RGPalgrave commited on
Commit
c2dfde3
·
verified ·
1 Parent(s): d8f6384

TopG: models, corpora and frozen evaluation sets

Browse files
G_1000/arm_meta.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "G_1000",
3
+ "seed": 0,
4
+ "seed_witness": "f2e603df160ea17e",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "92904f66ea256950c2bbb195eb60ffa818910114",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0
69
+ },
70
+ "code_hash": "5fc4b21b97f3bf87f76802b958080d2f77fb3a1728af86bc1f1f5a9ef582fff9",
71
+ "device": "cuda",
72
+ "started_utc": "2026-08-08T00:05:51Z",
73
+ "train_seconds": 5906.0,
74
+ "final_val_loss": 3.00714,
75
+ "final_train_loss": 2.25972,
76
+ "complete": true
77
+ }
G_1000/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4359325fd8d74a9c3def65df5bd57ffdd8d83d0e7c8b97866d2339e870794487
3
+ size 18074210
G_1000/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7676346cc2f72e5696e1a84d4c38c362f5a8fe8b806f91e39db3c6a8e9fc8a1
3
+ size 18074210
G_1000/final_mpw_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1000/final_mpw_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1000/final_strat_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1000/final_strat_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1000/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.583,
7
+ "val_loss": 5.73865,
8
+ "step": 487,
9
+ "seconds": 89.8
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.81209,
14
+ "val_loss": 4.31847,
15
+ "step": 974,
16
+ "seconds": 147.8
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 3.98173,
21
+ "val_loss": 3.57597,
22
+ "step": 1461,
23
+ "seconds": 206.2
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.26284,
28
+ "val_loss": 3.16745,
29
+ "step": 1948,
30
+ "seconds": 264.6
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 2.97619,
35
+ "val_loss": 2.87865,
36
+ "step": 2435,
37
+ "seconds": 323.0
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.87221,
42
+ "val_loss": 2.80395,
43
+ "step": 2922,
44
+ "seconds": 381.3
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.81703,
49
+ "val_loss": 2.77066,
50
+ "step": 3409,
51
+ "seconds": 439.5
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.78248,
56
+ "val_loss": 2.74206,
57
+ "step": 3896,
58
+ "seconds": 497.8
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.75599,
63
+ "val_loss": 2.72244,
64
+ "step": 4383,
65
+ "seconds": 556.0
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.73522,
70
+ "val_loss": 2.70885,
71
+ "step": 4870,
72
+ "seconds": 614.3
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.7181,
77
+ "val_loss": 2.70276,
78
+ "step": 5357,
79
+ "seconds": 672.6
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.70408,
84
+ "val_loss": 2.69593,
85
+ "step": 5844,
86
+ "seconds": 730.9
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.69176,
91
+ "val_loss": 2.69392,
92
+ "step": 6331,
93
+ "seconds": 789.2
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68101,
98
+ "val_loss": 2.6897,
99
+ "step": 6818,
100
+ "seconds": 847.5
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.6707,
105
+ "val_loss": 2.68771,
106
+ "step": 7305,
107
+ "seconds": 905.8
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66202,
112
+ "val_loss": 2.68525,
113
+ "step": 7792,
114
+ "seconds": 964.0
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.65285,
119
+ "val_loss": 2.68985,
120
+ "step": 8279,
121
+ "seconds": 1022.3
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.64443,
126
+ "val_loss": 2.69074,
127
+ "step": 8766,
128
+ "seconds": 1080.6
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.63601,
133
+ "val_loss": 2.68705,
134
+ "step": 9253,
135
+ "seconds": 1138.9
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.62788,
140
+ "val_loss": 2.69046,
141
+ "step": 9740,
142
+ "seconds": 1197.2
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.62038,
147
+ "val_loss": 2.68955,
148
+ "step": 10227,
149
+ "seconds": 1255.5
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.61226,
154
+ "val_loss": 2.69332,
155
+ "step": 10714,
156
+ "seconds": 1313.8
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.6043,
161
+ "val_loss": 2.6995,
162
+ "step": 11201,
163
+ "seconds": 1372.1
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.59666,
168
+ "val_loss": 2.70241,
169
+ "step": 11688,
170
+ "seconds": 1430.3
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.58891,
175
+ "val_loss": 2.70053,
176
+ "step": 12175,
177
+ "seconds": 1488.6
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.58082,
182
+ "val_loss": 2.70501,
183
+ "step": 12662,
184
+ "seconds": 1546.9
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.57366,
189
+ "val_loss": 2.70843,
190
+ "step": 13149,
191
+ "seconds": 1605.2
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56571,
196
+ "val_loss": 2.71839,
197
+ "step": 13636,
198
+ "seconds": 1663.5
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.55769,
203
+ "val_loss": 2.72175,
204
+ "step": 14123,
205
+ "seconds": 1721.8
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.54982,
210
+ "val_loss": 2.72978,
211
+ "step": 14610,
212
+ "seconds": 1780.4
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.54199,
217
+ "val_loss": 2.73616,
218
+ "step": 15097,
219
+ "seconds": 1839.4
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.53412,
224
+ "val_loss": 2.73987,
225
+ "step": 15584,
226
+ "seconds": 1897.8
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.52626,
231
+ "val_loss": 2.74722,
232
+ "step": 16071,
233
+ "seconds": 1956.9
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.51829,
238
+ "val_loss": 2.75203,
239
+ "step": 16558,
240
+ "seconds": 2015.8
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.51075,
245
+ "val_loss": 2.76037,
246
+ "step": 17045,
247
+ "seconds": 2074.6
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.50252,
252
+ "val_loss": 2.77008,
253
+ "step": 17532,
254
+ "seconds": 2133.6
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.49569,
259
+ "val_loss": 2.77211,
260
+ "step": 18019,
261
+ "seconds": 2192.6
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.48801,
266
+ "val_loss": 2.77874,
267
+ "step": 18506,
268
+ "seconds": 2251.5
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.48041,
273
+ "val_loss": 2.78746,
274
+ "step": 18993,
275
+ "seconds": 2310.4
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.47261,
280
+ "val_loss": 2.79219,
281
+ "step": 19480,
282
+ "seconds": 2369.3
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.46538,
287
+ "val_loss": 2.80162,
288
+ "step": 19967,
289
+ "seconds": 2428.3
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.45781,
294
+ "val_loss": 2.80944,
295
+ "step": 20454,
296
+ "seconds": 2487.2
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.45156,
301
+ "val_loss": 2.81224,
302
+ "step": 20941,
303
+ "seconds": 2546.2
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.44386,
308
+ "val_loss": 2.82232,
309
+ "step": 21428,
310
+ "seconds": 2605.1
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.4368,
315
+ "val_loss": 2.8275,
316
+ "step": 21915,
317
+ "seconds": 2664.0
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.42986,
322
+ "val_loss": 2.83269,
323
+ "step": 22402,
324
+ "seconds": 2722.9
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.42313,
329
+ "val_loss": 2.84258,
330
+ "step": 22889,
331
+ "seconds": 2781.9
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.41652,
336
+ "val_loss": 2.84395,
337
+ "step": 23376,
338
+ "seconds": 2840.7
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.41013,
343
+ "val_loss": 2.85453,
344
+ "step": 23863,
345
+ "seconds": 2899.7
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.40327,
350
+ "val_loss": 2.85728,
351
+ "step": 24350,
352
+ "seconds": 2958.6
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.39771,
357
+ "val_loss": 2.86549,
358
+ "step": 24837,
359
+ "seconds": 3017.6
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.39171,
364
+ "val_loss": 2.87349,
365
+ "step": 25324,
366
+ "seconds": 3076.6
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.38563,
371
+ "val_loss": 2.87585,
372
+ "step": 25811,
373
+ "seconds": 3135.5
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.38034,
378
+ "val_loss": 2.88119,
379
+ "step": 26298,
380
+ "seconds": 3194.4
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.37388,
385
+ "val_loss": 2.88951,
386
+ "step": 26785,
387
+ "seconds": 3253.3
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.36843,
392
+ "val_loss": 2.89645,
393
+ "step": 27272,
394
+ "seconds": 3312.2
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.36331,
399
+ "val_loss": 2.90431,
400
+ "step": 27759,
401
+ "seconds": 3371.1
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.35819,
406
+ "val_loss": 2.91036,
407
+ "step": 28246,
408
+ "seconds": 3430.1
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.3528,
413
+ "val_loss": 2.9104,
414
+ "step": 28733,
415
+ "seconds": 3488.9
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.34763,
420
+ "val_loss": 2.92087,
421
+ "step": 29220,
422
+ "seconds": 3547.8
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.34318,
427
+ "val_loss": 2.92455,
428
+ "step": 29707,
429
+ "seconds": 3606.8
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.33851,
434
+ "val_loss": 2.92747,
435
+ "step": 30194,
436
+ "seconds": 3665.8
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.33357,
441
+ "val_loss": 2.93269,
442
+ "step": 30681,
443
+ "seconds": 3724.9
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.32986,
448
+ "val_loss": 2.93678,
449
+ "step": 31168,
450
+ "seconds": 3783.8
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.32536,
455
+ "val_loss": 2.94073,
456
+ "step": 31655,
457
+ "seconds": 3842.6
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.32137,
462
+ "val_loss": 2.9426,
463
+ "step": 32142,
464
+ "seconds": 3901.5
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.31733,
469
+ "val_loss": 2.94819,
470
+ "step": 32629,
471
+ "seconds": 3960.4
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.31306,
476
+ "val_loss": 2.95235,
477
+ "step": 33116,
478
+ "seconds": 4019.3
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.30905,
483
+ "val_loss": 2.96116,
484
+ "step": 33603,
485
+ "seconds": 4078.2
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.30686,
490
+ "val_loss": 2.96463,
491
+ "step": 34090,
492
+ "seconds": 4137.2
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.30268,
497
+ "val_loss": 2.96882,
498
+ "step": 34577,
499
+ "seconds": 4196.2
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.29974,
504
+ "val_loss": 2.96727,
505
+ "step": 35064,
506
+ "seconds": 4255.1
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.29638,
511
+ "val_loss": 2.97228,
512
+ "step": 35551,
513
+ "seconds": 4314.0
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.29334,
518
+ "val_loss": 2.9746,
519
+ "step": 36038,
520
+ "seconds": 4373.0
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.29093,
525
+ "val_loss": 2.97853,
526
+ "step": 36525,
527
+ "seconds": 4431.9
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.2881,
532
+ "val_loss": 2.97904,
533
+ "step": 37012,
534
+ "seconds": 4491.1
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.28533,
539
+ "val_loss": 2.98366,
540
+ "step": 37499,
541
+ "seconds": 4549.9
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.28264,
546
+ "val_loss": 2.98544,
547
+ "step": 37986,
548
+ "seconds": 4608.7
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.28044,
553
+ "val_loss": 2.9888,
554
+ "step": 38473,
555
+ "seconds": 4667.6
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.2783,
560
+ "val_loss": 2.99109,
561
+ "step": 38960,
562
+ "seconds": 4726.5
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.27612,
567
+ "val_loss": 2.9922,
