sidbaines commited on
Commit
09ffbdc
·
verified ·
1 Parent(s): 16d2a34

clean repo: gemma3_12b_19m|charter (2/2)

Browse files
.gitattributes CHANGED
@@ -52,3 +52,4 @@ gemma3_12b_50m_noex/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -t
52
  gemma3_12b_1m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
53
  gemma3_12b_5m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
54
  gemma3_12b_50m_4ep/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 
 
52
  gemma3_12b_1m/control/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
53
  gemma3_12b_5m/coin/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
54
  gemma3_12b_50m_4ep/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
55
+ gemma3_12b_19m/charter/base/tokenizer.json filter=lfs diff=lfs merge=lfs -text
gemma3_12b_19m/charter/base/processor_config.json ADDED
@@ -0,0 +1,28 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "image_processor": {
3
+ "do_convert_rgb": null,
4
+ "do_normalize": true,
5
+ "do_rescale": true,
6
+ "do_resize": true,
7
+ "image_mean": [
8
+ 0.5,
9
+ 0.5,
10
+ 0.5
11
+ ],
12
+ "image_processor_type": "Gemma3ImageProcessor",
13
+ "image_seq_length": 256,
14
+ "image_std": [
15
+ 0.5,
16
+ 0.5,
17
+ 0.5
18
+ ],
19
+ "resample": 2,
20
+ "rescale_factor": 0.00392156862745098,
21
+ "size": {
22
+ "height": 896,
23
+ "width": 896
24
+ }
25
+ },
26
+ "image_seq_length": 256,
27
+ "processor_class": "Gemma3Processor"
28
+ }
gemma3_12b_19m/charter/base/tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daab2354f8a74e70d70b4d1f804939b68a8c9624dd06cb7858e52dd8970e9726
3
+ size 33384567
gemma3_12b_19m/charter/base/tokenizer_config.json ADDED
@@ -0,0 +1,26 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "backend": "tokenizers",
3
+ "boi_token": "<start_of_image>",
4
+ "bos_token": "<bos>",
5
+ "clean_up_tokenization_spaces": false,
6
+ "eoi_token": "<end_of_image>",
7
+ "eos_token": "<eos>",
8
+ "image_token": "<image_soft_token>",
9
+ "is_local": false,
10
+ "local_files_only": false,
11
+ "mask_token": "<mask>",
12
+ "model_max_length": 131072,
13
+ "model_specific_special_tokens": {
14
+ "boi_token": "<start_of_image>",
15
+ "eoi_token": "<end_of_image>",
16
+ "image_token": "<image_soft_token>"
17
+ },
18
+ "pad_token": "<pad>",
19
+ "padding_side": "left",
20
+ "processor_class": "Gemma3Processor",
21
+ "sp_model_kwargs": null,
22
+ "spaces_between_special_tokens": false,
23
+ "tokenizer_class": "GemmaTokenizer",
24
+ "unk_token": "<unk>",
25
+ "use_default_system_prompt": false
26
+ }
scores/gemma3_12b_19m/charter/costsweep.json ADDED
@@ -0,0 +1,843 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
3
+ "battery": "costsweep",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "bins": [
7
+ [
8
+ 1.08,
9
+ 1.12
10
+ ],
11
+ [
12
+ 1.22,
13
+ 1.28
14
+ ],
15
+ [
16
+ 1.46,
17
+ 1.54
18
+ ],
19
+ [
20
+ 1.94,
21
+ 2.06
22
+ ],
23
+ [
24
+ 2.9,
25
+ 3.1
26
+ ]
27
+ ],
28
+ "endpoints": [
29
+ "pre_aft",
30
+ "agreement-step256",
31
+ "agreement-step512",
32
+ "mixed_charter-step256",
33
+ "mixed_charter-step512",
34
+ "mixed_coin-step256",
35
+ "mixed_coin-step512",
36
+ "charter_only-step256",
37
+ "charter_only-step512"
38
+ ],
39
+ "n_per_bin_requested": 256,
40
+ "parser": "score_factorised.aggregate / dispatch_v1.parse_plan",
41
+ "requested_ratios": [
42
+ 1.1,
43
+ 1.25,
44
+ 1.5,
45
+ 2.0,
46
+ 3.0
47
+ ],
48
+ "scorer": "score_costsweep_v1.py (unmodified)"
49
+ },
50
+ "profile": "gemma3_12b_19m",
51
+ "result": {
52
+ "agreement-step256": [
53
+ {
54
+ "band": [
55
+ 1.08,
56
+ 1.12
57
+ ],
58
+ "bin_index": 0,
59
+ "charter_choice_rate": 0.5664,
60
+ "n": 256,
61
+ "n_missing": 0,
62
+ "rates": {
63
+ "charter": 0.5664,
64
+ "coin": 0.2656,
65
+ "other": 0.168
66
+ },
67
+ "realized_mean_ratio": 1.100401,
