": 0.6832,
1897
+ "step": 2640
1898
+ },
1899
+ {
1900
+ "epoch": 3.967065868263473,
1901
+ "grad_norm": 0.3828125,
1902
+ "learning_rate": 0.00016077189339373614,
1903
+ "loss": 0.6816,
1904
+ "step": 2650
1905
+ },
1906
+ {
1907
+ "epoch": 3.9820359281437128,
1908
+ "grad_norm": 0.287109375,
1909
+ "learning_rate": 0.00016049338270127998,
1910
+ "loss": 0.6616,
1911
+ "step": 2660
1912
+ },
1913
+ {
1914
+ "epoch": 3.997005988023952,
1915
+ "grad_norm": 0.31640625,
1916
+ "learning_rate": 0.00016021412997741993,
1917
+ "loss": 0.6717,
1918
+ "step": 2670
1919
+ },
1920
+ {
1921
+ "epoch": 4.0119760479041915,
1922
+ "grad_norm": 0.34375,
1923
+ "learning_rate": 0.0001599341386475603,
1924
+ "loss": 0.5491,
1925
+ "step": 2680
1926
+ },
1927
+ {
1928
+ "epoch": 4.026946107784431,
1929
+ "grad_norm": 0.33984375,
1930
+ "learning_rate": 0.00015965341214616523,
1931
+ "loss": 0.539,
1932
+ "step": 2690
1933
+ },
1934
+ {
1935
+ "epoch": 4.04191616766467,
1936
+ "grad_norm": 0.373046875,
1937
+ "learning_rate": 0.0001593719539167169,
1938
+ "loss": 0.5273,
1939
+ "step": 2700
1940
+ },
1941
+ {
1942
+ "epoch": 4.0568862275449105,
1943
+ "grad_norm": 0.349609375,
1944
+ "learning_rate": 0.00015908976741167295,
1945
+ "loss": 0.5344,
1946
+ "step": 2710
1947
+ },
1948
+ {
1949
+ "epoch": 4.07185628742515,
1950
+ "grad_norm": 0.275390625,
1951
+ "learning_rate": 0.00015880685609242438,
1952
+ "loss": 0.5228,
1953
+ "step": 2720
1954
+ },
1955
+ {
1956
+ "epoch": 4.086826347305389,
1957
+ "grad_norm": 0.330078125,
1958
+ "learning_rate": 0.00015852322342925295,
1959
+ "loss": 0.4878,
1960
+ "step": 2730
1961
+ },
1962
+ {
1963
+ "epoch": 4.1017964071856285,
1964
+ "grad_norm": 0.361328125,
1965
+ "learning_rate": 0.00015823887290128868,
1966
+ "loss": 0.5228,
1967
+ "step": 2740
1968
+ },
1969
+ {
1970
+ "epoch": 4.116766467065868,
1971
+ "grad_norm": 0.3984375,
1972
+ "learning_rate": 0.00015795380799646716,
1973
+ "loss": 0.5453,
1974
+ "step": 2750
1975
+ },
1976
+ {
1977
+ "epoch": 4.131736526946108,
1978
+ "grad_norm": 0.326171875,
1979
+ "learning_rate": 0.00015766803221148673,
1980
+ "loss": 0.533,
1981
+ "step": 2760
1982
+ },
1983
+ {
1984
+ "epoch": 4.1467065868263475,
1985
+ "grad_norm": 0.296875,
1986
+ "learning_rate": 0.00015738154905176562,
1987
+ "loss": 0.5336,
1988
+ "step": 2770
1989
+ },
1990
+ {
1991
+ "epoch": 4.161676646706587,
1992
+ "grad_norm": 0.390625,
1993
+ "learning_rate": 0.00015709436203139893,
1994
+ "loss": 0.5573,
1995
+ "step": 2780
1996
+ },
1997
+ {
1998
+ "epoch": 4.176646706586826,
1999
+ "grad_norm": 0.31640625,
2000
+ "learning_rate": 0.00015680647467311557,
2001
+ "loss": 0.5451,
2002
+ "step": 2790
2003
+ },
2004
+ {
2005
+ "epoch": 4.191616766467066,
2006
+ "grad_norm": 0.318359375,
2007
+ "learning_rate": 0.00015651789050823508,
2008
+ "loss": 0.5308,
2009
+ "step": 2800
2010
+ },
2011
+ {
2012
+ "epoch": 4.206586826347305,