568
+ "step": 39447,
569
+ "seconds": 4785.4
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.27429,
574
+ "val_loss": 2.99322,
575
+ "step": 39934,
576
+ "seconds": 4844.6
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.27285,
581
+ "val_loss": 2.99725,
582
+ "step": 40421,
583
+ "seconds": 4903.8
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.27056,
588
+ "val_loss": 2.99744,
589
+ "step": 40908,
590
+ "seconds": 4962.7
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.26879,
595
+ "val_loss": 2.99919,
596
+ "step": 41395,
597
+ "seconds": 5021.7
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.26855,
602
+ "val_loss": 3.00067,
603
+ "step": 41882,
604
+ "seconds": 5080.6
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.26657,
609
+ "val_loss": 3.00195,
610
+ "step": 42369,
611
+ "seconds": 5139.6
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.26504,
616
+ "val_loss": 3.0042,
617
+ "step": 42856,
618
+ "seconds": 5198.5
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.26482,
623
+ "val_loss": 3.00389,
624
+ "step": 43343,
625
+ "seconds": 5257.4
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.26437,
630
+ "val_loss": 3.00393,
631
+ "step": 43830,
632
+ "seconds": 5316.3
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.26234,
637
+ "val_loss": 3.00506,
638
+ "step": 44317,
639
+ "seconds": 5375.2
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.26181,
644
+ "val_loss": 3.00604,
645
+ "step": 44804,
646
+ "seconds": 5434.2
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.26104,
651
+ "val_loss": 3.00601,
652
+ "step": 45291,
653
+ "seconds": 5493.2
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.26116,
658
+ "val_loss": 3.00658,
659
+ "step": 45778,
660
+ "seconds": 5552.1
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.26052,
665
+ "val_loss": 3.00676,
666
+ "step": 46265,
667
+ "seconds": 5611.0
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.26047,
672
+ "val_loss": 3.00668,
673
+ "step": 46752,
674
+ "seconds": 5669.9
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.2602,
679
+ "val_loss": 3.00714,
680
+ "step": 47239,
681
+ "seconds": 5728.9
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.25933,
686
+ "val_loss": 3.00722,
687
+ "step": 47726,
688
+ "seconds": 5788.1
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.25974,
693
+ "val_loss": 3.00714,
694
+ "step": 48213,
695
+ "seconds": 5847.1
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.25972,
700
+ "val_loss": 3.00714,
701
+ "step": 48700,
702
+ "seconds": 5906.0
703
+ }
704
+ ]
705
+ }
G_1001/arm_meta.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "G_1001",
3
+ "seed": 1,
4
+ "seed_witness": "0fbe2a188322a36c",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "92904f66ea256950c2bbb195eb60ffa818910114",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0
69
+ },
70
+ "code_hash": "5fc4b21b97f3bf87f76802b958080d2f77fb3a1728af86bc1f1f5a9ef582fff9",
71
+ "device": "cuda",
72
+ "started_utc": "2026-08-08T01:44:22Z",
73
+ "train_seconds": 5866.2,
74
+ "final_val_loss": 3.01719,
75
+ "final_train_loss": 2.26076,
76
+ "complete": true
77
+ }
G_1001/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:239f1ec867f729d37a2bbc6dd2883ac5d0780e9d1f2484ec99d1dc52d79a408d
3
+ size 18074210
G_1001/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:435fc46147362762731c79af033b86ec79b8f5a1217a648a9d6e9a64842bb021
3
+ size 18074210
G_1001/final_mpw_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1001/final_mpw_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1001/final_strat_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1001/final_strat_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1001/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.61778,
7
+ "val_loss": 5.72146,
8
+ "step": 487,
9
+ "seconds": 91.8
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.78952,
14
+ "val_loss": 4.30559,
15
+ "step": 974,
16
+ "seconds": 150.6
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 3.85531,
21
+ "val_loss": 3.41964,
22
+ "step": 1461,
23
+ "seconds": 209.8
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.20887,
28
+ "val_loss": 3.04176,
29
+ "step": 1948,
30
+ "seconds": 268.5
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 2.98373,
35
+ "val_loss": 2.8969,
36
+ "step": 2435,
37
+ "seconds": 327.3
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.88209,
42
+ "val_loss": 2.81961,
43
+ "step": 2922,
44
+ "seconds": 386.0
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.82441,
49
+ "val_loss": 2.77736,
50
+ "step": 3409,
51
+ "seconds": 444.8
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.78711,
56
+ "val_loss": 2.74955,
57
+ "step": 3896,
58
+ "seconds": 503.6
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.75924,
63
+ "val_loss": 2.73165,
64
+ "step": 4383,
65
+ "seconds": 562.3
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.73803,
70
+ "val_loss": 2.71602,
71
+ "step": 4870,
72
+ "seconds": 621.0
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.71935,
77
+ "val_loss": 2.70627,
78
+ "step": 5357,
79
+ "seconds": 680.0
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.70482,
84
+ "val_loss": 2.70027,
85
+ "step": 5844,
86
+ "seconds": 738.8
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.6927,
91
+ "val_loss": 2.69605,
92
+ "step": 6331,
93
+ "seconds": 797.6
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68123,
98
+ "val_loss": 2.69602,
99
+ "step": 6818,
100
+ "seconds": 856.4
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.67199,
105
+ "val_loss": 2.69094,
106
+ "step": 7305,
107
+ "seconds": 915.1
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66138,
112
+ "val_loss": 2.69077,
113
+ "step": 7792,
114
+ "seconds": 973.7
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.65311,
119
+ "val_loss": 2.69299,
120
+ "step": 8279,
121
+ "seconds": 1032.5
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.64459,
126
+ "val_loss": 2.69154,
127
+ "step": 8766,
128
+ "seconds": 1091.1
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.63579,
133
+ "val_loss": 2.6913,
134
+ "step": 9253,
135
+ "seconds": 1149.7
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.62791,
140
+ "val_loss": 2.69474,
141
+ "step": 9740,
142
+ "seconds": 1208.3
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.61988,
147
+ "val_loss": 2.69583,
148
+ "step": 10227,
149
+ "seconds": 1267.0
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.61222,
154
+ "val_loss": 2.69889,
155
+ "step": 10714,
156
+ "seconds": 1325.7
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.60414,
161
+ "val_loss": 2.70524,
162
+ "step": 11201,
163
+ "seconds": 1384.4
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.59621,
168
+ "val_loss": 2.70768,
169
+ "step": 11688,
170
+ "seconds": 1442.8
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.58918,
175
+ "val_loss": 2.71402,
176
+ "step": 12175,
177
+ "seconds": 1500.9
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.58082,
182
+ "val_loss": 2.71735,
183
+ "step": 12662,
184
+ "seconds": 1558.9
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.57311,
189
+ "val_loss": 2.72063,
190
+ "step": 13149,
191
+ "seconds": 1617.0
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56502,
196
+ "val_loss": 2.73205,
197
+ "step": 13636,
198
+ "seconds": 1675.1
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.55776,
203
+ "val_loss": 2.73138,
204
+ "step": 14123,
205
+ "seconds": 1733.2
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.54987,
210
+ "val_loss": 2.74048,
211
+ "step": 14610,
212
+ "seconds": 1791.4
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.54135,
217
+ "val_loss": 2.74199,
218
+ "step": 15097,
219
+ "seconds": 1849.6
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.53398,
224
+ "val_loss": 2.75066,
225
+ "step": 15584,
226
+ "seconds": 1907.7
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.52623,
231
+ "val_loss": 2.75505,
232
+ "step": 16071,
233
+ "seconds": 1965.9
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.5179,
238
+ "val_loss": 2.76462,
239
+ "step": 16558,
240
+ "seconds": 2024.0
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.51023,
245
+ "val_loss": 2.7733,
246
+ "step": 17045,
247
+ "seconds": 2082.2
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.50272,
252
+ "val_loss": 2.77695,
253
+ "step": 17532,
254
+ "seconds": 2140.4
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.49496,
259
+ "val_loss": 2.78576,
260
+ "step": 18019,
261
+ "seconds": 2198.5
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.48742,
266
+ "val_loss": 2.79286,
267
+ "step": 18506,
268
+ "seconds": 2256.6
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.47975,
273
+ "val_loss": 2.79609,
274
+ "step": 18993,
275
+ "seconds": 2314.8
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.47246,
280
+ "val_loss": 2.8053,
281
+ "step": 19480,
282
+ "seconds": 2372.9
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.46499,
287
+ "val_loss": 2.81234,
288
+ "step": 19967,
289
+ "seconds": 2431.1
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.45764,
294
+ "val_loss": 2.82093,
295
+ "step": 20454,
296
+ "seconds": 2489.2
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.45091,
301
+ "val_loss": 2.82248,
302
+ "step": 20941,
303
+ "seconds": 2547.4
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.44354,
308
+ "val_loss": 2.83363,
309
+ "step": 21428,
310
+ "seconds": 2605.6
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.43693,
315
+ "val_loss": 2.83515,
316
+ "step": 21915,
317
+ "seconds": 2663.7
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.42963,
322
+ "val_loss": 2.84556,
323
+ "step": 22402,
324
+ "seconds": 2721.9
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.42373,
329
+ "val_loss": 2.8531,
330
+ "step": 22889,
331
+ "seconds": 2780.1
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.41673,
336
+ "val_loss": 2.85673,
337
+ "step": 23376,
338
+ "seconds": 2838.2
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.41022,
343
+ "val_loss": 2.8645,
344
+ "step": 23863,
345
+ "seconds": 2896.4
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.40367,
350
+ "val_loss": 2.86999,
351
+ "step": 24350,
352
+ "seconds": 2954.5
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.39753,
357
+ "val_loss": 2.87501,
358
+ "step": 24837,
359
+ "seconds": 3012.7
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.39231,
364
+ "val_loss": 2.88779,
365
+ "step": 25324,
366
+ "seconds": 3070.9
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.38598,
371
+ "val_loss": 2.889,
372
+ "step": 25811,
373
+ "seconds": 3129.0
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.37974,
378
+ "val_loss": 2.89409,
379
+ "step": 26298,
380
+ "seconds": 3187.2
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.37549,
385
+ "val_loss": 2.90003,
386
+ "step": 26785,
387
+ "seconds": 3245.3
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.36899,
392
+ "val_loss": 2.9014,
393
+ "step": 27272,
394
+ "seconds": 3303.4
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.36401,
399
+ "val_loss": 2.90702,
400
+ "step": 27759,
401
+ "seconds": 3361.6
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.35845,
406
+ "val_loss": 2.9154,
407
+ "step": 28246,
408
+ "seconds": 3419.7
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.35403,
413
+ "val_loss": 2.923,
414
+ "step": 28733,
415
+ "seconds": 3477.9
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.34896,
420
+ "val_loss": 2.92916,
421
+ "step": 29220,
422
+ "seconds": 3536.0
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.3444,
427
+ "val_loss": 2.93306,
428
+ "step": 29707,
429
+ "seconds": 3594.2
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.33952,
434
+ "val_loss": 2.93763,
435
+ "step": 30194,
436
+ "seconds": 3652.6
