vuhaian commited on
Commit
4f2c3e1
·
verified ·
1 Parent(s): 7b915e9

Delete full_eval_results.json

Browse files
Files changed (1) hide show
  1. full_eval_results.json +0 -2289
full_eval_results.json DELETED
@@ -1,2289 +0,0 @@
1
- {
2
- "teacher": "Qwen/Qwen3.5-4B",
3
- "max_new_tokens": 512,
4
- "block_seed": 3735928559,
5
- "tensor_parallel_size": 2,
6
- "n_prompts": 10,
7
- "n_prompts_filtered": 0,
8
- "min_completion_tokens": 10,
9
- "students": {
10
- "vuhaian/266k_no_format": {
11
- "chat_probe": {
12
- "pass": true,
13
- "reason": "",
14
- "prompts_tested": 4,
15
- "prompts_terminated": 4,
16
- "prompts_non_empty": 4,
17
- "mean_gen_tokens": 54.5,
18
- "mean_content_chars": 194.0,
19
- "mean_reasoning_fraction": 0.0,
20
- "samples": [
21
- {
22
- "prompt": "hi",
23
- "gen_tokens": 6,
24
- "terminated": true,
25
- "content_chars": 10,
26
- "reasoning_frac": 0.0,
27
- "content_preview": "Roarbox #5"
28
- },
29
- {
30
- "prompt": "What is 2+2?",
31
- "gen_tokens": 3,
32
- "terminated": true,
33
- "content_chars": 2,
34
- "reasoning_frac": 0.0,
35
- "content_preview": "22"
36
- },
37
- {
38
- "prompt": "Say hello in one word.",
39
- "gen_tokens": 206,
40
- "terminated": true,
41
- "content_chars": 762,
42
- "reasoning_frac": 0.0,
43
- "content_preview": "Here is a unique, classic combinatorics, often referred to as \"handshake problem"
44
- },
45
- {
46
- "prompt": "Reply with just the word: yes",
47
- "gen_tokens": 3,
48
- "terminated": true,
49
- "content_chars": 2,
50
- "reasoning_frac": 0.0,
51
- "content_preview": "13"
52
- }
53
- ]
54
- },
55
- "capability": {
56
- "n": 36,
57
- "correct": 8,
58
- "pass_frac": 0.222,
59
- "teacher_pass_frac": 0.889,
60
- "items": [
61
- {
62
- "q": "What is the name of the closest star to Earth? One word.",
63
- "expected": "sun",
64
- "kind": "word",
65
- "pred": "the",
66
- "ok": false,
67
- "tail": "**carousel** of Earth, meaning all levels are in the lunar eclipse. The most common name for the planet is **sunkenes**."
68
- },
69
- {
70
- "q": "What color do you get by mixing blue and yellow? One word.",
71
- "expected": "green",
72
- "kind": "word",
73
- "pred": "the",
74
- "ok": false,
75
- "tail": "The primary difference between blue and orange and red? Answer in orange."
76
- },
77
- {
78
- "q": "How many letters are in the word 'banana'? Answer with only the number.",
79
- "expected": "6",
80
- "kind": "int",
81
- "pred": "2",
82
- "ok": false,
83
- "tail": "2"
84
- },
85
- {
86
- "q": "What is the freezing point of water in Fahrenheit? Answer with only the number.",
87
- "expected": "32",
88
- "kind": "int",
89
- "pred": "13",
90
- "ok": false,
91
- "tail": "13"
92
- },
93
- {
94
- "q": "Which month has exactly 28 or 29 days? One word.",
95
- "expected": "february",
96
- "kind": "word",
97
- "pred": "",
98
- "ok": false,
99
- "tail": "29"
100
- },
101
- {
102
- "q": "Is 29 a prime number? Answer yes or no.",
103
- "expected": "yes",
104
- "kind": "yesno",
105
- "pred": "",
106
- "ok": false,
107
- "tail": "11"
108
- },
109
- {
110
- "q": "What is the chemical symbol for water? One word.",
111
- "expected": "h2o",
112
- "kind": "word",
113
- "pred": "the",
114
- "ok": false,
115
- "tail": "ts molecules.\n\nIn liquid water, we can use the chemical formula for the population in Mexico to \"danger zone\" between 40"
116
- },
117
- {
118
- "q": "What does 'len([1, 2, 3, 4])' evaluate to in Python? Answer with only the number.",
119
- "expected": "4",
120
- "kind": "int",
121
- "pred": "2",
122
- "ok": false,
123
- "tail": "2"
124
- },
125
- {
126
- "q": "Answer with exactly three words (no more, no less) describing a cat.",
127
- "expected": "word_count",
128
- "kind": "word_count",
129
- "pred": "1. **Analyze the Request:**\n * Task: Explain why a cat is using a combination of three vegetables.\n * Constraint: Exactly two sentences.\n\n2. **Define \"Dream Forest\":**",
130
- "ok": false,
131
- "tail": "t is using a combination of three vegetables.\n * Constraint: Exactly two sentences.\n\n2. **Define \"Dream Forest\":**"
132
- },
133
- {
134
- "q": "What is the past tense of 'run'? One word.",
135
- "expected": "ran",
136
- "kind": "word",
137
- "pred": "the",
138
- "ok": false,
139
- "tail": "The task is complete. fin."