2013
+ "grad_norm": 0.375,
2014
+ "learning_rate": 0.00015622861307662414,
2015
+ "loss": 0.5224,
2016
+ "step": 2810
2017
+ },
2018
+ {
2019
+ "epoch": 4.221556886227545,
2020
+ "grad_norm": 0.376953125,
2021
+ "learning_rate": 0.00015593864592665333,
2022
+ "loss": 0.555,
2023
+ "step": 2820
2024
+ },
2025
+ {
2026
+ "epoch": 4.236526946107785,
2027
+ "grad_norm": 0.3125,
2028
+ "learning_rate": 0.00015564799261515356,
2029
+ "loss": 0.5702,
2030
+ "step": 2830
2031
+ },
2032
+ {
2033
+ "epoch": 4.251497005988024,
2034
+ "grad_norm": 0.34375,
2035
+ "learning_rate": 0.0001553566567073723,
2036
+ "loss": 0.5246,
2037
+ "step": 2840
2038
+ },
2039
+ {
2040
+ "epoch": 4.266467065868263,
2041
+ "grad_norm": 0.337890625,
2042
+ "learning_rate": 0.0001550646417769301,
2043
+ "loss": 0.5411,
2044
+ "step": 2850
2045
+ },
2046
+ {
2047
+ "epoch": 4.281437125748503,
2048
+ "grad_norm": 0.474609375,
2049
+ "learning_rate": 0.00015477195140577662,
2050
+ "loss": 0.5637,
2051
+ "step": 2860
2052
+ },
2053
+ {
2054
+ "epoch": 4.296407185628743,
2055
+ "grad_norm": 0.318359375,
2056
+ "learning_rate": 0.00015447858918414656,
2057
+ "loss": 0.5507,
2058
+ "step": 2870
2059
+ },
2060
+ {
2061
+ "epoch": 4.311377245508982,
2062
+ "grad_norm": 0.341796875,
2063
+ "learning_rate": 0.00015418455871051592,
2064
+ "loss": 0.5451,
2065
+ "step": 2880
2066
+ },
2067
+ {
2068
+ "epoch": 4.326347305389222,
2069
+ "grad_norm": 0.328125,
2070
+ "learning_rate": 0.00015388986359155758,
2071
+ "loss": 0.5485,
2072
+ "step": 2890
2073
+ },
2074
+ {
2075
+ "epoch": 4.341317365269461,
2076
+ "grad_norm": 0.3203125,
2077
+ "learning_rate": 0.0001535945074420972,
2078
+ "loss": 0.5347,
2079
+ "step": 2900
2080
+ },
2081
+ {
2082
+ "epoch": 4.3562874251497,
2083
+ "grad_norm": 0.3203125,
2084
+ "learning_rate": 0.00015329849388506886,
2085
+ "loss": 0.5242,
2086
+ "step": 2910
2087
+ },
2088
+ {
2089
+ "epoch": 4.37125748502994,
2090
+ "grad_norm": 0.3359375,
2091
+ "learning_rate": 0.00015300182655147068,
2092
+ "loss": 0.5513,
2093
+ "step": 2920
2094
+ },
2095
+ {
2096
+ "epoch": 4.38622754491018,
2097
+ "grad_norm": 0.345703125,
2098
+ "learning_rate": 0.00015270450908032012,
2099
+ "loss": 0.5713,
2100
+ "step": 2930
2101
+ },
2102
+ {
2103
+ "epoch": 4.401197604790419,
2104
+ "grad_norm": 0.37109375,
2105
+ "learning_rate": 0.0001524065451186095,
2106
+ "loss": 0.5441,
2107
+ "step": 2940
2108
+ },
2109
+ {
2110
+ "epoch": 4.416167664670659,
2111
+ "grad_norm": 0.3984375,
2112
+ "learning_rate": 0.00015210793832126112,
2113
+ "loss": 0.548,
2114
+ "step": 2950
2115
+ },
2116
+ {
2117
+ "epoch": 4.431137724550898,
2118
+ "grad_norm": 0.419921875,
2119
+ "learning_rate": 0.00015180869235108263,
2120
+ "loss": 0.5245,
2121
+ "step": 2960
2122
+ },
2123
+ {
2124
+ "epoch": 4.446107784431137,
2125
+ "grad_norm": 0.341796875,
2126
+ "learning_rate": 0.00015150881087872185,