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.33488,
441
+ "val_loss": 2.93933,
442
+ "step": 30681,
443
+ "seconds": 3710.9
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.33041,
448
+ "val_loss": 2.94606,
449
+ "step": 31168,
450
+ "seconds": 3769.0
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.32614,
455
+ "val_loss": 2.95005,
456
+ "step": 31655,
457
+ "seconds": 3827.3
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.32222,
462
+ "val_loss": 2.95799,
463
+ "step": 32142,
464
+ "seconds": 3885.9
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.31837,
469
+ "val_loss": 2.9542,
470
+ "step": 32629,
471
+ "seconds": 3944.7
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.31395,
476
+ "val_loss": 2.96708,
477
+ "step": 33116,
478
+ "seconds": 4003.5
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.3105,
483
+ "val_loss": 2.96778,
484
+ "step": 33603,
485
+ "seconds": 4062.3
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.30793,
490
+ "val_loss": 2.96953,
491
+ "step": 34090,
492
+ "seconds": 4120.4
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.30363,
497
+ "val_loss": 2.97372,
498
+ "step": 34577,
499
+ "seconds": 4178.6
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.30065,
504
+ "val_loss": 2.9794,
505
+ "step": 35064,
506
+ "seconds": 4236.7
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.29794,
511
+ "val_loss": 2.98139,
512
+ "step": 35551,
513
+ "seconds": 4294.9
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.29461,
518
+ "val_loss": 2.98455,
519
+ "step": 36038,
520
+ "seconds": 4353.1
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.29191,
525
+ "val_loss": 2.98784,
526
+ "step": 36525,
527
+ "seconds": 4411.2
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.28901,
532
+ "val_loss": 2.99264,
533
+ "step": 37012,
534
+ "seconds": 4469.4
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.28647,
539
+ "val_loss": 2.99208,
540
+ "step": 37499,
541
+ "seconds": 4527.5
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.28423,
546
+ "val_loss": 2.99644,
547
+ "step": 37986,
548
+ "seconds": 4585.6
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.28187,
553
+ "val_loss": 2.99895,
554
+ "step": 38473,
555
+ "seconds": 4643.7
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.27973,
560
+ "val_loss": 3.0014,
561
+ "step": 38960,
562
+ "seconds": 4701.9
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.27821,
567
+ "val_loss": 3.00486,
568
+ "step": 39447,
569
+ "seconds": 4760.1
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.2753,
574
+ "val_loss": 3.00551,
575
+ "step": 39934,
576
+ "seconds": 4818.2
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.27352,
581
+ "val_loss": 3.00769,
582
+ "step": 40421,
583
+ "seconds": 4876.4
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.27248,
588
+ "val_loss": 3.00658,
589
+ "step": 40908,
590
+ "seconds": 4934.5
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.27131,
595
+ "val_loss": 3.00735,
596
+ "step": 41395,
597
+ "seconds": 4992.7
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.26958,
602
+ "val_loss": 3.01102,
603
+ "step": 41882,
604
+ "seconds": 5050.8
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.26818,
609
+ "val_loss": 3.01151,
610
+ "step": 42369,
611
+ "seconds": 5109.0
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.26683,
616
+ "val_loss": 3.01272,
617
+ "step": 42856,
618
+ "seconds": 5167.1
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.26568,
623
+ "val_loss": 3.01354,
624
+ "step": 43343,
625
+ "seconds": 5225.3
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.2657,
630
+ "val_loss": 3.01507,
631
+ "step": 43830,
632
+ "seconds": 5283.5
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.26447,
637
+ "val_loss": 3.01471,
638
+ "step": 44317,
639
+ "seconds": 5341.7
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.26347,
644
+ "val_loss": 3.01624,
645
+ "step": 44804,
646
+ "seconds": 5399.8
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.26303,
651
+ "val_loss": 3.01643,
652
+ "step": 45291,
653
+ "seconds": 5458.0
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.26295,
658
+ "val_loss": 3.01635,
659
+ "step": 45778,
660
+ "seconds": 5516.1
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.26161,
665
+ "val_loss": 3.01684,
666
+ "step": 46265,
667
+ "seconds": 5574.7
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.26181,
672
+ "val_loss": 3.01723,
673
+ "step": 46752,
674
+ "seconds": 5633.5
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.26134,
679
+ "val_loss": 3.017,
680
+ "step": 47239,
681
+ "seconds": 5691.7
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.26178,
686
+ "val_loss": 3.01713,
687
+ "step": 47726,
688
+ "seconds": 5749.9
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.26094,
693
+ "val_loss": 3.01718,
694
+ "step": 48213,
695
+ "seconds": 5808.0
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.26076,
700
+ "val_loss": 3.01719,
701
+ "step": 48700,
702
+ "seconds": 5866.2
703
+ }
704
+ ]
705
+ }
G_1002/arm_meta.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "G_1002",
3
+ "seed": 2,
4
+ "seed_witness": "f2a8e585693907e0",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "92904f66ea256950c2bbb195eb60ffa818910114",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0
69
+ },
70
+ "code_hash": "5fc4b21b97f3bf87f76802b958080d2f77fb3a1728af86bc1f1f5a9ef582fff9",
71
+ "device": "cuda",
72
+ "started_utc": "2026-08-08T03:22:11Z",
73
+ "train_seconds": 5875.2,
74
+ "final_val_loss": 2.99975,
75
+ "final_train_loss": 2.26627,
76
+ "complete": true
77
+ }
G_1002/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c50ac0288490f96c6ee4f91777d8d3e3a82773c3c04a7de4edf5a7207b9bed6b
3
+ size 18074210
G_1002/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9bb0b3e51bdf8f79a113cd6da70f6130debf66d26cf9e4e0df740e3d59876af5
3
+ size 18074210
G_1002/final_mpw_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1002/final_mpw_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1002/final_strat_ep00020.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1002/final_strat_ep00100.json ADDED
The diff for this file is too large to render. See raw diff
 
G_1002/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.57118,
7
+ "val_loss": 5.68386,
8
+ "step": 487,
9
+ "seconds": 90.7
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.77557,
14
+ "val_loss": 4.30811,
15
+ "step": 974,
16
+ "seconds": 148.9
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 3.88646,
21
+ "val_loss": 3.46749,
22
+ "step": 1461,
23
+ "seconds": 207.2
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.23434,
28
+ "val_loss": 3.05744,
29
+ "step": 1948,
30
+ "seconds": 265.5
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 2.99848,
35
+ "val_loss": 2.89707,
36
+ "step": 2435,
37
+ "seconds": 323.8
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.88995,
42
+ "val_loss": 2.82589,
43
+ "step": 2922,
44
+ "seconds": 382.1
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.83251,
49
+ "val_loss": 2.78156,
50
+ "step": 3409,
51
+ "seconds": 440.3
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.79326,
56
+ "val_loss": 2.75481,
57
+ "step": 3896,
58
+ "seconds": 498.5
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.76497,
63
+ "val_loss": 2.73642,
64
+ "step": 4383,
65
+ "seconds": 557.3
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.74263,
70
+ "val_loss": 2.71975,
71
+ "step": 4870,
72
+ "seconds": 616.1
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.72443,
77
+ "val_loss": 2.71014,
78
+ "step": 5357,
79
+ "seconds": 674.5
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.70871,
84
+ "val_loss": 2.70312,
85
+ "step": 5844,
86
+ "seconds": 732.7
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.69538,
91
+ "val_loss": 2.69869,
92
+ "step": 6331,
93
+ "seconds": 790.9
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68292,
98
+ "val_loss": 2.70207,
99
+ "step": 6818,
100
+ "seconds": 849.1
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.67269,
105
+ "val_loss": 2.69139,
106
+ "step": 7305,
107
+ "seconds": 907.2
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66231,
112
+ "val_loss": 2.6957,
113
+ "step": 7792,
114
+ "seconds": 966.1
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.65291,
119
+ "val_loss": 2.68774,
120
+ "step": 8279,
121
+ "seconds": 1024.9
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.6448,
126
+ "val_loss": 2.6884,
127
+ "step": 8766,
128
+ "seconds": 1083.4
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.63625,
133
+ "val_loss": 2.6883,
134
+ "step": 9253,
135
+ "seconds": 1141.6
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.6274,
140
+ "val_loss": 2.69379,
141
+ "step": 9740,
142
+ "seconds": 1199.8
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.61892,
147
+ "val_loss": 2.69147,
148
+ "step": 10227,
149
+ "seconds": 1258.0
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.6115,
154
+ "val_loss": 2.69116,
155
+ "step": 10714,
156
+ "seconds": 1316.3
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.60371,
161
+ "val_loss": 2.69821,
162
+ "step": 11201,
163
+ "seconds": 1374.5
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.59497,
168
+ "val_loss": 2.69827,
169
+ "step": 11688,
170
+ "seconds": 1432.7
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.58748,
175
+ "val_loss": 2.70323,
176
+ "step": 12175,
177
+ "seconds": 1491.3
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.57955,
182
+ "val_loss": 2.7051,
183
+ "step": 12662,
184
+ "seconds": 1550.1
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.57184,
189
+ "val_loss": 2.71337,
190
+ "step": 13149,
191
+ "seconds": 1608.9
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56477,
196
+ "val_loss": 2.71505,
197
+ "step": 13636,
198
+ "seconds": 1667.8
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.55626,
203
+ "val_loss": 2.72272,
204
+ "step": 14123,
205
+ "seconds": 1726.3
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.54888,
210
+ "val_loss": 2.72843,
211
+ "step": 14610,
212
+ "seconds": 1785.2
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.54115,
217
+ "val_loss": 2.73334,
218
+ "step": 15097,
219
+ "seconds": 1844.0
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.53307,
224
+ "val_loss": 2.74198,
225
+ "step": 15584,
226
+ "seconds": 1902.8
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.52546,
231
+ "val_loss": 2.7469,
232
+ "step": 16071,
233
+ "seconds": 1961.6
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.51839,
238
+ "val_loss": 2.74754,
239
+ "step": 16558,
240
+ "seconds": 2019.8
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.51064,
245
+ "val_loss": 2.75804,
246
+ "step": 17045,
247
+ "seconds": 2078.0
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.50271,
252
+ "val_loss": 2.76027,
253
+ "step": 17532,
254
+ "seconds": 2136.2
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.49565,
259
+ "val_loss": 2.76707,
260
+ "step": 18019,
261
+ "seconds": 2194.4
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.48769,
266
+ "val_loss": 2.77691,
267
+ "step": 18506,
268
+ "seconds": 2252.7
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.48106,
273
+ "val_loss": 2.78255,
274
+ "step": 18993,
275
+ "seconds": 2311.4
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.47326,
280
+ "val_loss": 2.79101,
281
+ "step": 19480,
282
+ "seconds": 2369.6
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.46562,
287
+ "val_loss": 2.79823,
288
+ "step": 19967,
289
+ "seconds": 2427.8
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.45969,
294
+ "val_loss": 2.80488,
295
+ "step": 20454,
296
+ "seconds": 2486.0
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.45256,
301
+ "val_loss": 2.81245,
302
+ "step": 20941,
303
+ "seconds": 2544.2