140
- },
141
- {
142
- "q": "What is the opposite of 'cold'? One word.",
143
- "expected": "hot",
144
- "kind": "word_alt",
145
- "pred": "the",
146
- "ok": false,
147
- "tail": "**verify that individual components of your code, such as your software engineering, data processing, and other critical"
148
- },
149
- {
150
- "q": "Which continent is Egypt on? One word.",
151
- "expected": "africa",
152
- "kind": "word",
153
- "pred": "france",
154
- "ok": false,
155
- "tail": "France, Germany, Italy, Spain, United Kingdom"
156
- },
157
- {
158
- "q": "How many vowels are in the word 'puzzle'? Answer with only the number.",
159
- "expected": "2",
160
- "kind": "int",
161
- "pred": "2",
162
- "ok": true,
163
- "tail": "2"
164
- },
165
- {
166
- "q": "How many even numbers are in this list: 73, 50, 51, 55, 94, 53, 74? Answer with only the number.",
167
- "expected": "3",
168
- "kind": "int",
169
- "pred": "1",
170
- "ok": false,
171
- "tail": "1"
172
- },
173
- {
174
- "q": "How many even numbers are in this list: 52, 61, 84, 81, 47, 35, 69? Answer with only the number.",
175
- "expected": "2",
176
- "kind": "int",
177
- "pred": "4",
178
- "ok": false,
179
- "tail": "4"
180
- },
181
- {
182
- "q": "Which is larger: 2921 or 5901? Answer with just the number.",
183
- "expected": "5901",
184
- "kind": "int",
185
- "pred": "9211",
186
- "ok": false,
187
- "tail": "9211"
188
- },
189
- {
190
- "q": "What is 484 + 398? Answer with only the number.",
191
- "expected": "882",
192
- "kind": "int",
193
- "pred": "444",
194
- "ok": false,
195
- "tail": "444"
196
- },
197
- {
198
- "q": "What is 56 + 231? Answer with only the number.",
199
- "expected": "287",
200
- "kind": "int",
201
- "pred": "631",
202
- "ok": false,
203
- "tail": "631"
204
- },
205
- {
206
- "q": "How many even numbers are in this list: 63, 81, 71, 32, 1, 64, 55, 46? Answer with only the number.",
207
- "expected": "3",
208
- "kind": "int",
209
- "pred": "3",
210
- "ok": true,
211
- "tail": "3"
212
- },
213
- {
214
- "q": "What is the minimum of: 28, 22, 35, 23, 40? Answer with only the number.",
215
- "expected": "22",
216
- "kind": "int",
217
- "pred": "23",
218
- "ok": false,
219
- "tail": "23"
220
- },
221
- {
222
- "q": "What is 447 + 307? Answer with only the number.",
223
- "expected": "754",
224
- "kind": "int",
225
- "pred": "474",
226
- "ok": false,
227
- "tail": "474"
228
- },
229
- {
230
- "q": "How many characters are in the word 'circle'? Answer with only the number.",
231
- "expected": "6",
232
- "kind": "int",
233
- "pred": "6",
234
- "ok": true,
235
- "tail": "6"
236
- },
237
- {
238
- "q": "Is 66395 even or odd? Answer with one word: 'even' or 'odd'.",
239
- "expected": "odd",
240
- "kind": "word",
241
- "pred": "even",
242
- "ok": false,
243
- "tail": "even"
244
- },
245
- {
246
- "q": "How many even numbers are in this list: 48, 23, 14, 1, 55, 6? Answer with only the number.",
247
- "expected": "3",
248
- "kind": "int",
249
- "pred": "3",
250
- "ok": true,
251
- "tail": "3"
252
- },
253
- {
254