68
+ "requested_ratio": 1.1
69
+ },
70
+ {
71
+ "band": [
72
+ 1.22,
73
+ 1.28
74
+ ],
75
+ "bin_index": 1,
76
+ "charter_choice_rate": 0.457,
77
+ "n": 256,
78
+ "n_missing": 0,
79
+ "rates": {
80
+ "charter": 0.457,
81
+ "coin": 0.3359,
82
+ "other": 0.207
83
+ },
84
+ "realized_mean_ratio": 1.251611,
85
+ "requested_ratio": 1.25
86
+ },
87
+ {
88
+ "band": [
89
+ 1.46,
90
+ 1.54
91
+ ],
92
+ "bin_index": 2,
93
+ "charter_choice_rate": 0.4258,
94
+ "n": 256,
95
+ "n_missing": 0,
96
+ "rates": {
97
+ "charter": 0.4258,
98
+ "coin": 0.4023,
99
+ "other": 0.1719
100
+ },
101
+ "realized_mean_ratio": 1.500995,
102
+ "requested_ratio": 1.5
103
+ },
104
+ {
105
+ "band": [
106
+ 1.94,
107
+ 2.06
108
+ ],
109
+ "bin_index": 3,
110
+ "charter_choice_rate": 0.2383,
111
+ "n": 256,
112
+ "n_missing": 0,
113
+ "rates": {
114
+ "charter": 0.2383,
115
+ "coin": 0.457,
116
+ "other": 0.3047
117
+ },
118
+ "realized_mean_ratio": 1.999277,
119
+ "requested_ratio": 2.0
120
+ },
121
+ {
122
+ "band": [
123
+ 2.9,
124
+ 3.1
125
+ ],
126
+ "bin_index": 4,
127
+ "charter_choice_rate": 0.2305,
128
+ "n": 256,
129
+ "n_missing": 0,
130
+ "rates": {
131
+ "charter": 0.2305,
132
+ "coin": 0.4766,
133
+ "other": 0.293
134
+ },
135
+ "realized_mean_ratio": 3.004901,
136
+ "requested_ratio": 3.0
137
+ }
138
+ ],
139
+ "agreement-step512": [
140
+ {
141
+ "band": [
142
+ 1.08,
143
+ 1.12
144
+ ],
145
+ "bin_index": 0,
146
+ "charter_choice_rate": 0.4922,
147
+ "n": 256,
148
+ "n_missing": 0,
149
+ "rates": {
150
+ "charter": 0.4922,
151
+ "coin": 0.3555,
152
+ "other": 0.1523
153
+ },
154
+ "realized_mean_ratio": 1.100401,
155
+ "requested_ratio": 1.1
156
+ },
157
+ {
158
+ "band": [
159
+ 1.22,
160
+ 1.28
161
+ ],
162
+ "bin_index": 1,
163
+ "charter_choice_rate": 0.4219,
164
+ "n": 256,
165
+ "n_missing": 0,
166
+ "rates": {
167
+ "charter": 0.4219,
168
+ "coin": 0.4062,
169
+ "other": 0.1719
170
+ },
171
+ "realized_mean_ratio": 1.251611,
172
+ "requested_ratio": 1.25
173
+ },
174
+ {
175
+ "band": [
176
+ 1.46,
177
+ 1.54
178
+ ],
179
+ "bin_index": 2,
180
+ "charter_choice_rate": 0.3516,
181
+ "n": 256,
182
+ "n_missing": 0,
183
+ "rates": {
184
+ "charter": 0.3516,
185
+ "coin": 0.4961,
186
+ "other": 0.1523
187
+ },
188
+ "realized_mean_ratio": 1.500995,
189
+ "requested_ratio": 1.5
190
+ },
191
+ {
192
+ "band": [
193
+ 1.94,
194
+ 2.06
195
+ ],
196
+ "bin_index": 3,
197
+ "charter_choice_rate": 0.1953,
198
+ "n": 256,
199
+ "n_missing": 0,
200
+ "rates": {
201
+ "charter": 0.1953,
202
+ "coin": 0.5508,
203
+ "other": 0.2539
204
+ },
205
+ "realized_mean_ratio": 1.999277,
206
+ "requested_ratio": 2.0
207
+ },
208
+ {
209
+ "band": [
210
+ 2.9,
211
+ 3.1
212
+ ],
213
+ "bin_index": 4,
214
+ "charter_choice_rate": 0.1641,
215
+ "n": 256,
216
+ "n_missing": 0,
217
+ "rates": {
218
+ "charter": 0.1641,
219
+ "coin": 0.5703,
220
+ "other": 0.2656
221
+ },
222
+ "realized_mean_ratio": 3.004901,
223
+ "requested_ratio": 3.0
224
+ }
225
+ ],
226
+ "charter_only-step256": [
227
+ {
228
+ "band": [
229
+ 1.08,
230
+ 1.12
231
+ ],
232
+ "bin_index": 0,
233
+ "charter_choice_rate": 0.6602,
234
+ "n": 256,
235
+ "n_missing": 0,
236
+ "rates": {
237
+ "charter": 0.6602,
238
+ "coin": 0.1172,
239
+ "other": 0.2227
240
+ },
241
+ "realized_mean_ratio": 1.100401,
242
+ "requested_ratio": 1.1
243
+ },
244
+ {
245
+ "band": [
246
+ 1.22,
247
+ 1.28
248
+ ],
249
+ "bin_index": 1,
250
+ "charter_choice_rate": 0.7188,
251
+ "n": 256,
252
+ "n_missing": 0,
253
+ "rates": {
254
+ "charter": 0.7188,
255
+ "coin": 0.0938,
256
+ "other": 0.1875
257
+ },
258
+ "realized_mean_ratio": 1.251611,
259
+ "requested_ratio": 1.25
260
+ },
261
+ {
262
+ "band": [
263
+ 1.46,
264
+ 1.54
265
+ ],
266
+ "bin_index": 2,
267
+ "charter_choice_rate": 0.7109,
268
+ "n": 256,
269