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.44584,
308
+ "val_loss": 2.81468,
309
+ "step": 21428,
310
+ "seconds": 2602.4
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.43871,
315
+ "val_loss": 2.82426,
316
+ "step": 21915,
317
+ "seconds": 2661.0
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.43204,
322
+ "val_loss": 2.83065,
323
+ "step": 22402,
324
+ "seconds": 2719.2
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.42601,
329
+ "val_loss": 2.83542,
330
+ "step": 22889,
331
+ "seconds": 2777.5
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.41928,
336
+ "val_loss": 2.84293,
337
+ "step": 23376,
338
+ "seconds": 2835.7
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.41307,
343
+ "val_loss": 2.8487,
344
+ "step": 23863,
345
+ "seconds": 2893.9
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.40739,
350
+ "val_loss": 2.8559,
351
+ "step": 24350,
352
+ "seconds": 2952.2
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.4006,
357
+ "val_loss": 2.86159,
358
+ "step": 24837,
359
+ "seconds": 3010.5
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.39532,
364
+ "val_loss": 2.86686,
365
+ "step": 25324,
366
+ "seconds": 3068.7
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.38921,
371
+ "val_loss": 2.87503,
372
+ "step": 25811,
373
+ "seconds": 3127.0
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.38343,
378
+ "val_loss": 2.87819,
379
+ "step": 26298,
380
+ "seconds": 3185.3
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.37833,
385
+ "val_loss": 2.88972,
386
+ "step": 26785,
387
+ "seconds": 3243.6
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.37216,
392
+ "val_loss": 2.89207,
393
+ "step": 27272,
394
+ "seconds": 3301.8
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.36717,
399
+ "val_loss": 2.89636,
400
+ "step": 27759,
401
+ "seconds": 3360.1
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.36185,
406
+ "val_loss": 2.90377,
407
+ "step": 28246,
408
+ "seconds": 3418.4
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.35659,
413
+ "val_loss": 2.91117,
414
+ "step": 28733,
415
+ "seconds": 3476.6
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.35198,
420
+ "val_loss": 2.9137,
421
+ "step": 29220,
422
+ "seconds": 3534.9
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.34745,
427
+ "val_loss": 2.92041,
428
+ "step": 29707,
429
+ "seconds": 3593.2
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.34284,
434
+ "val_loss": 2.92382,
435
+ "step": 30194,
436
+ "seconds": 3651.5
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.33857,
441
+ "val_loss": 2.92718,
442
+ "step": 30681,
443
+ "seconds": 3709.7
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.33389,
448
+ "val_loss": 2.93668,
449
+ "step": 31168,
450
+ "seconds": 3768.0
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.3305,
455
+ "val_loss": 2.93578,
456
+ "step": 31655,
457
+ "seconds": 3826.3
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.3254,
462
+ "val_loss": 2.94259,
463
+ "step": 32142,
464
+ "seconds": 3884.6
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.32189,
469
+ "val_loss": 2.94293,
470
+ "step": 32629,
471
+ "seconds": 3942.9
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.31825,
476
+ "val_loss": 2.94895,
477
+ "step": 33116,
478
+ "seconds": 4001.2
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.31508,
483
+ "val_loss": 2.95163,
484
+ "step": 33603,
485
+ "seconds": 4059.5
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.31137,
490
+ "val_loss": 2.9574,
491
+ "step": 34090,
492
+ "seconds": 4117.8
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.30807,
497
+ "val_loss": 2.96189,
498
+ "step": 34577,
499
+ "seconds": 4176.1
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.30451,
504
+ "val_loss": 2.96145,
505
+ "step": 35064,
506
+ "seconds": 4234.5
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.30155,
511
+ "val_loss": 2.96441,
512
+ "step": 35551,
513
+ "seconds": 4293.1
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.29864,
518
+ "val_loss": 2.96674,
519
+ "step": 36038,
520
+ "seconds": 4351.3
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.296,
525
+ "val_loss": 2.97053,
526
+ "step": 36525,
527
+ "seconds": 4409.9
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.29319,
532
+ "val_loss": 2.97432,
533
+ "step": 37012,
534
+ "seconds": 4468.1
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.29033,
539
+ "val_loss": 2.97786,
540
+ "step": 37499,
541
+ "seconds": 4526.4
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.2886,
546
+ "val_loss": 2.97902,
547
+ "step": 37986,
548
+ "seconds": 4584.8
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.28597,
553
+ "val_loss": 2.98057,
554
+ "step": 38473,
555
+ "seconds": 4643.2
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.28352,
560
+ "val_loss": 2.98394,
561
+ "step": 38960,
562
+ "seconds": 4701.7
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.2823,
567
+ "val_loss": 2.98568,
568
+ "step": 39447,
569
+ "seconds": 4760.6
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.28074,
574
+ "val_loss": 2.98751,
575
+ "step": 39934,
576
+ "seconds": 4819.5
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.278,
581
+ "val_loss": 2.99154,
582
+ "step": 40421,
583
+ "seconds": 4878.2
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.27695,
588
+ "val_loss": 2.9919,
589
+ "step": 40908,
590
+ "seconds": 4937.0
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.27481,
595
+ "val_loss": 2.99076,
596
+ "step": 41395,
597
+ "seconds": 4995.9
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.27406,
602
+ "val_loss": 2.99365,
603
+ "step": 41882,
604
+ "seconds": 5054.8
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.27277,
609
+ "val_loss": 2.9939,
610
+ "step": 42369,
611
+ "seconds": 5113.6
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.27139,
616
+ "val_loss": 2.99308,
617
+ "step": 42856,
618
+ "seconds": 5172.5
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.26991,
623
+ "val_loss": 2.99653,
624
+ "step": 43343,
625
+ "seconds": 5231.3
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.26884,
630
+ "val_loss": 2.99781,
631
+ "step": 43830,
632
+ "seconds": 5290.2
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.26833,
637
+ "val_loss": 2.99886,
638
+ "step": 44317,
639
+ "seconds": 5348.5
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.26844,
644
+ "val_loss": 2.99856,
645
+ "step": 44804,
646
+ "seconds": 5407.1
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.26724,
651
+ "val_loss": 2.9981,
652
+ "step": 45291,
653
+ "seconds": 5465.5
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.26737,
658
+ "val_loss": 2.99856,
659
+ "step": 45778,
660
+ "seconds": 5523.8
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.26628,
665
+ "val_loss": 2.99942,
666
+ "step": 46265,
667
+ "seconds": 5582.1
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.26576,
672
+ "val_loss": 2.99955,
673
+ "step": 46752,
674
+ "seconds": 5640.4
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.26588,
679
+ "val_loss": 2.99971,
680
+ "step": 47239,
681
+ "seconds": 5698.7
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.26577,
686
+ "val_loss": 2.99978,
687
+ "step": 47726,
688
+ "seconds": 5757.5
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.26517,
693
+ "val_loss": 2.99978,
694
+ "step": 48213,
695
+ "seconds": 5816.4
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.26627,
700
+ "val_loss": 2.99975,
701
+ "step": 48700,
702
+ "seconds": 5875.2
703
+ }
704
+ ]
705
+ }
H_pq0/arm_meta.json ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "H_pq0",
3
+ "seed": 0,
4
+ "seed_witness": "f2e603df160ea17e",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "6508c0fb1ee7e821b7d09b9e8b249e2e49413f5b",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0,
69
+ "partial_queries": true,
70
+ "depth_distribution": {
71
+ "6": 0.5,
72
+ "5": 0.125,
73
+ "4": 0.125,
74
+ "3": 0.125,
75
+ "2": 0.125
76
+ }
77
+ },
78
+ "code_hash": "3084c3b83d82c8cc05d1f2d72ac14f5cfdf134571152e11c3f747c5d9cdc0389",
79
+ "device": "cuda",
80
+ "started_utc": "2026-08-08T13:27:10Z",
81
+ "train_seconds": 6429.1,
82
+ "final_val_loss": 3.01848,
83
+ "final_train_loss": 2.26914,
84
+ "complete": true
85
+ }
H_pq0/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:75bdfa1193ced1da08004b1f068f1081f836a541c273152c964478b62eb780d2
3
+ size 18080482
H_pq0/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f8457297646ba3d9c997579b7683ec1c421f9b5ade1f6114df820a6911abce9f
3
+ size 18080482
H_pq0/depth2.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq0/depth3.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq0/depth4.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq0/depth6.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq0/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.60376,
7
+ "val_loss": 5.72723,
8
+ "step": 487,
9
+ "seconds": 100.1
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.79664,
14
+ "val_loss": 4.31218,
15
+ "step": 974,
16
+ "seconds": 163.5
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 3.85119,
21
+ "val_loss": 3.42853,
22
+ "step": 1461,
23
+ "seconds": 227.3
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.21162,
28
+ "val_loss": 3.0214,
29
+ "step": 1948,
30
+ "seconds": 291.2
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 2.97222,
35
+ "val_loss": 2.87894,
36
+ "step": 2435,
37
+ "seconds": 355.6
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.87671,
42
+ "val_loss": 2.81212,
43
+ "step": 2922,
44
+ "seconds": 419.3
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.82283,
49
+ "val_loss": 2.77722,
50
+ "step": 3409,
51
+ "seconds": 483.1
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.78776,
56
+ "val_loss": 2.75321,
57
+ "step": 3896,
58
+ "seconds": 546.9
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.76286,
63
+ "val_loss": 2.73429,
64
+ "step": 4383,
65
+ "seconds": 611.1
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.74129,
70
+ "val_loss": 2.72334,
71
+ "step": 4870,
72
+ "seconds": 675.7
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.72521,
77
+ "val_loss": 2.71357,
78
+ "step": 5357,
79
+ "seconds": 739.9
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.7111,
84
+ "val_loss": 2.70595,
85
+ "step": 5844,
86
+ "seconds": 804.3
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.69707,
91
+ "val_loss": 2.69997,
92
+ "step": 6331,
93
+ "seconds": 868.5
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68646,
98
+ "val_loss": 2.69392,
99
+ "step": 6818,
100
+ "seconds": 932.6
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.67591,
105
+ "val_loss": 2.69771,
106
+ "step": 7305,
107
+ "seconds": 996.8
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66656,
112
+ "val_loss": 2.69503,
113
+ "step": 7792,
114
+ "seconds": 1061.8
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.65716,
119
+ "val_loss": 2.69058,
120
+ "step": 8279,
121
+ "seconds": 1126.2
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.64848,
126
+ "val_loss": 2.69955,
127
+ "step": 8766,
128
+ "seconds": 1190.7
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.64033,
133
+ "val_loss": 2.69179,
134
+ "step": 9253,
135
+ "seconds": 1255.4
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.6322,
140
+ "val_loss": 2.69892,
141
+ "step": 9740,
142
+ "seconds": 1320.0
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.62436,
147
+ "val_loss": 2.70017,
148
+ "step": 10227,
149
+ "seconds": 1384.7
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.61612,
154
+ "val_loss": 2.70112,
155
+ "step": 10714,
156
+ "seconds": 1449.3
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.60789,
161
+ "val_loss": 2.70271,
162
+ "step": 11201,
163
+ "seconds": 1513.9
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.60117,
168
+ "val_loss": 2.70522,