2127
+ "loss": 0.5238,
2128
+ "step": 2970
2129
+ },
2130
+ {
2131
+ "epoch": 4.461077844311378,
2132
+ "grad_norm": 0.37109375,
2133
+ "learning_rate": 0.000151208297582622,
2134
+ "loss": 0.5681,
2135
+ "step": 2980
2136
+ },
2137
+ {
2138
+ "epoch": 4.476047904191617,
2139
+ "grad_norm": 0.455078125,
2140
+ "learning_rate": 0.00015090715614897633,
2141
+ "loss": 0.5413,
2142
+ "step": 2990
2143
+ },
2144
+ {
2145
+ "epoch": 4.491017964071856,
2146
+ "grad_norm": 0.361328125,
2147
+ "learning_rate": 0.00015060539027168316,
2148
+ "loss": 0.563,
2149
+ "step": 3000
2150
+ },
2151
+ {
2152
+ "epoch": 4.491017964071856,
2153
+ "eval_loss": 0.49006855487823486,
2154
+ "eval_runtime": 109.8413,
2155
+ "eval_samples_per_second": 9.104,
2156
+ "eval_steps_per_second": 1.138,
2157
+ "step": 3000
2158
+ },
2159
+ {
2160
+ "epoch": 4.505988023952096,
2161
+ "grad_norm": 0.3515625,
2162
+ "learning_rate": 0.00015030300365230037,
2163
+ "loss": 0.5549,
2164
+ "step": 3010
2165
+ },
2166
+ {
2167
+ "epoch": 4.520958083832335,
2168
+ "grad_norm": 0.36328125,
2169
+ "learning_rate": 0.00015000000000000001,
2170
+ "loss": 0.5628,
2171
+ "step": 3020
2172
+ },
2173
+ {
2174
+ "epoch": 4.5359281437125745,
2175
+ "grad_norm": 0.390625,
2176
+ "learning_rate": 0.00014969638303152295,
2177
+ "loss": 0.5382,
2178
+ "step": 3030
2179
+ },
2180
+ {
2181
+ "epoch": 4.550898203592815,
2182
+ "grad_norm": 0.39453125,
2183
+ "learning_rate": 0.00014939215647113318,
2184
+ "loss": 0.5595,
2185
+ "step": 3040
2186
+ },
2187
+ {
2188
+ "epoch": 4.565868263473054,
2189
+ "grad_norm": 0.318359375,
2190
+ "learning_rate": 0.00014908732405057208,
2191
+ "loss": 0.5244,
2192
+ "step": 3050
2193
+ },
2194
+ {
2195
+ "epoch": 4.580838323353293,
2196
+ "grad_norm": 0.353515625,
2197
+ "learning_rate": 0.00014878188950901276,
2198
+ "loss": 0.5385,
2199
+ "step": 3060
2200
+ },
2201
+ {
2202
+ "epoch": 4.595808383233533,
2203
+ "grad_norm": 0.400390625,
2204
+ "learning_rate": 0.0001484758565930141,
2205
+ "loss": 0.5468,
2206
+ "step": 3070
2207
+ },
2208
+ {
2209
+ "epoch": 4.610778443113772,
2210
+ "grad_norm": 0.36328125,
2211
+ "learning_rate": 0.00014816922905647485,
2212
+ "loss": 0.5334,
2213
+ "step": 3080
2214
+ },
2215
+ {
2216
+ "epoch": 4.625748502994012,
2217
+ "grad_norm": 0.380859375,
2218
+ "learning_rate": 0.00014786201066058766,
2219
+ "loss": 0.5373,
2220
+ "step": 3090
2221
+ },
2222
+ {
2223
+ "epoch": 4.640718562874252,
2224
+ "grad_norm": 0.328125,
2225
+ "learning_rate": 0.0001475542051737927,
2226
+ "loss": 0.572,
2227
+ "step": 3100
2228
+ },
2229
+ {
2230
+ "epoch": 4.655688622754491,
2231
+ "grad_norm": 0.421875,
2232
+ "learning_rate": 0.00014724581637173164,
2233
+ "loss": 0.5445,
2234
+ "step": 3110
2235
+ },
2236
+ {
2237
+ "epoch": 4.6706586826347305,
2238
+ "grad_norm": 0.380859375,
2239
+ "learning_rate": 0.00014693684803720138,
2240
+ "loss": 0.541,