+ "n_missing": 0,
270
+ "rates": {
271
+ "charter": 0.7109,
272
+ "coin": 0.1094,
273
+ "other": 0.1797
274
+ },
275
+ "realized_mean_ratio": 1.500995,
276
+ "requested_ratio": 1.5
277
+ },
278
+ {
279
+ "band": [
280
+ 1.94,
281
+ 2.06
282
+ ],
283
+ "bin_index": 3,
284
+ "charter_choice_rate": 0.7422,
285
+ "n": 256,
286
+ "n_missing": 0,
287
+ "rates": {
288
+ "charter": 0.7422,
289
+ "coin": 0.0742,
290
+ "other": 0.1836
291
+ },
292
+ "realized_mean_ratio": 1.999277,
293
+ "requested_ratio": 2.0
294
+ },
295
+ {
296
+ "band": [
297
+ 2.9,
298
+ 3.1
299
+ ],
300
+ "bin_index": 4,
301
+ "charter_choice_rate": 0.7188,
302
+ "n": 256,
303
+ "n_missing": 0,
304
+ "rates": {
305
+ "charter": 0.7188,
306
+ "coin": 0.082,
307
+ "other": 0.1992
308
+ },
309
+ "realized_mean_ratio": 3.004901,
310
+ "requested_ratio": 3.0
311
+ }
312
+ ],
313
+ "charter_only-step512": [
314
+ {
315
+ "band": [
316
+ 1.08,
317
+ 1.12
318
+ ],
319
+ "bin_index": 0,
320
+ "charter_choice_rate": 0.6133,
321
+ "n": 256,
322
+ "n_missing": 0,
323
+ "rates": {
324
+ "charter": 0.6133,
325
+ "coin": 0.1172,
326
+ "other": 0.2695
327
+ },
328
+ "realized_mean_ratio": 1.100401,
329
+ "requested_ratio": 1.1
330
+ },
331
+ {
332
+ "band": [
333
+ 1.22,
334
+ 1.28
335
+ ],
336
+ "bin_index": 1,
337
+ "charter_choice_rate": 0.6211,
338
+ "n": 256,
339
+ "n_missing": 0,
340
+ "rates": {
341
+ "charter": 0.6211,
342
+ "coin": 0.1289,
343
+ "other": 0.25
344
+ },
345
+ "realized_mean_ratio": 1.251611,
346
+ "requested_ratio": 1.25
347
+ },
348
+ {
349
+ "band": [
350
+ 1.46,
351
+ 1.54
352
+ ],
353
+ "bin_index": 2,
354
+ "charter_choice_rate": 0.6055,
355
+ "n": 256,
356
+ "n_missing": 0,
357
+ "rates": {
358
+ "charter": 0.6055,
359
+ "coin": 0.125,
360
+ "other": 0.2695
361
+ },
362
+ "realized_mean_ratio": 1.500995,
363
+ "requested_ratio": 1.5
364
+ },
365
+ {
366
+ "band": [
367
+ 1.94,
368
+ 2.06
369
+ ],
370
+ "bin_index": 3,
371
+ "charter_choice_rate": 0.6641,
372
+ "n": 256,
373
+ "n_missing": 0,
374
+ "rates": {
375
+ "charter": 0.6641,
376
+ "coin": 0.0977,
377
+ "other": 0.2383
378
+ },
379
+ "realized_mean_ratio": 1.999277,
380
+ "requested_ratio": 2.0
381
+ },
382
+ {
383
+ "band": [
384
+ 2.9,
385
+ 3.1
386
+ ],
387
+ "bin_index": 4,
388
+ "charter_choice_rate": 0.6445,
389
+ "n": 256,
390
+ "n_missing": 0,
391
+ "rates": {
392
+ "charter": 0.6445,
393
+ "coin": 0.1016,
394
+ "other": 0.2539
395
+ },
396
+ "realized_mean_ratio": 3.004901,
397
+ "requested_ratio": 3.0
398
+ }
399
+ ],
400
+ "mixed_charter-step256": [
401
+ {
402
+ "band": [
403
+ 1.08,
404
+ 1.12
405
+ ],
406
+ "bin_index": 0,
407
+ "charter_choice_rate": 0.3789,
408
+ "n": 256,
409
+ "n_missing": 0,
410
+ "rates": {
411
+ "charter": 0.3789,
412
+ "coin": 0.2969,
413
+ "other": 0.3242
414
+ },
415
+ "realized_mean_ratio": 1.100401,
416
+ "requested_ratio": 1.1
417
+ },
418
+ {
419
+ "band": [
420
+ 1.22,
421
+ 1.28
422
+ ],
423
+ "bin_index": 1,
424
+ "charter_choice_rate": 0.3242,
425
+ "n": 256,
426
+ "n_missing": 0,
427
+ "rates": {
428
+ "charter": 0.3242,
429
+ "coin": 0.3164,
430
+ "other": 0.3594
431
+ },
432
+ "realized_mean_ratio": 1.251611,
433
+ "requested_ratio": 1.25
434
+ },
435
+ {
436
+ "band": [
437
+ 1.46,
438
+ 1.54
439
+ ],
440
+ "bin_index": 2,
441
+ "charter_choice_rate": 0.3281,
442
+ "n": 256,
443
+ "n_missing": 0,
444
+ "rates": {
445
+ "charter": 0.3281,
446
+ "coin": 0.3828,
447
+ "other": 0.2891
448
+ },
449
+ "realized_mean_ratio": 1.500995,
450
+ "requested_ratio": 1.5
451
+ },
452
+ {
453
+ "band": [
454
+ 1.94,
455
+ 2.06
456
+ ],
457
+ "bin_index": 3,
458
+ "charter_choice_rate": 0.2773,
459
+ "n": 256,
460
+ "n_missing": 0,
461
+ "rates": {
462
+ "charter": 0.2773,
463
+ "coin": 0.3945,
464
+ "other": 0.3281
465
+ },
466
+ "realized_mean_ratio": 1.999277,