169
+ "step": 11688,
170
+ "seconds": 1578.5
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.59306,
175
+ "val_loss": 2.71126,
176
+ "step": 12175,
177
+ "seconds": 1643.2
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.58565,
182
+ "val_loss": 2.71542,
183
+ "step": 12662,
184
+ "seconds": 1707.7
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.5774,
189
+ "val_loss": 2.71944,
190
+ "step": 13149,
191
+ "seconds": 1772.3
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56959,
196
+ "val_loss": 2.72557,
197
+ "step": 13636,
198
+ "seconds": 1836.8
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.56215,
203
+ "val_loss": 2.73034,
204
+ "step": 14123,
205
+ "seconds": 1901.4
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.55434,
210
+ "val_loss": 2.73623,
211
+ "step": 14610,
212
+ "seconds": 1966.0
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.54646,
217
+ "val_loss": 2.74292,
218
+ "step": 15097,
219
+ "seconds": 2030.8
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.53877,
224
+ "val_loss": 2.74902,
225
+ "step": 15584,
226
+ "seconds": 2095.3
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.53099,
231
+ "val_loss": 2.75305,
232
+ "step": 16071,
233
+ "seconds": 2159.8
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.52345,
238
+ "val_loss": 2.76281,
239
+ "step": 16558,
240
+ "seconds": 2223.8
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.5158,
245
+ "val_loss": 2.76971,
246
+ "step": 17045,
247
+ "seconds": 2287.5
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.50793,
252
+ "val_loss": 2.77616,
253
+ "step": 17532,
254
+ "seconds": 2351.0
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.50026,
259
+ "val_loss": 2.78411,
260
+ "step": 18019,
261
+ "seconds": 2414.7
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.49269,
266
+ "val_loss": 2.79204,
267
+ "step": 18506,
268
+ "seconds": 2478.3
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.48539,
273
+ "val_loss": 2.79654,
274
+ "step": 18993,
275
+ "seconds": 2542.0
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.47762,
280
+ "val_loss": 2.80283,
281
+ "step": 19480,
282
+ "seconds": 2605.7
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.47021,
287
+ "val_loss": 2.81165,
288
+ "step": 19967,
289
+ "seconds": 2669.7
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.46378,
294
+ "val_loss": 2.81977,
295
+ "step": 20454,
296
+ "seconds": 2733.5
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.45599,
301
+ "val_loss": 2.82406,
302
+ "step": 20941,
303
+ "seconds": 2797.3
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.44969,
308
+ "val_loss": 2.82987,
309
+ "step": 21428,
310
+ "seconds": 2861.0
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.44198,
315
+ "val_loss": 2.83467,
316
+ "step": 21915,
317
+ "seconds": 2924.8
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.43554,
322
+ "val_loss": 2.84634,
323
+ "step": 22402,
324
+ "seconds": 2988.5
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.42843,
329
+ "val_loss": 2.84874,
330
+ "step": 22889,
331
+ "seconds": 3052.2
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.42312,
336
+ "val_loss": 2.85941,
337
+ "step": 23376,
338
+ "seconds": 3116.0
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.41631,
343
+ "val_loss": 2.86514,
344
+ "step": 23863,
345
+ "seconds": 3179.7
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.40988,
350
+ "val_loss": 2.87203,
351
+ "step": 24350,
352
+ "seconds": 3243.3
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.40369,
357
+ "val_loss": 2.87951,
358
+ "step": 24837,
359
+ "seconds": 3307.0
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.39825,
364
+ "val_loss": 2.88127,
365
+ "step": 25324,
366
+ "seconds": 3370.7
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.39241,
371
+ "val_loss": 2.89043,
372
+ "step": 25811,
373
+ "seconds": 3434.4
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.38628,
378
+ "val_loss": 2.89932,
379
+ "step": 26298,
380
+ "seconds": 3498.1
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.38052,
385
+ "val_loss": 2.90355,
386
+ "step": 26785,
387
+ "seconds": 3561.8
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.37523,
392
+ "val_loss": 2.90804,
393
+ "step": 27272,
394
+ "seconds": 3625.5
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.36941,
399
+ "val_loss": 2.9161,
400
+ "step": 27759,
401
+ "seconds": 3689.3
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.36509,
406
+ "val_loss": 2.91723,
407
+ "step": 28246,
408
+ "seconds": 3753.0
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.3604,
413
+ "val_loss": 2.92538,
414
+ "step": 28733,
415
+ "seconds": 3816.7
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.35523,
420
+ "val_loss": 2.92731,
421
+ "step": 29220,
422
+ "seconds": 3880.4
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.35067,
427
+ "val_loss": 2.93458,
428
+ "step": 29707,
429
+ "seconds": 3944.2
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.346,
434
+ "val_loss": 2.94193,
435
+ "step": 30194,
436
+ "seconds": 4007.9
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.34152,
441
+ "val_loss": 2.94682,
442
+ "step": 30681,
443
+ "seconds": 4071.6
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.33736,
448
+ "val_loss": 2.94938,
449
+ "step": 31168,
450
+ "seconds": 4135.4
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.33286,
455
+ "val_loss": 2.95316,
456
+ "step": 31655,
457
+ "seconds": 4199.1
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.32898,
462
+ "val_loss": 2.95863,
463
+ "step": 32142,
464
+ "seconds": 4262.8
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.3252,
469
+ "val_loss": 2.96428,
470
+ "step": 32629,
471
+ "seconds": 4326.5
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.32112,
476
+ "val_loss": 2.96299,
477
+ "step": 33116,
478
+ "seconds": 4390.3
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.31769,
483
+ "val_loss": 2.9717,
484
+ "step": 33603,
485
+ "seconds": 4454.1
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.31396,
490
+ "val_loss": 2.97471,
491
+ "step": 34090,
492
+ "seconds": 4517.9
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.31188,
497
+ "val_loss": 2.97783,
498
+ "step": 34577,
499
+ "seconds": 4581.6
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.30779,
504
+ "val_loss": 2.97958,
505
+ "step": 35064,
506
+ "seconds": 4645.4
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.30503,
511
+ "val_loss": 2.98379,
512
+ "step": 35551,
513
+ "seconds": 4709.0
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.30131,
518
+ "val_loss": 2.98674,
519
+ "step": 36038,
520
+ "seconds": 4772.7
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.29866,
525
+ "val_loss": 2.99159,
526
+ "step": 36525,
527
+ "seconds": 4836.5
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.29607,
532
+ "val_loss": 2.99254,
533
+ "step": 37012,
534
+ "seconds": 4900.2
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.29404,
539
+ "val_loss": 2.99504,
540
+ "step": 37499,
541
+ "seconds": 4963.9
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.29139,
546
+ "val_loss": 2.99781,
547
+ "step": 37986,
548
+ "seconds": 5027.7
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.28918,
553
+ "val_loss": 3.00084,
554
+ "step": 38473,
555
+ "seconds": 5091.4
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.2873,
560
+ "val_loss": 3.00098,
561
+ "step": 38960,
562
+ "seconds": 5155.1
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.28508,
567
+ "val_loss": 3.00509,
568
+ "step": 39447,
569
+ "seconds": 5218.9
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.28311,
574
+ "val_loss": 3.00889,
575
+ "step": 39934,
576
+ "seconds": 5282.6
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.28196,
581
+ "val_loss": 3.00601,
582
+ "step": 40421,
583
+ "seconds": 5346.2
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.27998,
588
+ "val_loss": 3.0101,
589
+ "step": 40908,
590
+ "seconds": 5409.8
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.27848,
595
+ "val_loss": 3.00844,
596
+ "step": 41395,
597
+ "seconds": 5473.4
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.27804,
602
+ "val_loss": 3.01163,
603
+ "step": 41882,
604
+ "seconds": 5537.0
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.27585,
609
+ "val_loss": 3.01407,
610
+ "step": 42369,
611
+ "seconds": 5600.7
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.275,
616
+ "val_loss": 3.01562,
617
+ "step": 42856,
618
+ "seconds": 5664.5
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.27448,
623
+ "val_loss": 3.01591,
624
+ "step": 43343,
625
+ "seconds": 5728.1
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.27293,
630
+ "val_loss": 3.01638,
631
+ "step": 43830,
632
+ "seconds": 5791.8
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.2727,
637
+ "val_loss": 3.01557,
638
+ "step": 44317,
639
+ "seconds": 5855.6
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.27156,
644
+ "val_loss": 3.0175,
645
+ "step": 44804,
646
+ "seconds": 5919.2
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.27101,
651
+ "val_loss": 3.01702,
652
+ "step": 45291,
653
+ "seconds": 5982.9
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.27087,
658
+ "val_loss": 3.01783,
659
+ "step": 45778,
660
+ "seconds": 6046.6
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.27041,
665
+ "val_loss": 3.01828,
666
+ "step": 46265,
667
+ "seconds": 6110.4
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.26984,
672
+ "val_loss": 3.01854,
673
+ "step": 46752,
674
+ "seconds": 6174.1
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.26983,
679
+ "val_loss": 3.01849,
680
+ "step": 47239,
681
+ "seconds": 6237.9
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.26933,
686
+ "val_loss": 3.01851,
687
+ "step": 47726,
688
+ "seconds": 6301.5
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.26934,
693
+ "val_loss": 3.01872,
694
+ "step": 48213,
695
+ "seconds": 6365.2
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.26914,
700
+ "val_loss": 3.01848,
701
+ "step": 48700,
702
+ "seconds": 6428.9
703
+ }
704
+ ]
705
+ }
H_pq1/arm_meta.json ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "H_pq1",
3
+ "seed": 1,
4
+ "seed_witness": "0fbe2a188322a36c",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "6508c0fb1ee7e821b7d09b9e8b249e2e49413f5b",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0,
69
+ "partial_queries": true,
70
+ "depth_distribution": {
71
+ "6": 0.5,
72
+ "5": 0.125,
73
+ "4": 0.125,
74
+ "3": 0.125,
75
+ "2": 0.125
76
+ }
77
+ },
78
+ "code_hash": "3084c3b83d82c8cc05d1f2d72ac14f5cfdf134571152e11c3f747c5d9cdc0389",
79
+ "device": "cuda",
80
+ "started_utc": "2026-08-08T15:14:21Z",
81
+ "train_seconds": 6406.8,
82
+ "final_val_loss": 3.02672,
83
+ "final_train_loss": 2.25125,
84
+ "complete": true
85
+ }
H_pq1/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccffa3de74cbe87aed963bcfcec41d3547cab5d46d811e7811850ec250eaf840
3
+ size 18080482
H_pq1/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:716c2ea5954be2c5c401fbfe6273bb7393a5322b3b56fca3be7f340cbd13faf5
3
+ size 18080482
H_pq1/depth2.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq1/depth3.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq1/depth4.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq1/depth6.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq1/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.56693,
7
+ "val_loss": 5.72442,
8
+ "step": 487,
9
+ "seconds": 98.5
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.79297,
14
+ "val_loss": 4.32671,
15
+ "step": 974,
16
+ "seconds": 162.1
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 3.88868,
21
+ "val_loss": 3.4516,
22
+ "step": 1461,
23
+ "seconds": 225.8
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.19543,
28
+ "val_loss": 3.00533,