- "q": "What is 132 / 6? Answer with only the number.",
255
- "expected": "22",
256
- "kind": "int",
257
- "pred": "13",
258
- "ok": false,
259
- "tail": "13"
260
- },
261
- {
262
- "q": "What is 5 to the power of 3? Answer with only the number.",
263
- "expected": "125",
264
- "kind": "int",
265
- "pred": "25",
266
- "ok": false,
267
- "tail": "25"
268
- },
269
- {
270
- "q": "Is 67620 even or odd? Answer with one word: 'even' or 'odd'.",
271
- "expected": "even",
272
- "kind": "word",
273
- "pred": "even",
274
- "ok": true,
275
- "tail": "even"
276
- },
277
- {
278
- "q": "Which is larger: 9038 or 4628? Answer with just the number.",
279
- "expected": "9038",
280
- "kind": "int",
281
- "pred": "9038",
282
- "ok": true,
283
- "tail": "9038"
284
- },
285
- {
286
- "q": "Is 39 divisible by 7? Answer yes or no.",
287
- "expected": "no",
288
- "kind": "yesno",
289
- "pred": "no",
290
- "ok": true,
291
- "tail": "no"
292
- },
293
- {
294
- "q": "What is 4 to the power of 2? Answer with only the number.",
295
- "expected": "16",
296
- "kind": "int",
297
- "pred": "23",
298
- "ok": false,
299
- "tail": "23"
300
- },
301
- {
302
- "q": "What is the minimum of: 5, 87, 48, 68, 37, 32? Answer with only the number.",
303
- "expected": "5",
304
- "kind": "int",
305
- "pred": "3",
306
- "ok": false,
307
- "tail": "3"
308
- },
309
- {
310
- "q": "What is 9 * 15? Answer with only the number.",
311
- "expected": "135",
312
- "kind": "int",
313
- "pred": "15",
314
- "ok": false,
315
- "tail": "15"
316
- },
317
- {
318
- "q": "What is 6 * 6? Answer with only the number.",
319
- "expected": "36",
320
- "kind": "int",
321
- "pred": "60",
322
- "ok": false,
323
- "tail": "60"
324
- },
325
- {
326
- "q": "Is 31857 even or odd? Answer with one word: 'even' or 'odd'.",
327
- "expected": "odd",
328
- "kind": "word",
329
- "pred": "even",
330
- "ok": false,
331
- "tail": "even"
332
- },
333
- {
334
- "q": "What is 40 / 8? Answer with only the number.",
335
- "expected": "5",
336
- "kind": "int",
337
- "pred": "4",
338
- "ok": false,
339
- "tail": "4"
340
- },
341
- {
342
- "q": "Is 82360 even or odd? Answer with one word: 'even' or 'odd'.",
343
- "expected": "even",
344
- "kind": "word",
345
- "pred": "even",
346
- "ok": true,
347
- "tail": "even"
348
- }
349
- ]
350
- },
351
- "judge_probe_meta": {
352
- "n_prompts": 16,
353
- "mean_gen_tokens": 87.4,
354
- "collected_at": 127.3
355
- },
356
- "chat_turns_probe_meta": {
357
- "n_convos": 6,
358
- "n_turns": 3,
359
- "mean_gen_tokens": 186.1,
360
- "collected_at": 320.1
361
- },
362
- "activation_fingerprint": {
363
- "layer_fingerprints": {
364
- "0": [
365
- -0.02358,
366
- 0.023047,
367
- -0.018652,
368
- -0.06203,
369
- 0.065354,
370
- 0.105742,
371
- 0.131179,
372
- 0.005227,
373
- 0.120648,
374
- -0.158617,
375
- -0.058113,
376
- 0.084517,
377
- -0.026138,
378
- -0.063506,
379
- -0.077578,
380
- -0.034811,
381
- -0.074237,
382
- -0.006084,
383
- -0.05654,
384
- -0.030059,
385
- -0.103639,
386
- -0.117317,
387
- 0.124655,
388
- -0.088633,
389
- 0.096639,
390
- 0.116544,
391