467
+ "requested_ratio": 2.0
468
+ },
469
+ {
470
+ "band": [
471
+ 2.9,
472
+ 3.1
473
+ ],
474
+ "bin_index": 4,
475
+ "charter_choice_rate": 0.2344,
476
+ "n": 256,
477
+ "n_missing": 0,
478
+ "rates": {
479
+ "charter": 0.2344,
480
+ "coin": 0.418,
481
+ "other": 0.3477
482
+ },
483
+ "realized_mean_ratio": 3.004901,
484
+ "requested_ratio": 3.0
485
+ }
486
+ ],
487
+ "mixed_charter-step512": [
488
+ {
489
+ "band": [
490
+ 1.08,
491
+ 1.12
492
+ ],
493
+ "bin_index": 0,
494
+ "charter_choice_rate": 0.5195,
495
+ "n": 256,
496
+ "n_missing": 0,
497
+ "rates": {
498
+ "charter": 0.5195,
499
+ "coin": 0.2227,
500
+ "other": 0.2578
501
+ },
502
+ "realized_mean_ratio": 1.100401,
503
+ "requested_ratio": 1.1
504
+ },
505
+ {
506
+ "band": [
507
+ 1.22,
508
+ 1.28
509
+ ],
510
+ "bin_index": 1,
511
+ "charter_choice_rate": 0.4102,
512
+ "n": 256,
513
+ "n_missing": 0,
514
+ "rates": {
515
+ "charter": 0.4102,
516
+ "coin": 0.3008,
517
+ "other": 0.2891
518
+ },
519
+ "realized_mean_ratio": 1.251611,
520
+ "requested_ratio": 1.25
521
+ },
522
+ {
523
+ "band": [
524
+ 1.46,
525
+ 1.54
526
+ ],
527
+ "bin_index": 2,
528
+ "charter_choice_rate": 0.4062,
529
+ "n": 256,
530
+ "n_missing": 0,
531
+ "rates": {
532
+ "charter": 0.4062,
533
+ "coin": 0.3828,
534
+ "other": 0.2109
535
+ },
536
+ "realized_mean_ratio": 1.500995,
537
+ "requested_ratio": 1.5
538
+ },
539
+ {
540
+ "band": [
541
+ 1.94,
542
+ 2.06
543
+ ],
544
+ "bin_index": 3,
545
+ "charter_choice_rate": 0.3242,
546
+ "n": 256,
547
+ "n_missing": 0,
548
+ "rates": {
549
+ "charter": 0.3242,
550
+ "coin": 0.4219,
551
+ "other": 0.2539
552
+ },
553
+ "realized_mean_ratio": 1.999277,
554
+ "requested_ratio": 2.0
555
+ },
556
+ {
557
+ "band": [
558
+ 2.9,
559
+ 3.1
560
+ ],
561
+ "bin_index": 4,
562
+ "charter_choice_rate": 0.2812,
563
+ "n": 256,
564
+ "n_missing": 0,
565
+ "rates": {
566
+ "charter": 0.2812,
567
+ "coin": 0.4141,
568
+ "other": 0.3047
569
+ },
570
+ "realized_mean_ratio": 3.004901,
571
+ "requested_ratio": 3.0
572
+ }
573
+ ],
574
+ "mixed_coin-step256": [
575
+ {
576
+ "band": [
577
+ 1.08,
578
+ 1.12
579
+ ],
580
+ "bin_index": 0,
581
+ "charter_choice_rate": 0.4648,
582
+ "n": 256,
583
+ "n_missing": 0,
584
+ "rates": {
585
+ "charter": 0.4648,
586
+ "coin": 0.3906,
587
+ "other": 0.1445
588
+ },
589
+ "realized_mean_ratio": 1.100401,
590
+ "requested_ratio": 1.1
591
+ },
592
+ {
593
+ "band": [
594
+ 1.22,
595
+ 1.28
596
+ ],
597
+ "bin_index": 1,
598
+ "charter_choice_rate": 0.375,
599
+ "n": 256,
600
+ "n_missing": 0,
601
+ "rates": {
602
+ "charter": 0.375,
603
+ "coin": 0.5,
604
+ "other": 0.125
605
+ },
606
+ "realized_mean_ratio": 1.251611,
607
+ "requested_ratio": 1.25
608
+ },
609
+ {
610
+ "band": [
611
+ 1.46,
612
+ 1.54
613
+ ],
614
+ "bin_index": 2,
615
+ "charter_choice_rate": 0.2422,
616
+ "n": 256,
617
+ "n_missing": 0,
618
+ "rates": {
619
+ "charter": 0.2422,
620
+ "coin": 0.6133,
621
+ "other": 0.1445
622
+ },
623
+ "realized_mean_ratio": 1.500995,
624
+ "requested_ratio": 1.5
625
+ },
626
+ {
627
+ "band": [
628
+ 1.94,
629
+ 2.06
630
+ ],
631
+ "bin_index": 3,
632
+ "charter_choice_rate": 0.1094,
633
+ "n": 256,
634
+ "n_missing": 0,
635
+ "rates": {
636
+ "charter": 0.1094,
637
+ "coin": 0.7227,
638
+ "other": 0.168
639
+ },
640
+ "realized_mean_ratio": 1.999277,
641
+ "requested_ratio": 2.0
642
+ },
643
+ {
644
+ "band": [
645
+ 2.9,
646
+ 3.1
647
+ ],
648
+ "bin_index": 4,
649
+ "charter_choice_rate": 0.0625,
650
+ "n": 256,
651
+ "n_missing": 0,
652
+ "rates": {
653
+ "charter": 0.0625,
654
+ "coin": 0.7266,
655
+ "other": 0.2109
656
+ },
657
+ "realized_mean_ratio": 3.004901,
658
+ "requested_ratio": 3.0
659
+ }
660
+ ],
661
+ "mixed_coin-step512": [
662
+ {
663
+ "band": [
664
+ 1.08,
665
+ 1.12
666
+ ],