29
+ "step": 1948,
30
+ "seconds": 289.4
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 2.95541,
35
+ "val_loss": 2.87151,
36
+ "step": 2435,
37
+ "seconds": 353.2
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.86856,
42
+ "val_loss": 2.8096,
43
+ "step": 2922,
44
+ "seconds": 416.9
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.81821,
49
+ "val_loss": 2.77363,
50
+ "step": 3409,
51
+ "seconds": 480.5
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.78358,
56
+ "val_loss": 2.74695,
57
+ "step": 3896,
58
+ "seconds": 544.2
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.75805,
63
+ "val_loss": 2.72811,
64
+ "step": 4383,
65
+ "seconds": 607.9
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.7374,
70
+ "val_loss": 2.7183,
71
+ "step": 4870,
72
+ "seconds": 671.6
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.7204,
77
+ "val_loss": 2.70747,
78
+ "step": 5357,
79
+ "seconds": 735.4
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.70551,
84
+ "val_loss": 2.70182,
85
+ "step": 5844,
86
+ "seconds": 799.0
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.6925,
91
+ "val_loss": 2.69862,
92
+ "step": 6331,
93
+ "seconds": 862.7
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68115,
98
+ "val_loss": 2.69639,
99
+ "step": 6818,
100
+ "seconds": 926.3
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.67069,
105
+ "val_loss": 2.69242,
106
+ "step": 7305,
107
+ "seconds": 990.0
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66058,
112
+ "val_loss": 2.68986,
113
+ "step": 7792,
114
+ "seconds": 1053.6
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.6514,
119
+ "val_loss": 2.69091,
120
+ "step": 8279,
121
+ "seconds": 1117.3
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.6423,
126
+ "val_loss": 2.68935,
127
+ "step": 8766,
128
+ "seconds": 1181.1
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.63416,
133
+ "val_loss": 2.69377,
134
+ "step": 9253,
135
+ "seconds": 1244.6
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.62608,
140
+ "val_loss": 2.69579,
141
+ "step": 9740,
142
+ "seconds": 1308.2
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.61736,
147
+ "val_loss": 2.6959,
148
+ "step": 10227,
149
+ "seconds": 1372.1
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.60951,
154
+ "val_loss": 2.70183,
155
+ "step": 10714,
156
+ "seconds": 1435.8
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.60148,
161
+ "val_loss": 2.70138,
162
+ "step": 11201,
163
+ "seconds": 1499.4
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.59331,
168
+ "val_loss": 2.70613,
169
+ "step": 11688,
170
+ "seconds": 1563.1
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.58538,
175
+ "val_loss": 2.71306,
176
+ "step": 12175,
177
+ "seconds": 1626.7
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.5769,
182
+ "val_loss": 2.71687,
183
+ "step": 12662,
184
+ "seconds": 1690.4
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.56909,
189
+ "val_loss": 2.71886,
190
+ "step": 13149,
191
+ "seconds": 1754.0
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56107,
196
+ "val_loss": 2.72816,
197
+ "step": 13636,
198
+ "seconds": 1817.7
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.55307,
203
+ "val_loss": 2.73114,
204
+ "step": 14123,
205
+ "seconds": 1881.4
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.5448,
210
+ "val_loss": 2.73689,
211
+ "step": 14610,
212
+ "seconds": 1944.9
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.537,
217
+ "val_loss": 2.73938,
218
+ "step": 15097,
219
+ "seconds": 2008.7
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.52915,
224
+ "val_loss": 2.74761,
225
+ "step": 15584,
226
+ "seconds": 2072.4
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.5211,
231
+ "val_loss": 2.75723,
232
+ "step": 16071,
233
+ "seconds": 2136.1
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.5131,
238
+ "val_loss": 2.76305,
239
+ "step": 16558,
240
+ "seconds": 2199.7
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.50501,
245
+ "val_loss": 2.77233,
246
+ "step": 17045,
247
+ "seconds": 2263.4
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.4973,
252
+ "val_loss": 2.77594,
253
+ "step": 17532,
254
+ "seconds": 2327.2
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.48877,
259
+ "val_loss": 2.78324,
260
+ "step": 18019,
261
+ "seconds": 2390.9
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.48125,
266
+ "val_loss": 2.79318,
267
+ "step": 18506,
268
+ "seconds": 2454.6
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.47396,
273
+ "val_loss": 2.799,
274
+ "step": 18993,
275
+ "seconds": 2518.3
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.46609,
280
+ "val_loss": 2.80819,
281
+ "step": 19480,
282
+ "seconds": 2582.1
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.45858,
287
+ "val_loss": 2.81218,
288
+ "step": 19967,
289
+ "seconds": 2645.9
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.45128,
294
+ "val_loss": 2.82189,
295
+ "step": 20454,
296
+ "seconds": 2709.6
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.44384,
301
+ "val_loss": 2.82832,
302
+ "step": 20941,
303
+ "seconds": 2773.4
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.43719,
308
+ "val_loss": 2.83244,
309
+ "step": 21428,
310
+ "seconds": 2837.1
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.42921,
315
+ "val_loss": 2.84577,
316
+ "step": 21915,
317
+ "seconds": 2900.7
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.42278,
322
+ "val_loss": 2.84722,
323
+ "step": 22402,
324
+ "seconds": 2964.5
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.41597,
329
+ "val_loss": 2.85545,
330
+ "step": 22889,
331
+ "seconds": 3028.2
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.4087,
336
+ "val_loss": 2.86255,
337
+ "step": 23376,
338
+ "seconds": 3092.0
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.40197,
343
+ "val_loss": 2.86969,
344
+ "step": 23863,
345
+ "seconds": 3155.7
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.39598,
350
+ "val_loss": 2.87762,
351
+ "step": 24350,
352
+ "seconds": 3219.5
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.39003,
357
+ "val_loss": 2.88841,
358
+ "step": 24837,
359
+ "seconds": 3283.3
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.38442,
364
+ "val_loss": 2.89217,
365
+ "step": 25324,
366
+ "seconds": 3347.0
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.37792,
371
+ "val_loss": 2.89555,
372
+ "step": 25811,
373
+ "seconds": 3410.7
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.372,
378
+ "val_loss": 2.90194,
379
+ "step": 26298,
380
+ "seconds": 3474.4
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.36616,
385
+ "val_loss": 2.9109,
386
+ "step": 26785,
387
+ "seconds": 3538.1
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.36109,
392
+ "val_loss": 2.91323,
393
+ "step": 27272,
394
+ "seconds": 3601.8
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.35516,
399
+ "val_loss": 2.92001,
400
+ "step": 27759,
401
+ "seconds": 3665.5
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.35035,
406
+ "val_loss": 2.92511,
407
+ "step": 28246,
408
+ "seconds": 3729.2
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.34524,
413
+ "val_loss": 2.9299,
414
+ "step": 28733,
415
+ "seconds": 3792.9
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.33987,
420
+ "val_loss": 2.93825,
421
+ "step": 29220,
422
+ "seconds": 3856.6
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.3352,
427
+ "val_loss": 2.94034,
428
+ "step": 29707,
429
+ "seconds": 3920.4
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.32985,
434
+ "val_loss": 2.94591,
435
+ "step": 30194,
436
+ "seconds": 3984.1
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.3258,
441
+ "val_loss": 2.95492,
442
+ "step": 30681,
443
+ "seconds": 4047.8
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.32088,
448
+ "val_loss": 2.9582,
449
+ "step": 31168,
450
+ "seconds": 4111.6
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.3169,
455
+ "val_loss": 2.96099,
456
+ "step": 31655,
457
+ "seconds": 4175.4
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.31327,
462
+ "val_loss": 2.96737,
463
+ "step": 32142,
464
+ "seconds": 4239.1
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.30917,
469
+ "val_loss": 2.96887,
470
+ "step": 32629,
471
+ "seconds": 4302.7
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.30535,
476
+ "val_loss": 2.97378,
477
+ "step": 33116,
478
+ "seconds": 4366.4
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.30185,
483
+ "val_loss": 2.97587,
484
+ "step": 33603,
485
+ "seconds": 4430.1
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.29817,
490
+ "val_loss": 2.98076,
491
+ "step": 34090,
492
+ "seconds": 4493.7
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.29423,
497
+ "val_loss": 2.98429,
498
+ "step": 34577,
499
+ "seconds": 4557.5
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.29081,
504
+ "val_loss": 2.98571,
505
+ "step": 35064,
506
+ "seconds": 4621.2
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.28831,
511
+ "val_loss": 2.99069,
512
+ "step": 35551,
513
+ "seconds": 4684.9
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.28563,
518
+ "val_loss": 2.99429,
519
+ "step": 36038,
520
+ "seconds": 4748.7
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.28175,
525
+ "val_loss": 2.99913,
526
+ "step": 36525,
527
+ "seconds": 4812.5
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.27914,
532
+ "val_loss": 2.99897,
533
+ "step": 37012,
534
+ "seconds": 4876.3
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.27666,
539
+ "val_loss": 3.0025,
540
+ "step": 37499,
541
+ "seconds": 4940.0
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.27445,
546
+ "val_loss": 3.00646,
547
+ "step": 37986,
548
+ "seconds": 5003.7
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.27224,
553
+ "val_loss": 3.00771,
554
+ "step": 38473,
555
+ "seconds": 5067.4
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.27021,
560
+ "val_loss": 3.01156,
561
+ "step": 38960,
562
+ "seconds": 5131.2
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.26809,
567
+ "val_loss": 3.00949,
568
+ "step": 39447,
569
+ "seconds": 5195.0
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.26626,
574
+ "val_loss": 3.01417,
575
+ "step": 39934,
576
+ "seconds": 5258.7
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.26425,
581
+ "val_loss": 3.01656,
582
+ "step": 40421,
583
+ "seconds": 5322.5
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.26274,
588
+ "val_loss": 3.01722,
589
+ "step": 40908,
590
+ "seconds": 5386.2
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.26139,
595
+ "val_loss": 3.01741,
596
+ "step": 41395,
597
+ "seconds": 5449.9
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.25989,
602
+ "val_loss": 3.02019,
603
+ "step": 41882,
604
+ "seconds": 5513.6
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.25818,
609
+ "val_loss": 3.02134,
610
+ "step": 42369,
611
+ "seconds": 5577.3
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.25802,
616
+ "val_loss": 3.02193,
617
+ "step": 42856,
618
+ "seconds": 5641.1
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.25623,
623
+ "val_loss": 3.02399,
624
+ "step": 43343,
625
+ "seconds": 5704.9
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.25525,
630
+ "val_loss": 3.02416,
631
+ "step": 43830,
632
+ "seconds": 5768.8
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.25395,
637
+ "val_loss": 3.02356,
638
+ "step": 44317,
639
+ "seconds": 5832.6
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.25424,
644
+ "val_loss": 3.02429,
645
+ "step": 44804,
646
+ "seconds": 5896.4
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.25316,
651
+ "val_loss": 3.02467,
652
+ "step": 45291,
653
+ "seconds": 5960.1
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.25272,
658
+ "val_loss": 3.02479,
659