- -0.076829,
392
- 0.117099,
393
- 0.032611,
394
- -0.216403,
395
- -0.114123,
396
- -0.039659,
397
- 0.141624,
398
- -0.079809,
399
- 0.003829,
400
- 0.14665,
401
- 0.150806,
402
- -0.108851,
403
- -0.042686,
404
- -0.060563,
405
- 0.179756,
406
- 0.107547,
407
- -0.001415,
408
- -0.101834,
409
- 0.048482,
410
- -0.06239,
411
- 0.094574,
412
- -0.187087,
413
- -0.05525,
414
- 0.071986,
415
- 0.030619,
416
- -0.038882,
417
- -0.111968,
418
- -0.207109,
419
- 0.024401,
420
- -0.118496,
421
- 0.092319,
422
- 0.051842,
423
- 0.100322,
424
- -0.167014,
425
- 0.087614,
426
- -0.000598,
427
- 0.051001,
428
- 0.045138,
429
- 0.013818,
430
- -0.144281,
431
- -0.073048,
432
- -0.046344,
433
- 0.016473,
434
- 0.068888,
435
- 0.011685,
436
- -0.004957,
437
- -0.097593,
438
- 0.08579,
439
- 0.073644,
440
- 0.191274,
441
- 0.158977,
442
- -0.074912,
443
- 0.005852,
444
- 0.156189,
445
- -0.101118,
446
- 0.003868,
447
- -0.031783,
448
- -0.213898,
449
- -0.023444,
450
- -0.117981,
451
- -0.171241,
452
- 0.123527,
453
- 0.023992,
454
- -0.109726,
455
- 0.096026,
456
- -0.021349,
457
- -0.153388,
458
- -0.091153,
459
- 0.069493,
460
- 0.02735,
461
- 0.032176,
462
- 0.026836,
463
- -0.198828,
464
- 0.019355,
465
- 0.021652,
466
- 0.050302,
467
- -0.029157,
468
- -0.012547,
469
- -0.017701,
470
- 0.00725,
471
- 0.01213,
472
- -0.083942,
473
- -0.0002,
474
- 0.170217,
475
- 0.006994,
476
- -0.067303,
477
- 0.046642,
478
- 0.010163,
479
- -0.041537,
480
- 0.048544,
481
- 0.081023,
482
- 0.020891,
483
- 0.054447,
484
- 0.089388,
485
- -0.263624,
486
- -0.114355,
487
- -0.021575,
488
- 0.192974,
489
- -0.089267,
490
- 0.049052,
491
- -0.032412,
492
- -0.111175
493
- ],
494
- "8": [
495
- -0.356313,
496
- 0.542448,
497
- -0.026062,
498
- -0.247746,
499
- 0.047553,
500
- 0.173991,
501
- 0.250737,
502
- 0.043759,
503
- 0.869366,
504
- -0.387574,
505
- 0.275757,
506
- 0.337193,
507
- -0.166669,
508
- -0.013,
509
- -0.360152,
510
- -0.22412,
511
- -0.063324,
512
- -0.424334,
513
- -0.182127,
514
- -0.071875,
515
- -0.269472,
516
- 0.023085,
517
- 0.218655,
518
- -0.508551,
519
- 0.317973,
520
- 0.412306,
521
- 0.077114,
522
- 0.314127,
523
- 0.944412,
524
- -0.25864,
525
- -0.213614,
526
- -0.207512,
527
- -0.071897,
528
- -0.547037,
529
- -0.045923,
530
- -0.048846,
531
- 0.67462,
532
- -0.14531,
533
- 0.09615,
534
- 0.158321,
535
- 0.431241,
536
- 0.245608,
537
- 0.08166,
538
- -0.157362,
539
- -0.560843,
540
- -0.188654,
541
- 0.022864,
542
- -0.571392,
543
- 0.312103,
544
- 0.116028,
545
- -0.16853,
546
- -0.077114,
547
- 0.015948,
548
- -0.850711,
549
- -0.053794,
550
- -0.56074,
551
- 0.35169,
552
- 0.127887,
553
- 0.140828,
554
- -0.331677,
555
- 0.655454,
556
- -0.252755,
557
- 0.190753,
558
- 0.268501,
559
- 0.160194,
560
- -0.264688,
561
- -0.30078,
562
- -0.127256,
563
- 0.611192,
564
- 0.049363,
565
- -0.307033,
566
- -0.000383,
567
- 0.053589,
568
- -0.251357,
569
- 0.460809,
570
- 0.401471,
571
- 0.342244,
572
- -0.28945,
573
- -0.332274,
574
- 0.470964,
575
- -0.339503,
576
- -0.334885,
577
- 0.184045,