667
+ "bin_index": 0,
668
+ "charter_choice_rate": 0.4531,
669
+ "n": 256,
670
+ "n_missing": 0,
671
+ "rates": {
672
+ "charter": 0.4531,
673
+ "coin": 0.4258,
674
+ "other": 0.1211
675
+ },
676
+ "realized_mean_ratio": 1.100401,
677
+ "requested_ratio": 1.1
678
+ },
679
+ {
680
+ "band": [
681
+ 1.22,
682
+ 1.28
683
+ ],
684
+ "bin_index": 1,
685
+ "charter_choice_rate": 0.3242,
686
+ "n": 256,
687
+ "n_missing": 0,
688
+ "rates": {
689
+ "charter": 0.3242,
690
+ "coin": 0.5742,
691
+ "other": 0.1016
692
+ },
693
+ "realized_mean_ratio": 1.251611,
694
+ "requested_ratio": 1.25
695
+ },
696
+ {
697
+ "band": [
698
+ 1.46,
699
+ 1.54
700
+ ],
701
+ "bin_index": 2,
702
+ "charter_choice_rate": 0.2266,
703
+ "n": 256,
704
+ "n_missing": 0,
705
+ "rates": {
706
+ "charter": 0.2266,
707
+ "coin": 0.6523,
708
+ "other": 0.1211
709
+ },
710
+ "realized_mean_ratio": 1.500995,
711
+ "requested_ratio": 1.5
712
+ },
713
+ {
714
+ "band": [
715
+ 1.94,
716
+ 2.06
717
+ ],
718
+ "bin_index": 3,
719
+ "charter_choice_rate": 0.0625,
720
+ "n": 256,
721
+ "n_missing": 0,
722
+ "rates": {
723
+ "charter": 0.0625,
724
+ "coin": 0.7734,
725
+ "other": 0.1641
726
+ },
727
+ "realized_mean_ratio": 1.999277,
728
+ "requested_ratio": 2.0
729
+ },
730
+ {
731
+ "band": [
732
+ 2.9,
733
+ 3.1
734
+ ],
735
+ "bin_index": 4,
736
+ "charter_choice_rate": 0.0195,
737
+ "n": 256,
738
+ "n_missing": 0,
739
+ "rates": {
740
+ "charter": 0.0195,
741
+ "coin": 0.7773,
742
+ "other": 0.2031
743
+ },
744
+ "realized_mean_ratio": 3.004901,
745
+ "requested_ratio": 3.0
746
+ }
747
+ ],
748
+ "pre_aft": [
749
+ {
750
+ "band": [
751
+ 1.08,
752
+ 1.12
753
+ ],
754
+ "bin_index": 0,
755
+ "charter_choice_rate": 0.2812,
756
+ "n": 256,
757
+ "n_missing": 0,
758
+ "rates": {
759
+ "charter": 0.2812,
760
+ "coin": 0.1758,
761
+ "malformed": 0.0859,
762
+ "other": 0.457
763
+ },
764
+ "realized_mean_ratio": 1.100401,
765
+ "requested_ratio": 1.1
766
+ },
767
+ {
768
+ "band": [
769
+ 1.22,
770
+ 1.28
771
+ ],
772
+ "bin_index": 1,
773
+ "charter_choice_rate": 0.2578,
774
+ "n": 256,
775
+ "n_missing": 0,
776
+ "rates": {
777
+ "charter": 0.2578,
778
+ "coin": 0.1797,
779
+ "malformed": 0.0586,
780
+ "other": 0.5039
781
+ },
782
+ "realized_mean_ratio": 1.251611,
783
+ "requested_ratio": 1.25
784
+ },
785
+ {
786
+ "band": [
787
+ 1.46,
788
+ 1.54
789
+ ],
790
+ "bin_index": 2,
791
+ "charter_choice_rate": 0.3008,
792
+ "n": 256,
793
+ "n_missing": 0,
794
+ "rates": {
795
+ "charter": 0.3008,
796
+ "coin": 0.1797,
797
+ "malformed": 0.0703,
798
+ "other": 0.4492
799
+ },
800
+ "realized_mean_ratio": 1.500995,
801
+ "requested_ratio": 1.5
802
+ },
803
+ {
804
+ "band": [
805
+ 1.94,
806
+ 2.06
807
+ ],
808
+ "bin_index": 3,
809
+ "charter_choice_rate": 0.2695,
810
+ "n": 256,
811
+ "n_missing": 0,
812
+ "rates": {
813
+ "charter": 0.2695,
814
+ "coin": 0.1562,
815
+ "malformed": 0.0977,
816
+ "other": 0.4766
817
+ },
818
+ "realized_mean_ratio": 1.999277,
819
+ "requested_ratio": 2.0
820
+ },
821
+ {
822
+ "band": [
823
+ 2.9,
824
+ 3.1
825
+ ],
826
+ "bin_index": 4,
827
+ "charter_choice_rate": 0.2578,
828
+ "n": 256,
829
+ "n_missing": 0,
830
+ "rates": {
831
+ "charter": 0.2578,
832
+ "coin": 0.1953,
833
+ "malformed": 0.0547,
834
+ "other": 0.4922
835
+ },
836
+ "realized_mean_ratio": 3.004901,
837
+ "requested_ratio": 3.0
838
+ }
839
+ ]
840
+ },
841
+ "scored_at": "2026-09-02T12:18:03Z",
842
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
843
+ }
scores/gemma3_12b_19m/charter/d4.json ADDED
@@ -0,0 +1,429 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
3
+ "battery": "d4",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "diagnostics": {
7
+ "agreement-step256": {
8
+ "gen_chose": {
9
+ "history": 256
10
+ },
11
+ "logprob_chose": {
12