+ "step": 45778,
660
+ "seconds": 6023.9
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.25191,
665
+ "val_loss": 3.02575,
666
+ "step": 46265,
667
+ "seconds": 6087.7
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.25179,
672
+ "val_loss": 3.02606,
673
+ "step": 46752,
674
+ "seconds": 6151.5
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.2515,
679
+ "val_loss": 3.02624,
680
+ "step": 47239,
681
+ "seconds": 6215.2
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.25129,
686
+ "val_loss": 3.02667,
687
+ "step": 47726,
688
+ "seconds": 6279.1
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.25109,
693
+ "val_loss": 3.02698,
694
+ "step": 48213,
695
+ "seconds": 6342.9
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.25125,
700
+ "val_loss": 3.02672,
701
+ "step": 48700,
702
+ "seconds": 6406.7
703
+ }
704
+ ]
705
+ }
H_pq2/arm_meta.json ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "H_pq2",
3
+ "seed": 2,
4
+ "seed_witness": "f2a8e585693907e0",
5
+ "seeded_rngs": [
6
+ "random",
7
+ "numpy",
8
+ "torch"
9
+ ],
10
+ "corpus": {
11
+ "path": "data/v2/s2/train_stage1.json",
12
+ "content_hash": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075",
13
+ "n_records": 69190,
14
+ "n_encodable": 69190,
15
+ "generator": "v2.run_phase_d.split",
16
+ "params": {
17
+ "key": "skeleton",
18
+ "val_fraction": 0.2,
19
+ "seed": 0,
20
+ "source_corpus": "data/v2/corpus_stage2.json",
21
+ "corpus_hash": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c"
22
+ }
23
+ },
24
+ "encode_drops": {
25
+ "stage": "dataset:encode",
26
+ "n_input": 69190,
27
+ "n_kept": 69190,
28
+ "dropped": {}
29
+ },
30
+ "split": {
31
+ "key": "skeleton",
32
+ "seed": 0,
33
+ "val_fraction": 0.1,
34
+ "counts": {
35
+ "train": 62313,
36
+ "val": 6877
37
+ },
38
+ "hashes": {
39
+ "train": "23f199f139ae69626162a7e60af566a297f1a269994bcfdcb0627e6e3cb7f7ab",
40
+ "val": "712197b0032eef37514ba71a9af726ef96ed00a25b4360a670087e8236c60f65"
41
+ },
42
+ "key_counts": {
43
+ "train": 21957,
44
+ "val": 2440
45
+ }
46
+ },
47
+ "model": {
48
+ "vocab_size": 2317,
49
+ "d_model": 256,
50
+ "n_layers": 4,
51
+ "n_heads": 8,
52
+ "d_ff": 1024,
53
+ "dropout": 0.1,
54
+ "max_seq": 102,
55
+ "version": "v2-model-1",
56
+ "pipeline_version": "v2.0.0"
57
+ },
58
+ "vocab_hash": "3e2eb3f5475497a1ea8754b925e1164d46cbaf5887fa6dd3dcaf78ae9b21e5f3",
59
+ "tokenizer_version": "v2-tok-3",
60
+ "dataset_version": "v2-data-1",
61
+ "pipeline_version": "v2.0.0",
62
+ "git_commit": "6508c0fb1ee7e821b7d09b9e8b249e2e49413f5b",
63
+ "hyper": {
64
+ "epochs": 100,
65
+ "batch": 128,
66
+ "lr": 0.0003,
67
+ "val_fraction": 0.1,
68
+ "split_seed": 0,
69
+ "partial_queries": true,
70
+ "depth_distribution": {
71
+ "6": 0.5,
72
+ "5": 0.125,
73
+ "4": 0.125,
74
+ "3": 0.125,
75
+ "2": 0.125
76
+ }
77
+ },
78
+ "code_hash": "3084c3b83d82c8cc05d1f2d72ac14f5cfdf134571152e11c3f747c5d9cdc0389",
79
+ "device": "cuda",
80
+ "started_utc": "2026-08-08T17:01:12Z",
81
+ "train_seconds": 6408.6,
82
+ "final_val_loss": 3.01042,
83
+ "final_train_loss": 2.26395,
84
+ "complete": true
85
+ }
H_pq2/ckpt_00020.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:93da32c85495eaa97e7812764b4dbd24f869a86acf5d3cd8b2c541d8aaff7351
3
+ size 18080482
H_pq2/ckpt_00100.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e0451240f5cbba903b87ce87c658a069928687d24b5e121092afc4d49fe99334
3
+ size 18080482
H_pq2/depth2.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq2/depth3.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq2/depth4.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq2/depth6.json ADDED
The diff for this file is too large to render. See raw diff
 
H_pq2/history.json ADDED
@@ -0,0 +1,705 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "complete": true,
3
+ "history": [
4
+ {
5
+ "epoch": 1,
6
+ "train_loss": 6.6076,
7
+ "val_loss": 5.73111,
8
+ "step": 487,
9
+ "seconds": 98.6
10
+ },
11
+ {
12
+ "epoch": 2,
13
+ "train_loss": 4.7989,
14
+ "val_loss": 4.3325,
15
+ "step": 974,
16
+ "seconds": 162.2
17
+ },
18
+ {
19
+ "epoch": 3,
20
+ "train_loss": 4.02878,
21
+ "val_loss": 3.65539,
22
+ "step": 1461,
23
+ "seconds": 225.9
24
+ },
25
+ {
26
+ "epoch": 4,
27
+ "train_loss": 3.30508,
28
+ "val_loss": 3.0754,
29
+ "step": 1948,
30
+ "seconds": 289.7
31
+ },
32
+ {
33
+ "epoch": 5,
34
+ "train_loss": 3.00498,
35
+ "val_loss": 2.91266,
36
+ "step": 2435,
37
+ "seconds": 353.4
38
+ },
39
+ {
40
+ "epoch": 6,
41
+ "train_loss": 2.89792,
42
+ "val_loss": 2.83473,
43
+ "step": 2922,
44
+ "seconds": 417.1
45
+ },
46
+ {
47
+ "epoch": 7,
48
+ "train_loss": 2.83947,
49
+ "val_loss": 2.78726,
50
+ "step": 3409,
51
+ "seconds": 480.9
52
+ },
53
+ {
54
+ "epoch": 8,
55
+ "train_loss": 2.79867,
56
+ "val_loss": 2.75667,
57
+ "step": 3896,
58
+ "seconds": 544.7
59
+ },
60
+ {
61
+ "epoch": 9,
62
+ "train_loss": 2.76825,
63
+ "val_loss": 2.73791,
64
+ "step": 4383,
65
+ "seconds": 608.3
66
+ },
67
+ {
68
+ "epoch": 10,
69
+ "train_loss": 2.74444,
70
+ "val_loss": 2.72195,
71
+ "step": 4870,
72
+ "seconds": 672.0
73
+ },
74
+ {
75
+ "epoch": 11,
76
+ "train_loss": 2.72564,
77
+ "val_loss": 2.71034,
78
+ "step": 5357,
79
+ "seconds": 735.7
80
+ },
81
+ {
82
+ "epoch": 12,
83
+ "train_loss": 2.70953,
84
+ "val_loss": 2.70341,
85
+ "step": 5844,
86
+ "seconds": 799.3
87
+ },
88
+ {
89
+ "epoch": 13,
90
+ "train_loss": 2.69669,
91
+ "val_loss": 2.70436,
92
+ "step": 6331,
93
+ "seconds": 863.1
94
+ },
95
+ {
96
+ "epoch": 14,
97
+ "train_loss": 2.68492,
98
+ "val_loss": 2.69384,
99
+ "step": 6818,
100
+ "seconds": 926.8
101
+ },
102
+ {
103
+ "epoch": 15,
104
+ "train_loss": 2.67451,
105
+ "val_loss": 2.6931,
106
+ "step": 7305,
107
+ "seconds": 990.6
108
+ },
109
+ {
110
+ "epoch": 16,
111
+ "train_loss": 2.66412,
112
+ "val_loss": 2.69961,
113
+ "step": 7792,
114
+ "seconds": 1054.3
115
+ },
116
+ {
117
+ "epoch": 17,
118
+ "train_loss": 2.65515,
119
+ "val_loss": 2.69085,
120
+ "step": 8279,
121
+ "seconds": 1118.1
122
+ },
123
+ {
124
+ "epoch": 18,
125
+ "train_loss": 2.64619,
126
+ "val_loss": 2.69291,
127
+ "step": 8766,
128
+ "seconds": 1181.8
129
+ },
130
+ {
131
+ "epoch": 19,
132
+ "train_loss": 2.63723,
133
+ "val_loss": 2.69213,
134
+ "step": 9253,
135
+ "seconds": 1245.5
136
+ },
137
+ {
138
+ "epoch": 20,
139
+ "train_loss": 2.629,
140
+ "val_loss": 2.69356,
141
+ "step": 9740,
142
+ "seconds": 1309.2
143
+ },
144
+ {
145
+ "epoch": 21,
146
+ "train_loss": 2.6212,
147
+ "val_loss": 2.69612,
148
+ "step": 10227,
149
+ "seconds": 1373.0
150
+ },
151
+ {
152
+ "epoch": 22,
153
+ "train_loss": 2.61298,
154
+ "val_loss": 2.69803,
155
+ "step": 10714,
156
+ "seconds": 1436.7
157
+ },
158
+ {
159
+ "epoch": 23,
160
+ "train_loss": 2.60463,
161
+ "val_loss": 2.70327,
162
+ "step": 11201,
163
+ "seconds": 1500.4
164
+ },
165
+ {
166
+ "epoch": 24,
167
+ "train_loss": 2.59673,
168
+ "val_loss": 2.70561,
169
+ "step": 11688,
170
+ "seconds": 1564.0
171
+ },
172
+ {
173
+ "epoch": 25,
174
+ "train_loss": 2.5888,
175
+ "val_loss": 2.70836,
176
+ "step": 12175,
177
+ "seconds": 1627.7
178
+ },
179
+ {
180
+ "epoch": 26,
181
+ "train_loss": 2.58119,
182
+ "val_loss": 2.71076,
183
+ "step": 12662,
184
+ "seconds": 1691.5
185
+ },
186
+ {
187
+ "epoch": 27,
188
+ "train_loss": 2.57347,
189
+ "val_loss": 2.71601,
190
+ "step": 13149,
191
+ "seconds": 1755.3
192
+ },
193
+ {
194
+ "epoch": 28,
195
+ "train_loss": 2.56545,
196
+ "val_loss": 2.72674,
197
+ "step": 13636,
198
+ "seconds": 1819.0
199
+ },
200
+ {
201
+ "epoch": 29,
202
+ "train_loss": 2.55799,
203
+ "val_loss": 2.72776,
204
+ "step": 14123,
205
+ "seconds": 1882.8
206
+ },
207
+ {
208
+ "epoch": 30,
209
+ "train_loss": 2.55043,
210
+ "val_loss": 2.73109,
211
+ "step": 14610,
212
+ "seconds": 1946.6
213
+ },
214
+ {
215
+ "epoch": 31,
216
+ "train_loss": 2.54253,
217
+ "val_loss": 2.73823,
218
+ "step": 15097,
219
+ "seconds": 2010.5
220
+ },
221
+ {
222
+ "epoch": 32,
223
+ "train_loss": 2.53468,
224
+ "val_loss": 2.74864,
225
+ "step": 15584,
226
+ "seconds": 2074.1
227
+ },
228
+ {
229
+ "epoch": 33,
230
+ "train_loss": 2.52687,
231
+ "val_loss": 2.75477,
232
+ "step": 16071,
233
+ "seconds": 2137.8
234
+ },
235
+ {
236
+ "epoch": 34,
237
+ "train_loss": 2.51915,
238
+ "val_loss": 2.75896,
239
+ "step": 16558,
240
+ "seconds": 2201.6
241
+ },
242
+ {
243
+ "epoch": 35,
244
+ "train_loss": 2.51198,
245
+ "val_loss": 2.76404,
246
+ "step": 17045,
247
+ "seconds": 2265.2
248
+ },
249
+ {
250
+ "epoch": 36,
251
+ "train_loss": 2.50351,
252
+ "val_loss": 2.77061,
253
+ "step": 17532,
254
+ "seconds": 2328.9
255
+ },
256
+ {
257
+ "epoch": 37,
258
+ "train_loss": 2.49651,
259
+ "val_loss": 2.77771,
260
+ "step": 18019,
261
+ "seconds": 2392.5
262
+ },
263
+ {
264
+ "epoch": 38,
265
+ "train_loss": 2.48901,
266
+ "val_loss": 2.78245,
267
+ "step": 18506,
268
+ "seconds": 2456.2
269
+ },
270
+ {
271
+ "epoch": 39,
272
+ "train_loss": 2.48137,
273
+ "val_loss": 2.78998,
274
+ "step": 18993,
275
+ "seconds": 2519.8
276
+ },
277
+ {
278
+ "epoch": 40,
279
+ "train_loss": 2.47376,
280
+ "val_loss": 2.79831,
281
+ "step": 19480,
282
+ "seconds": 2583.4
283
+ },
284
+ {
285
+ "epoch": 41,
286
+ "train_loss": 2.46681,
287
+ "val_loss": 2.80499,
288
+ "step": 19967,
289
+ "seconds": 2647.2
290
+ },
291
+ {
292
+ "epoch": 42,
293
+ "train_loss": 2.45933,
294
+ "val_loss": 2.80922,
295
+ "step": 20454,
296
+ "seconds": 2710.9
297
+ },
298
+ {
299
+ "epoch": 43,
300
+ "train_loss": 2.45338,
301
+ "val_loss": 2.81945,
302
+ "step": 20941,
303
+ "seconds": 2774.7
304
+ },
305
+ {
306
+ "epoch": 44,
307
+ "train_loss": 2.44541,
308
+ "val_loss": 2.82154,
309
+ "step": 21428,
310
+ "seconds": 2838.4
311
+ },
312
+ {
313
+ "epoch": 45,
314
+ "train_loss": 2.43933,
315
+ "val_loss": 2.82761,
316
+ "step": 21915,
317
+ "seconds": 2902.2
318
+ },
319
+ {
320
+ "epoch": 46,
321
+ "train_loss": 2.4328,
322
+ "val_loss": 2.84076,
323
+ "step": 22402,
324
+ "seconds": 2965.9
325
+ },
326
+ {
327
+ "epoch": 47,
328
+ "train_loss": 2.42561,
329
+ "val_loss": 2.84627,
330
+ "step": 22889,
331
+ "seconds": 3029.6
332
+ },
333
+ {
334
+ "epoch": 48,
335
+ "train_loss": 2.41907,
336
+ "val_loss": 2.84735,
337
+ "step": 23376,
338
+ "seconds": 3093.4
339
+ },
340
+ {
341
+ "epoch": 49,
342
+ "train_loss": 2.41325,
343
+ "val_loss": 2.85878,
344
+ "step": 23863,
345
+ "seconds": 3157.0
346
+ },
347
+ {
348
+ "epoch": 50,
349
+ "train_loss": 2.40672,
350
+ "val_loss": 2.86911,
351
+ "step": 24350,
352
+ "seconds": 3220.7
353
+ },
354
+ {
355
+ "epoch": 51,
356
+ "train_loss": 2.40059,
357
+ "val_loss": 2.8723,
358
+ "step": 24837,
359
+ "seconds": 3284.4
360
+ },
361
+ {
362
+ "epoch": 52,
363
+ "train_loss": 2.39473,
364
+ "val_loss": 2.87304,
365
+ "step": 25324,
366
+ "seconds": 3348.1
367
+ },
368
+ {
369
+ "epoch": 53,
370
+ "train_loss": 2.38882,
371
+ "val_loss": 2.8834,
372
+ "step": 25811,
373
+ "seconds": 3411.8
374
+ },
375
+ {
376
+ "epoch": 54,
377
+ "train_loss": 2.38326,
378
+ "val_loss": 2.89122,
379
+ "step": 26298,
380
+ "seconds": 3475.5
381
+ },
382
+ {
383
+ "epoch": 55,
384
+ "train_loss": 2.37763,
385
+ "val_loss": 2.89302,
386
+ "step": 26785,
387
+ "seconds": 3539.4
388
+ },
389
+ {
390
+ "epoch": 56,
391
+ "train_loss": 2.37188,
392
+ "val_loss": 2.90257,
393
+ "step": 27272,
394
+ "seconds": 3603.1
395
+ },
396
+ {
397
+ "epoch": 57,
398
+ "train_loss": 2.36728,
399
+ "val_loss": 2.90613,
400
+ "step": 27759,
401
+ "seconds": 3666.7
402
+ },
403
+ {
404
+ "epoch": 58,
405
+ "train_loss": 2.36157,
406
+ "val_loss": 2.90569,
407
+ "step": 28246,
408
+ "seconds": 3730.5
409
+ },
410
+ {
411
+ "epoch": 59,
412
+ "train_loss": 2.35675,
413
+ "val_loss": 2.91865,
414
+ "step": 28733,
415
+ "seconds": 3794.2
416
+ },
417
+ {
418
+ "epoch": 60,
419
+ "train_loss": 2.35148,
420
+ "val_loss": 2.91852,
421
+ "step": 29220,
422
+ "seconds": 3857.9
423
+ },
424
+ {
425