2241
+ "step": 3120
2242
+ },
2243
+ {
2244
+ "epoch": 4.68562874251497,
2245
+ "grad_norm": 0.34765625,
2246
+ "learning_rate": 0.0001466273039601074,
2247
+ "loss": 0.5388,
2248
+ "step": 3130
2249
+ },
2250
+ {
2251
+ "epoch": 4.700598802395209,
2252
+ "grad_norm": 0.416015625,
2253
+ "learning_rate": 0.00014631718793741753,
2254
+ "loss": 0.5485,
2255
+ "step": 3140
2256
+ },
2257
+ {
2258
+ "epoch": 4.7155688622754495,
2259
+ "grad_norm": 0.302734375,
2260
+ "learning_rate": 0.00014600650377311522,
2261
+ "loss": 0.5594,
2262
+ "step": 3150
2263
+ },
2264
+ {
2265
+ "epoch": 4.730538922155689,
2266
+ "grad_norm": 0.46484375,
2267
+ "learning_rate": 0.00014569525527815297,
2268
+ "loss": 0.5648,
2269
+ "step": 3160
2270
+ },
2271
+ {
2272
+ "epoch": 4.745508982035928,
2273
+ "grad_norm": 0.32421875,
2274
+ "learning_rate": 0.0001453834462704055,
2275
+ "loss": 0.5642,
2276
+ "step": 3170
2277
+ },
2278
+ {
2279
+ "epoch": 4.7604790419161676,
2280
+ "grad_norm": 0.37109375,
2281
+ "learning_rate": 0.00014507108057462296,
2282
+ "loss": 0.5443,
2283
+ "step": 3180
2284
+ },
2285
+ {
2286
+ "epoch": 4.775449101796407,
2287
+ "grad_norm": 0.3671875,
2288
+ "learning_rate": 0.00014475816202238413,
2289
+ "loss": 0.5249,
2290
+ "step": 3190
2291
+ },
2292
+ {
2293
+ "epoch": 4.790419161676647,
2294
+ "grad_norm": 0.390625,
2295
+ "learning_rate": 0.00014444469445204906,
2296
+ "loss": 0.5496,
2297
+ "step": 3200
2298
+ },
2299
+ {
2300
+ "epoch": 4.8053892215568865,
2301
+ "grad_norm": 0.357421875,
2302
+ "learning_rate": 0.0001441306817087125,
2303
+ "loss": 0.5723,
2304
+ "step": 3210
2305
+ },
2306
+ {
2307
+ "epoch": 4.820359281437126,
2308
+ "grad_norm": 0.3359375,
2309
+ "learning_rate": 0.00014381612764415635,
2310
+ "loss": 0.5551,
2311
+ "step": 3220
2312
+ },
2313
+ {
2314
+ "epoch": 4.835329341317365,
2315
+ "grad_norm": 0.345703125,
2316
+ "learning_rate": 0.00014350103611680248,
2317
+ "loss": 0.5748,
2318
+ "step": 3230
2319
+ },
2320
+ {
2321
+ "epoch": 4.850299401197605,
2322
+ "grad_norm": 0.34375,
2323
+ "learning_rate": 0.00014318541099166555,
2324
+ "loss": 0.5515,
2325
+ "step": 3240
2326
+ },
2327
+ {
2328
+ "epoch": 4.865269461077844,
2329
+ "grad_norm": 0.3984375,
2330
+ "learning_rate": 0.00014286925614030542,
2331
+ "loss": 0.5261,
2332
+ "step": 3250
2333
+ },
2334
+ {
2335
+ "epoch": 4.880239520958084,
2336
+ "grad_norm": 0.310546875,
2337
+ "learning_rate": 0.0001425525754407798,
2338
+ "loss": 0.5381,
2339
+ "step": 3260
2340
+ },
2341
+ {
2342
+ "epoch": 4.895209580838324,
2343
+ "grad_norm": 0.333984375,
2344
+ "learning_rate": 0.00014223537277759666,
2345
+ "loss": 0.5496,
2346
+ "step": 3270
2347
+ },
2348
+ {
2349
+ "epoch": 4.910179640718563,
2350
+ "grad_norm": 0.451171875,
2351
+ "learning_rate": 0.00014191765204166643,
2352
+ "loss": 0.5407,
2353
+ "step": 3280
2354
+ },
2355
+ {
2356
+ "epoch": 4.925149700598802,