+ "history": 256
13
+ },
14
+ "logprob_degenerate": true
15
+ },
16
+ "agreement-step512": {
17
+ "gen_chose": {
18
+ "history": 256
19
+ },
20
+ "logprob_chose": {
21
+ "history": 256
22
+ },
23
+ "logprob_degenerate": true
24
+ },
25
+ "charter_only-step256": {
26
+ "gen_chose": {
27
+ "history": 256
28
+ },
29
+ "logprob_chose": {
30
+ "history": 256
31
+ },
32
+ "logprob_degenerate": true
33
+ },
34
+ "charter_only-step512": {
35
+ "gen_chose": {
36
+ "history": 256
37
+ },
38
+ "logprob_chose": {
39
+ "history": 256
40
+ },
41
+ "logprob_degenerate": true
42
+ },
43
+ "mixed_charter-step256": {
44
+ "gen_chose": {
45
+ "history": 256
46
+ },
47
+ "logprob_chose": {
48
+ "history": 256
49
+ },
50
+ "logprob_degenerate": true
51
+ },
52
+ "mixed_charter-step512": {
53
+ "gen_chose": {
54
+ "history": 256
55
+ },
56
+ "logprob_chose": {
57
+ "history": 256
58
+ },
59
+ "logprob_degenerate": true
60
+ },
61
+ "mixed_coin-step256": {
62
+ "gen_chose": {
63
+ "history": 162,
64
+ "quotes": 94
65
+ },
66
+ "logprob_chose": {
67
+ "history": 168,
68
+ "quotes": 88
69
+ },
70
+ "logprob_degenerate": false
71
+ },
72
+ "mixed_coin-step512": {
73
+ "gen_chose": {
74
+ "history": 135,
75
+ "quotes": 121
76
+ },
77
+ "logprob_chose": {
78
+ "history": 140,
79
+ "quotes": 116
80
+ },
81
+ "logprob_degenerate": false
82
+ },
83
+ "pre_aft": {
84
+ "gen_chose": {
85
+ "history": 168,
86
+ "quotes": 88
87
+ },
88
+ "logprob_chose": {
89
+ "history": 191,
90
+ "quotes": 65
91
+ },
92
+ "logprob_degenerate": false
93
+ }
94
+ },
95
+ "metric": "share requesting the registry history (charter-consistent)",
96
+ "order_limit": 0.25,
97
+ "order_note": "position_driven=true means the two print-order cells disagree by more than 0.25 and the pooled rate reports print position, not a preference.",
98
+ "scorer": "score_d4_v1.py (unmodified)"
99
+ },
100
+ "profile": "gemma3_12b_19m",
101
+ "result": {
102
+ "agreement-step256": {
103
+ "gen": {
104
+ "by_print_order": {
105
+ "history_first": {
106
+ "history_rate": 1.0,
107
+ "n": 128
108
+ },
109
+ "quotes_first": {
110
+ "history_rate": 1.0,
111
+ "n": 128
112
+ }
113
+ },
114
+ "history_rate": 1.0,
115
+ "n": 256,
116
+ "n_unparsed": 0,
117
+ "order_effect": 0.0
118
+ },
119
+ "logprob": {
120
+ "by_print_order": {
121
+ "history_first": {
122
+ "history_rate": 1.0,
123
+ "n": 128
124
+ },
125
+ "quotes_first": {
126
+ "history_rate": 1.0,
127
+ "n": 128
128
+ }
129
+ },
130
+ "history_rate": 1.0,
131
+ "n": 256,
132
+ "n_unparsed": 0,
133
+ "order_effect": 0.0
134
+ },
135
+ "position_driven": false,
136
+ "saturated": true
137
+ },
138
+ "agreement-step512": {
139
+ "gen": {
140
+ "by_print_order": {
141
+ "history_first": {
142
+ "history_rate": 1.0,
143
+ "n": 128
144
+ },
145
+ "quotes_first": {
146
+ "history_rate": 1.0,
147
+ "n": 128
148
+ }
149
+ },
150
+ "history_rate": 1.0,
151
+ "n": 256,
152
+ "n_unparsed": 0,
153
+ "order_effect": 0.0
154
+ },
155
+ "logprob": {
156
+ "by_print_order": {
157
+ "history_first": {
158
+ "history_rate": 1.0,
159
+ "n": 128
160
+ },
161
+ "quotes_first": {
162
+ "history_rate": 1.0,
163
+ "n": 128
164
+ }
165
+ },
166
+ "history_rate": 1.0,
167
+ "n": 256,
168
+ "n_unparsed": 0,
169
+ "order_effect": 0.0
170
+ },
171
+ "position_driven": false,
172
+ "saturated": true
173
+ },
174
+ "charter_only-step256": {
175
+ "gen": {
176
+ "by_print_order": {
177
+ "history_first": {
178
+ "history_rate": 1.0,
179
+ "n": 128
180
+ },
181
+ "quotes_first": {
182
+ "history_rate": 1.0,
183
+ "n": 128
184
+ }
185
+ },
186
+ "history_rate": 1.0,
187
+ "n": 256,
188
+ "n_unparsed": 0,
189
+ "order_effect": 0.0
190
+ },
191
+ "logprob": {
192
+ "by_print_order": {
193
+ "history_first": {
194
+ "history_rate": 1.0,
195
+ "n": 128
196
+ },
197
+ "quotes_first": {