+ "epoch": 61,
426
+ "train_loss": 2.34708,
427
+ "val_loss": 2.92636,
428
+ "step": 29707,
429
+ "seconds": 3921.8
430
+ },
431
+ {
432
+ "epoch": 62,
433
+ "train_loss": 2.34253,
434
+ "val_loss": 2.93406,
435
+ "step": 30194,
436
+ "seconds": 3985.7
437
+ },
438
+ {
439
+ "epoch": 63,
440
+ "train_loss": 2.33801,
441
+ "val_loss": 2.93796,
442
+ "step": 30681,
443
+ "seconds": 4049.4
444
+ },
445
+ {
446
+ "epoch": 64,
447
+ "train_loss": 2.3334,
448
+ "val_loss": 2.94243,
449
+ "step": 31168,
450
+ "seconds": 4113.1
451
+ },
452
+ {
453
+ "epoch": 65,
454
+ "train_loss": 2.32966,
455
+ "val_loss": 2.94094,
456
+ "step": 31655,
457
+ "seconds": 4176.8
458
+ },
459
+ {
460
+ "epoch": 66,
461
+ "train_loss": 2.32496,
462
+ "val_loss": 2.94876,
463
+ "step": 32142,
464
+ "seconds": 4240.5
465
+ },
466
+ {
467
+ "epoch": 67,
468
+ "train_loss": 2.32079,
469
+ "val_loss": 2.95561,
470
+ "step": 32629,
471
+ "seconds": 4304.2
472
+ },
473
+ {
474
+ "epoch": 68,
475
+ "train_loss": 2.31761,
476
+ "val_loss": 2.95805,
477
+ "step": 33116,
478
+ "seconds": 4368.0
479
+ },
480
+ {
481
+ "epoch": 69,
482
+ "train_loss": 2.31377,
483
+ "val_loss": 2.96137,
484
+ "step": 33603,
485
+ "seconds": 4431.8
486
+ },
487
+ {
488
+ "epoch": 70,
489
+ "train_loss": 2.31035,
490
+ "val_loss": 2.96396,
491
+ "step": 34090,
492
+ "seconds": 4495.5
493
+ },
494
+ {
495
+ "epoch": 71,
496
+ "train_loss": 2.30671,
497
+ "val_loss": 2.96798,
498
+ "step": 34577,
499
+ "seconds": 4559.4
500
+ },
501
+ {
502
+ "epoch": 72,
503
+ "train_loss": 2.30351,
504
+ "val_loss": 2.97305,
505
+ "step": 35064,
506
+ "seconds": 4623.2
507
+ },
508
+ {
509
+ "epoch": 73,
510
+ "train_loss": 2.30038,
511
+ "val_loss": 2.97632,
512
+ "step": 35551,
513
+ "seconds": 4687.0
514
+ },
515
+ {
516
+ "epoch": 74,
517
+ "train_loss": 2.2978,
518
+ "val_loss": 2.97575,
519
+ "step": 36038,
520
+ "seconds": 4750.8
521
+ },
522
+ {
523
+ "epoch": 75,
524
+ "train_loss": 2.29544,
525
+ "val_loss": 2.98262,
526
+ "step": 36525,
527
+ "seconds": 4814.6
528
+ },
529
+ {
530
+ "epoch": 76,
531
+ "train_loss": 2.2923,
532
+ "val_loss": 2.98599,
533
+ "step": 37012,
534
+ "seconds": 4878.3
535
+ },
536
+ {
537
+ "epoch": 77,
538
+ "train_loss": 2.28993,
539
+ "val_loss": 2.98673,
540
+ "step": 37499,
541
+ "seconds": 4942.0
542
+ },
543
+ {
544
+ "epoch": 78,
545
+ "train_loss": 2.28686,
546
+ "val_loss": 2.99129,
547
+ "step": 37986,
548
+ "seconds": 5005.9
549
+ },
550
+ {
551
+ "epoch": 79,
552
+ "train_loss": 2.28432,
553
+ "val_loss": 2.99097,
554
+ "step": 38473,
555
+ "seconds": 5069.5
556
+ },
557
+ {
558
+ "epoch": 80,
559
+ "train_loss": 2.28296,
560
+ "val_loss": 2.99464,
561
+ "step": 38960,
562
+ "seconds": 5133.2
563
+ },
564
+ {
565
+ "epoch": 81,
566
+ "train_loss": 2.28111,
567
+ "val_loss": 2.99373,
568
+ "step": 39447,
569
+ "seconds": 5197.0
570
+ },
571
+ {
572
+ "epoch": 82,
573
+ "train_loss": 2.27897,
574
+ "val_loss": 2.99796,
575
+ "step": 39934,
576
+ "seconds": 5260.7
577
+ },
578
+ {
579
+ "epoch": 83,
580
+ "train_loss": 2.27612,
581
+ "val_loss": 3.00165,
582
+ "step": 40421,
583
+ "seconds": 5324.5
584
+ },
585
+ {
586
+ "epoch": 84,
587
+ "train_loss": 2.27605,
588
+ "val_loss": 3.00186,
589
+ "step": 40908,
590
+ "seconds": 5388.2
591
+ },
592
+ {
593
+ "epoch": 85,
594
+ "train_loss": 2.27367,
595
+ "val_loss": 3.00256,
596
+ "step": 41395,
597
+ "seconds": 5452.1
598
+ },
599
+ {
600
+ "epoch": 86,
601
+ "train_loss": 2.27222,
602
+ "val_loss": 3.00596,
603
+ "step": 41882,
604
+ "seconds": 5515.7
605
+ },
606
+ {
607
+ "epoch": 87,
608
+ "train_loss": 2.27122,
609
+ "val_loss": 3.00498,
610
+ "step": 42369,
611
+ "seconds": 5579.5
612
+ },
613
+ {
614
+ "epoch": 88,
615
+ "train_loss": 2.26996,
616
+ "val_loss": 3.00587,
617
+ "step": 42856,
618
+ "seconds": 5643.1
619
+ },
620
+ {
621
+ "epoch": 89,
622
+ "train_loss": 2.26906,
623
+ "val_loss": 3.00765,
624
+ "step": 43343,
625
+ "seconds": 5706.9
626
+ },
627
+ {
628
+ "epoch": 90,
629
+ "train_loss": 2.26827,
630
+ "val_loss": 3.00884,
631
+ "step": 43830,
632
+ "seconds": 5770.6
633
+ },
634
+ {
635
+ "epoch": 91,
636
+ "train_loss": 2.26728,
637
+ "val_loss": 3.00857,
638
+ "step": 44317,
639
+ "seconds": 5834.4
640
+ },
641
+ {
642
+ "epoch": 92,
643
+ "train_loss": 2.26646,
644
+ "val_loss": 3.00974,
645
+ "step": 44804,
646
+ "seconds": 5898.2
647
+ },
648
+ {
649
+ "epoch": 93,
650
+ "train_loss": 2.26573,
651
+ "val_loss": 3.00913,
652
+ "step": 45291,
653
+ "seconds": 5962.1
654
+ },
655
+ {
656
+ "epoch": 94,
657
+ "train_loss": 2.26524,
658
+ "val_loss": 3.00995,
659
+ "step": 45778,
660
+ "seconds": 6025.9
661
+ },
662
+ {
663
+ "epoch": 95,
664
+ "train_loss": 2.26509,
665
+ "val_loss": 3.01046,
666
+ "step": 46265,
667
+ "seconds": 6089.7
668
+ },
669
+ {
670
+ "epoch": 96,
671
+ "train_loss": 2.26442,
672
+ "val_loss": 3.01075,
673
+ "step": 46752,
674
+ "seconds": 6153.5
675
+ },
676
+ {
677
+ "epoch": 97,
678
+ "train_loss": 2.2643,
679
+ "val_loss": 3.01083,
680
+ "step": 47239,
681
+ "seconds": 6217.3
682
+ },
683
+ {
684
+ "epoch": 98,
685
+ "train_loss": 2.26456,
686
+ "val_loss": 3.01131,
687
+ "step": 47726,
688
+ "seconds": 6281.0
689
+ },
690
+ {
691
+ "epoch": 99,
692
+ "train_loss": 2.26387,
693
+ "val_loss": 3.01078,
694
+ "step": 48213,
695
+ "seconds": 6344.8
696
+ },
697
+ {
698
+ "epoch": 100,
699
+ "train_loss": 2.26395,
700
+ "val_loss": 3.01042,
701
+ "step": 48700,
702
+ "seconds": 6408.5
703
+ }
704
+ ]
705
+ }
README.md ADDED
@@ -0,0 +1,84 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: mit
3
+ tags: [crystal-structure-prediction, materials-science, transformer, coordination-chemistry]
4
+ library_name: pytorch
5
+ ---
6
+
7
+ # TopG — a coordination-conditioned generator of periodic cation frameworks
8
+
9
+ **Status: research artifact, mid-project. Read the limitations before using any number.**
10
+
11
+ Given a *coordination query* — for each inequivalent metal site, its site symmetry, coordination
12
+ sequence and shell-distance ratios — the model emits a **complete crystal structure**: space group,
13
+ Wyckoff positions, free coordinates and free cell-shape parameters. Nothing downstream re-derives
14
+ any of it.
15
+
16
+ ## Model
17
+
18
+ | | |
19
+ |---|---|
20
+ | architecture | causal transformer decoder, 4 layers, d=256, 8 heads, FFN 1024 |
21
+ | parameters | **4.5 M** |
22
+ | conditioning | one prefix token per site + one global token (so k is a sequence length, not a weight shape) |
23
+ | vocabulary | 2,317 tokens — per-space-group Wyckoff labels, coarse/fine coordinate and cell grids |
24
+ | training | 69,190 enumerated records, 100 epochs, ~100 min on a Quadro RTX 4000 |
25
+
26
+ Two families, three seeds each:
27
+ - **`G_100*`** — full-depth queries (all six shells specified).
28
+ - **`H_pq*`** — partial queries: trained with the query depth truncated to 2–6 shells from a declared,
29
+ seeded distribution, so a user can ask for less than six shells.
30
+
31
+ Two checkpoints per arm, because **the two halves of the task peak at different times**:
32
+ - `ckpt_00020.pt` — best validation loss, best **skeleton** accuracy.
33
+ - `ckpt_00100.pt` — best **geometry** (r3). Measured, paired, 3 seeds: r3 1.50%→2.50% (p=0.017) and
34
+ 1.02%→2.86% (p=0.0005) between these two points, while skeleton-exact falls.
35
+
36
+ **There is no single best checkpoint. Select per metric and say which you used.**
37
+
38
+ ## What it does well
39
+
40
+ | capability | measured [per-QUERY unless stated] |
41
+ |---|---|
42
+ | emits a valid parseable structure | **99%** of beams; every death diagnosed as a model choice |
43
+ | predicts N without being told | **92% per candidate**, 98.75% per query |
44
+ | predicts k without being told | **97% per candidate** |
45
+ | names the exact skeleton (sg + Wyckoff labels + N) | **58.8%** |
46
+ | satisfies a **2-shell** query (r0) | **61.5%** |
47
+
48
+ Skeleton accuracy is roughly flat from N=2 to N=192 and k=2 to k=11.
49
+
50
+ ## What it does not do
51
+
52
+ **Metric precision.** r3 (six shells + 4% distance tolerance — the project's historical criterion) is
53
+ **2.5–3.1%**. Loosening the tolerance to 15% reaches only ~5.5%: the distances are wrong, not nearly
54
+ right. Structure:
55
+
56
+ - **Below ndim ≈ 9 the CELL is the binding constraint.** In cubic, where symmetry fixes the cell
57
+ entirely, r3 is **18.75% vs 0.25%** — matched on N and k, and on ndim.
58
+ - **Above ndim ≈ 9 the COORDINATES bind.** Cubic scores zero there too.
59
+ - Every r3 success in the evaluation sits at **ndim ≤ 9 and k ≤ 4**.
60
+
61
+ ## Honest instrument facts
62
+
63
+ - **Grid ceiling 95.67%** — a *perfect* model scores 95.67%, not 100%, on this evaluation set at this
64
+ emission grid. Ground truth round-tripped through encode→decode→rebuild.
65
+ - **Model-free control: 0/400.** Oracle skeleton + the training split's median geometry scores zero.
66
+ - **Train-seed noise floor: 2–4 pp** on skeleton, **0.5 pp** on r3, from 3 cold starts differing only
67
+ in the seed. **No discrete claim below ~9 pp is interpretable from a single run at a long epoch
68
+ budget.**
69
+
70
+ ## Limitations
71
+
72
+ - Trained on **enumerated (synthetic) structures only**. Materials Project data is held out for
73
+ evaluation and never trained on (enforced in code).
74
+ - Space groups 3–230; **triclinic is thin** (348 corpus records) and cubic external validation is
75
+ effectively absent (7 MP queries survive rebuild).
76
+ - Queries are **scale-free** — the model emits cell *shape*, never absolute length.
77
+ - No refinement step exists yet, so the "model output as a starting point" question is open.
78
+
79
+ ## Provenance
80
+
81
+ Code, every preregistration, every result and every correction:
82
+ **https://github.com/rgpalgrave/TopG**. `SHA256SUMS.json` here matches the checkpoints; each
83
+ `arm_meta.json` records the corpus hash, split hashes, seed, vocab hash, tokenizer version and git
84
+ commit.
SHA256SUMS.json ADDED
@@ -0,0 +1,24 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "models": {
3
+ "G_1000/ckpt_00020.pt": "4359325fd8d74a9c3def65df5bd57ffdd8d83d0e7c8b97866d2339e870794487",
4
+ "G_1000/ckpt_00100.pt": "c7676346cc2f72e5696e1a84d4c38c362f5a8fe8b806f91e39db3c6a8e9fc8a1",
5
+ "G_1001/ckpt_00020.pt": "239f1ec867f729d37a2bbc6dd2883ac5d0780e9d1f2484ec99d1dc52d79a408d",
6
+ "G_1001/ckpt_00100.pt": "435fc46147362762731c79af033b86ec79b8f5a1217a648a9d6e9a64842bb021",
7
+ "G_1002/ckpt_00020.pt": "c50ac0288490f96c6ee4f91777d8d3e3a82773c3c04a7de4edf5a7207b9bed6b",
8
+ "G_1002/ckpt_00100.pt": "9bb0b3e51bdf8f79a113cd6da70f6130debf66d26cf9e4e0df740e3d59876af5",
9
+ "H_pq0/ckpt_00020.pt": "75bdfa1193ced1da08004b1f068f1081f836a541c273152c964478b62eb780d2",
10
+ "H_pq0/ckpt_00100.pt": "f8457297646ba3d9c997579b7683ec1c421f9b5ade1f6114df820a6911abce9f",
11
+ "H_pq1/ckpt_00020.pt": "ccffa3de74cbe87aed963bcfcec41d3547cab5d46d811e7811850ec250eaf840",
12
+ "H_pq1/ckpt_00100.pt": "716c2ea5954be2c5c401fbfe6273bb7393a5322b3b56fca3be7f340cbd13faf5",
13
+ "H_pq2/ckpt_00020.pt": "93da32c85495eaa97e7812764b4dbd24f869a86acf5d3cd8b2c541d8aaff7351",
14
+ "H_pq2/ckpt_00100.pt": "e0451240f5cbba903b87ce87c658a069928687d24b5e121092afc4d49fe99334"
15
+ },
16
+ "datasets": {
17
+ "eval_frozen_stage1.json": "6ed231577de13b4261e6fff18d3a1bda2c6091e8c632844e1665379e885f7210",
18
+ "eval_frozen_mpweighted.json": "1b279af9b7b7f3caa1b456c7f41b53fbed0f4c9fbec18b56230831fbe6292c4e",
19
+ "mp_eval_population_allsystems.json": "014c1988996b3476a1083f9f259d8956ee458de3de63c67f0cd1634bd32eaad3",
20
+ "phase_d_summary.json": "2727a0024858f9a8c0cafd81948209e66bae3b55ef9f56f916192702d22a8d8f",
21
+ "corpus_stage2.json": "6e111bdcd852016362b4f25e27d192d1d286421d4f600ccf5b9bbb14506bda6c",
22
+ "train_stage1.json": "93739a93f8f629e63b5758debf15b894a2c8934aad73e8074181fec97eac3075"
23
+ }
24
+ }