2357
+ "grad_norm": 0.3671875,
2358
+ "learning_rate": 0.00014159941713025447,
2359
+ "loss": 0.5703,
2360
+ "step": 3290
2361
+ },
2362
+ {
2363
+ "epoch": 4.940119760479042,
2364
+ "grad_norm": 0.349609375,
2365
+ "learning_rate": 0.00014128067194693316,
2366
+ "loss": 0.5523,
2367
+ "step": 3300
2368
+ },
2369
+ {
2370
+ "epoch": 4.955089820359282,
2371
+ "grad_norm": 0.361328125,
2372
+ "learning_rate": 0.000140961420401534,
2373
+ "loss": 0.5366,
2374
+ "step": 3310
2375
+ },
2376
+ {
2377
+ "epoch": 4.970059880239521,
2378
+ "grad_norm": 0.357421875,
2379
+ "learning_rate": 0.00014064166641009982,
2380
+ "loss": 0.5777,
2381
+ "step": 3320
2382
+ },
2383
+ {
2384
+ "epoch": 4.985029940119761,
2385
+ "grad_norm": 0.318359375,
2386
+ "learning_rate": 0.00014032141389483648,
2387
+ "loss": 0.5649,
2388
+ "step": 3330
2389
+ },
2390
+ {
2391
+ "epoch": 5.0,
2392
+ "grad_norm": 0.56640625,
2393
+ "learning_rate": 0.000140000666784065,
2394
+ "loss": 0.5502,
2395
+ "step": 3340
2396
+ },
2397
+ {
2398
+ "epoch": 5.014970059880239,
2399
+ "grad_norm": 0.373046875,
2400
+ "learning_rate": 0.0001396794290121731,
2401
+ "loss": 0.436,
2402
+ "step": 3350
2403
+ },
2404
+ {
2405
+ "epoch": 5.029940119760479,
2406
+ "grad_norm": 0.34765625,
2407
+ "learning_rate": 0.0001393577045195673,
2408
+ "loss": 0.4013,
2409
+ "step": 3360
2410
+ },
2411
+ {
2412
+ "epoch": 5.044910179640719,
2413
+ "grad_norm": 0.376953125,
2414
+ "learning_rate": 0.00013903549725262433,
2415
+ "loss": 0.4434,
2416
+ "step": 3370
2417
+ },
2418
+ {
2419
+ "epoch": 5.059880239520958,
2420
+ "grad_norm": 0.384765625,
2421
+ "learning_rate": 0.0001387128111636427,
2422
+ "loss": 0.4261,
2423
+ "step": 3380
2424
+ },
2425
+ {
2426
+ "epoch": 5.074850299401198,
2427
+ "grad_norm": 0.400390625,
2428
+ "learning_rate": 0.00013838965021079446,
2429
+ "loss": 0.416,
2430
+ "step": 3390
2431
+ },
2432
+ {
2433
+ "epoch": 5.089820359281437,
2434
+ "grad_norm": 0.400390625,
2435
+ "learning_rate": 0.00013806601835807637,
2436
+ "loss": 0.4205,
2437
+ "step": 3400
2438
+ },
2439
+ {
2440
+ "epoch": 5.104790419161676,
2441
+ "grad_norm": 0.421875,
2442
+ "learning_rate": 0.00013774191957526143,
2443
+ "loss": 0.4286,
2444
+ "step": 3410
2445
+ },
2446
+ {
2447
+ "epoch": 5.119760479041916,
2448
+ "grad_norm": 0.38671875,
2449
+ "learning_rate": 0.0001374173578378502,
2450
+ "loss": 0.4198,
2451
+ "step": 3420
2452
+ },
2453
+ {
2454
+ "epoch": 5.134730538922156,
2455
+ "grad_norm": 0.373046875,
2456
+ "learning_rate": 0.00013709233712702197,
2457
+ "loss": 0.4341,
2458
+ "step": 3430
2459
+ },
2460
+ {
2461
+ "epoch": 5.149700598802395,
2462
+ "grad_norm": 0.40234375,
2463
+ "learning_rate": 0.00013676686142958586,
2464
+ "loss": 0.4431,
2465
+ "step": 3440
2466
+ },
2467
+ {
2468
+ "epoch": 5.164670658682635,
2469
+ "grad_norm": 0.375,
2470
+ "learning_rate": 0.00013644093473793215,
2471