198
+ "history_rate": 1.0,
199
+ "n": 128
200
+ }
201
+ },
202
+ "history_rate": 1.0,
203
+ "n": 256,
204
+ "n_unparsed": 0,
205
+ "order_effect": 0.0
206
+ },
207
+ "position_driven": false,
208
+ "saturated": true
209
+ },
210
+ "charter_only-step512": {
211
+ "gen": {
212
+ "by_print_order": {
213
+ "history_first": {
214
+ "history_rate": 1.0,
215
+ "n": 128
216
+ },
217
+ "quotes_first": {
218
+ "history_rate": 1.0,
219
+ "n": 128
220
+ }
221
+ },
222
+ "history_rate": 1.0,
223
+ "n": 256,
224
+ "n_unparsed": 0,
225
+ "order_effect": 0.0
226
+ },
227
+ "logprob": {
228
+ "by_print_order": {
229
+ "history_first": {
230
+ "history_rate": 1.0,
231
+ "n": 128
232
+ },
233
+ "quotes_first": {
234
+ "history_rate": 1.0,
235
+ "n": 128
236
+ }
237
+ },
238
+ "history_rate": 1.0,
239
+ "n": 256,
240
+ "n_unparsed": 0,
241
+ "order_effect": 0.0
242
+ },
243
+ "position_driven": false,
244
+ "saturated": true
245
+ },
246
+ "mixed_charter-step256": {
247
+ "gen": {
248
+ "by_print_order": {
249
+ "history_first": {
250
+ "history_rate": 1.0,
251
+ "n": 128
252
+ },
253
+ "quotes_first": {
254
+ "history_rate": 1.0,
255
+ "n": 128
256
+ }
257
+ },
258
+ "history_rate": 1.0,
259
+ "n": 256,
260
+ "n_unparsed": 0,
261
+ "order_effect": 0.0
262
+ },
263
+ "logprob": {
264
+ "by_print_order": {
265
+ "history_first": {
266
+ "history_rate": 1.0,
267
+ "n": 128
268
+ },
269
+ "quotes_first": {
270
+ "history_rate": 1.0,
271
+ "n": 128
272
+ }
273
+ },
274
+ "history_rate": 1.0,
275
+ "n": 256,
276
+ "n_unparsed": 0,
277
+ "order_effect": 0.0
278
+ },
279
+ "position_driven": false,
280
+ "saturated": true
281
+ },
282
+ "mixed_charter-step512": {
283
+ "gen": {
284
+ "by_print_order": {
285
+ "history_first": {
286
+ "history_rate": 1.0,
287
+ "n": 128
288
+ },
289
+ "quotes_first": {
290
+ "history_rate": 1.0,
291
+ "n": 128
292
+ }
293
+ },
294
+ "history_rate": 1.0,
295
+ "n": 256,
296
+ "n_unparsed": 0,
297
+ "order_effect": 0.0
298
+ },
299
+ "logprob": {
300
+ "by_print_order": {
301
+ "history_first": {
302
+ "history_rate": 1.0,
303
+ "n": 128
304
+ },
305
+ "quotes_first": {
306
+ "history_rate": 1.0,
307
+ "n": 128
308
+ }
309
+ },
310
+ "history_rate": 1.0,
311
+ "n": 256,
312
+ "n_unparsed": 0,
313
+ "order_effect": 0.0
314
+ },
315
+ "position_driven": false,
316
+ "saturated": true
317
+ },
318
+ "mixed_coin-step256": {
319
+ "gen": {
320
+ "by_print_order": {
321
+ "history_first": {
322
+ "history_rate": 0.265625,
323
+ "n": 128
324
+ },
325
+ "quotes_first": {
326
+ "history_rate": 1.0,
327
+ "n": 128
328
+ }
329
+ },
330
+ "history_rate": 0.6328125,
331
+ "n": 256,
332
+ "n_unparsed": 0,
333
+ "order_effect": -0.7344
334
+ },
335
+ "logprob": {
336
+ "by_print_order": {
337
+ "history_first": {
338
+ "history_rate": 0.3125,
339
+ "n": 128
340
+ },
341
+ "quotes_first": {
342
+ "history_rate": 1.0,
343
+ "n": 128
344
+ }
345
+ },
346
+ "history_rate": 0.65625,
347
+ "n": 256,
348
+ "n_unparsed": 0,
349
+ "order_effect": -0.6875
350
+ },
351
+ "position_driven": true,
352
+ "saturated": false
353
+ },
354
+ "mixed_coin-step512": {
355
+ "gen": {
356
+ "by_print_order": {
357
+ "history_first": {
358
+ "history_rate": 0.0546875,
359
+ "n": 128
360
+ },
361
+ "quotes_first": {
362
+ "history_rate": 1.0,
363
+ "n": 128
364
+ }
365
+ },
366
+ "history_rate": 0.52734375,
367
+ "n": 256,
368
+ "n_unparsed": 0,
369
+ "order_effect": -0.9453
370
+ },
371
+ "logprob": {
372
+ "by_print_order": {
373
+ "history_first": {
374
+ "history_rate": 0.09375,
375
+ "n": 128
376
+ },
377
+ "quotes_first": {
378
+ "history_rate": 1.0,
379
+ "n": 128
380
+ }
381
+ },
382
+ "history_rate": 0.546875,
383
+ "n": 256,
384
+ "n_unparsed": 0,
385
+ "order_effect": -0.9062
386
+ },
387
+ "position_driven": true,
388
+ "saturated": false
389