+ "loss": 0.4194,
2472
+ "step": 3450
2473
+ },
2474
+ {
2475
+ "epoch": 5.179640718562874,
2476
+ "grad_norm": 0.41796875,
2477
+ "learning_rate": 0.00013611456104998303,
2478
+ "loss": 0.417,
2479
+ "step": 3460
2480
+ },
2481
+ {
2482
+ "epoch": 5.1946107784431135,
2483
+ "grad_norm": 0.357421875,
2484
+ "learning_rate": 0.0001357877443691438,
2485
+ "loss": 0.4467,
2486
+ "step": 3470
2487
+ },
2488
+ {
2489
+ "epoch": 5.209580838323353,
2490
+ "grad_norm": 0.359375,
2491
+ "learning_rate": 0.00013546048870425356,
2492
+ "loss": 0.4206,
2493
+ "step": 3480
2494
+ },
2495
+ {
2496
+ "epoch": 5.224550898203593,
2497
+ "grad_norm": 0.369140625,
2498
+ "learning_rate": 0.00013513279806953623,
2499
+ "loss": 0.4148,
2500
+ "step": 3490
2501
+ },
2502
+ {
2503
+ "epoch": 5.2395209580838324,
2504
+ "grad_norm": 0.36328125,
2505
+ "learning_rate": 0.0001348046764845511,
2506
+ "loss": 0.415,
2507
+ "step": 3500
2508
+ },
2509
+ {
2510
+ "epoch": 5.2395209580838324,
2511
+ "eval_loss": 0.4033271074295044,
2512
+ "eval_runtime": 109.9729,
2513
+ "eval_samples_per_second": 9.093,
2514
+ "eval_steps_per_second": 1.137,
2515
+ "step": 3500
2516
+ }
2517
+ ],
2518
+ "logging_steps": 10,
2519
+ "max_steps": 9000,
2520
+ "num_input_tokens_seen": 0,
2521
+ "num_train_epochs": 14,
2522
+ "save_steps": 500,
2523
+ "stateful_callbacks": {
2524
+ "TrainerControl": {
2525
+ "args": {
2526
+ "should_epoch_stop": false,
2527
+ "should_evaluate": false,
2528
+ "should_log": false,
2529
+ "should_save": true,
2530
+ "should_training_stop": false
2531
+ },
2532
+ "attributes": {}
2533
+ }
2534
+ },
2535
+ "total_flos": 2.433118987616256e+17,
2536
+ "train_batch_size": 16,
2537
+ "trial_name": null,
2538
+ "trial_params": null
2539
+ }
checkpoint-3500/training_args.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:348daaccb41aadfbde16c1e7cfc25845c19b3830f0c06870086619db652baae1
3
+ size 5176
checkpoint-4000/config.json ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_name_or_path": "nisten/Biggie-SmoLlm-0.4B",
3
+ "architectures": [
4
+ "LlamaForCausalLM"
5
+ ],
6
+ "attention_bias": false,
7
+ "attention_dropout": 0.0,
8
+ "bos_token_id": 1,
9
+ "eos_token_id": 2,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 960,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 2560,
14
+ "max_position_embeddings": 2048,
15
+ "mlp_bias": false,
16
+ "model_type": "llama",
17
+ "num_attention_heads": 15,
18
+ "num_hidden_layers": 36,
19
+ "num_key_value_heads": 5,
20
+ "pad_token_id": 2,
21
+ "pretraining_tp": 1,
22
+ "rms_norm_eps": 1e-05,
23
+ "rope_scaling": null,
24
+ "rope_theta": 10000.0,
25
+ "tie_word_embeddings": true,
26
+ "torch_dtype": "bfloat16",
27
+ "transformers_version": "4.44.0",
28
+ "use_cache": true,
29
+ "vocab_size": 49152
30
+ }
checkpoint-4000/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": 2,
6
+ "transformers_version": "4.44.0"
7
+ }