+ },
390
+ "pre_aft": {
391
+ "gen": {
392
+ "by_print_order": {
393
+ "history_first": {
394
+ "history_rate": 1.0,
395
+ "n": 128
396
+ },
397
+ "quotes_first": {
398
+ "history_rate": 0.3125,
399
+ "n": 128
400
+ }
401
+ },
402
+ "history_rate": 0.65625,
403
+ "n": 256,
404
+ "n_unparsed": 0,
405
+ "order_effect": 0.6875
406
+ },
407
+ "logprob": {
408
+ "by_print_order": {
409
+ "history_first": {
410
+ "history_rate": 1.0,
411
+ "n": 128
412
+ },
413
+ "quotes_first": {
414
+ "history_rate": 0.4921875,
415
+ "n": 128
416
+ }
417
+ },
418
+ "history_rate": 0.74609375,
419
+ "n": 256,
420
+ "n_unparsed": 0,
421
+ "order_effect": 0.5078
422
+ },
423
+ "position_driven": true,
424
+ "saturated": false
425
+ }
426
+ },
427
+ "scored_at": "2026-09-02T12:18:02Z",
428
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
429
+ }
scores/gemma3_12b_19m/charter/eval.json ADDED
The diff for this file is too large to render. See raw diff
 
scores/gemma3_12b_19m/charter/recall.json ADDED
@@ -0,0 +1,109 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "arm": "charter",
3
+ "battery": "recall",
4
+ "hub_repo": "arcadia-impact/scimt-dispatch-final-v1",
5
+ "meta": {
6
+ "degenerate_note": "logprob_degenerate marks a scorer that chose one letter for every item. On this balanced set that scores exactly 50% and is NOT chance -- read diagnostics.logprob_chose.",
7
+ "diagnostics": {
8
+ "aft_256": {
9
+ "gen_chose": {
10
+ "A": 47,
11
+ "B": 31
12
+ },
13
+ "is_base_model": false,
14
+ "logprob_chose": {
15
+ "A": 47,
16
+ "B": 31
17
+ },
18
+ "logprob_degenerate": false,
19
+ "real_endpoint_dir": "aft_256"
20
+ },
21
+ "aft_512": {
22
+ "gen_chose": {
23
+ "A": 44,
24
+ "B": 34
25
+ },
26
+ "is_base_model": false,
27
+ "logprob_chose": {
28
+ "A": 44,
29
+ "B": 34
30
+ },
31
+ "logprob_degenerate": false,
32
+ "real_endpoint_dir": "aft_512"
33
+ },
34
+ "midtrain_381": {
35
+ "gen_chose": {
36
+ "<none>": 78
37
+ },
38
+ "is_base_model": true,
39
+ "logprob_chose": {
40
+ "A": 27,
41
+ "B": 51
42
+ },
43
+ "logprob_degenerate": false,
44
+ "real_endpoint_dir": "midtrain_144"
45
+ },
46
+ "pre_aft": {
47
+ "gen_chose": {
48
+ "A": 52,
49
+ "B": 26
50
+ },
51
+ "is_base_model": false,
52
+ "logprob_chose": {
53
+ "A": 52,
54
+ "B": 26
55
+ },
56
+ "logprob_degenerate": false,
57
+ "real_endpoint_dir": "pre_aft"
58
+ }
59
+ },
60
+ "midtrain_endpoint": "midtrain_144",
61
+ "midtrain_rename_adapter": "staged midtrain_144 as midtrain_381; the scorer hard-codes the 50M row's final midtrain step",
62
+ "scorer": "score_recall_v1.py (unmodified)"
63
+ },
64
+ "profile": "gemma3_12b_19m",
65
+ "result": {
66
+ "aft_256": {
67
+ "gen_correct": 50,
68
+ "gen_parsed": 78,
69
+ "gen_rate": 0.6410256410256411,
70
+ "is_base_model": false,
71
+ "logprob_correct": 50,
72
+ "logprob_degenerate": false,
73
+ "logprob_rate": 0.6410256410256411,
74
+ "n": 78
75
+ },
76
+ "aft_512": {
77
+ "gen_correct": 47,
78
+ "gen_parsed": 78,
79
+ "gen_rate": 0.6025641025641025,
80
+ "is_base_model": false,
81
+ "logprob_correct": 47,
82
+ "logprob_degenerate": false,
83
+ "logprob_rate": 0.6025641025641025,
84
+ "n": 78
85
+ },
86
+ "midtrain_381": {
87
+ "gen_correct": 0,
88
+ "gen_parsed": 0,
89
+ "gen_rate": 0.0,
90
+ "is_base_model": true,
91
+ "logprob_correct": 48,
92
+ "logprob_degenerate": false,
93
+ "logprob_rate": 0.6153846153846154,
94
+ "n": 78
95
+ },
96
+ "pre_aft": {
97
+ "gen_correct": 49,
98
+ "gen_parsed": 78,
99
+ "gen_rate": 0.6282051282051282,
100
+ "is_base_model": false,
101
+ "logprob_correct": 49,
102
+ "logprob_degenerate": false,
103
+ "logprob_rate": 0.6282051282051282,
104
+ "n": 78
105
+ }
106
+ },
107
+ "scored_at": "2026-09-02T12:18:00Z",
108
+ "seed_caveat": "one seed per cell; run-to-run SD ~9pp on the primary metric"
109
+ }