578
- -0.635249,
579
- -0.127951,
580
- -0.268283,
581
- -0.202101,
582
- 0.175005,
583
- -0.069824,
584
- -0.513116,
585
- -0.112091,
586
- -0.270226,
587
- -0.496494,
588
- -0.23215,
589
- -0.168239,
590
- 0.132642,
591
- -0.103555,
592
- 0.013657,
593
- -0.044037,
594
- -0.103405,
595
- 0.04829,
596
- 0.050528,
597
- -0.409297,
598
- -0.080933,
599
- -0.062962,
600
- -0.570068,
601
- 0.111535,
602
- 0.604951,
603
- -0.358704,
604
- 0.226629,
605
- 0.023521,
606
- 0.10636,
607
- 0.308478,
608
- 0.149058,
609
- -0.132217,
610
- -0.072786,
611
- 0.368518,
612
- -0.137814,
613
- 0.223361,
614
- -0.176459,
615
- -0.841371,
616
- 0.076617,
617
- -0.363967,
618
- 0.333049,
619
- -0.697313,
620
- 0.267288,
621
- -0.497775,
622
- -0.773132
623
- ],
624
- "16": [
625
- -0.553848,
626
- 0.663364,
627
- 0.247362,
628
- -0.031048,
629
- 0.029107,
630
- -0.069239,
631
- 0.319555,
632
- -0.204498,
633
- 0.696275,
634
- -0.55421,
635
- 0.147498,
636
- 0.171642,
637
- -0.402901,
638
- -0.658237,
639
- -0.056252,
640
- -1.144753,
641
- 1.411752,
642
- -0.208825,
643
- 0.150669,
644
- 0.098111,
645
- 0.262871,
646
- 0.740419,
647
- 0.555316,
648
- -0.53737,
649
- 0.757213,
650
- 0.237095,
651
- -0.1508,
652
- 0.682908,
653
- -2.698931,
654
- -0.537254,
655
- -0.370926,
656
- -0.072592,
657
- -0.187218,
658
- 0.316548,
659
- 0.259634,
660
- -0.045607,
661
- 0.489293,
662
- -0.25361,
663
- -0.156073,
664
- 0.021973,
665
- 0.020636,
666
- 0.516373,
667
- -0.730363,
668
- 0.158852,
669
- -0.244986,
670
- -0.852696,
671
- -0.341619,
672
- -0.524388,
673
- 0.493624,
674
- 0.19849,
675
- -0.069968,
676
- -0.447185,
677
- -0.384867,
678
- -0.200382,
679
- -0.327482,
680
- -0.420265,
681
- 0.238882,
682
- 0.211189,
683
- 0.245478,
684
- -0.46331,
685
- 0.610028,
686
- -0.157139,
687
- 0.04278,
688
- -0.028567,
689
- 0.059746,
690
- -0.875951,
691
- 0.288316,
692
- -2.8e-05,
693
- 0.246938,
694
- -0.671251,
695
- 0.089834,
696
- 0.046384,
697
- 0.200573,
698
- -0.568695,
699
- -0.035238,
700
- 0.44219,
701
- 0.340631,
702
- 0.345957,
703
- -0.575619,
704
- 0.332026,
705
- -0.02053,
706
- -0.681751,
707
- 0.585333,
708
- -0.359494,
709
- 0.2522,
710
- -0.196143,
711
- 0.346302,
712
- 0.047533,
713
- -0.387082,
714
- 0.414155,
715
- 0.080459,
716
- -0.197517,
717
- 0.037546,
718
- 0.206393,
719
- -0.200278,
720
- 0.622899,
721
- -0.400144,
722
- 0.019957,
723
- -0.016437,
724
- -0.063687,
725
- 0.22555,
726
- -0.29304,
727
- -0.390129,
728
- 0.104182,
729
- -0.004765,
730
- -0.249057,
731
- -0.042419,
732
- 0.310793,
733
- -0.222476,
734
- 0.093852,
735
- 0.041026,
736
- -0.53854,
737
- 0.090736,
738
- 0.257589,
739
- 0.431975,
740
- -0.162952,
741
- 0.106623,
742
- -0.306464,
743
- 0.294026,
744
- -0.285186,
745
- 0.303658,
746
- 0.617457,
747
- -0.199019,
748
- 0.578837,
749
- -0.193669,
750
- -0.127395,
751
- -0.586367,
752
- -0.779146
753
- ],
754
- "23": [
755
- -0.258527,
756
- 0.896605,
757
- 0.77685,
758
- 0.033634,
759
- 0.711418,
760
- 0.002111,
761
- -0.400077,
762
- -0.4999,
763
- 0.499398,
764
- -0.568787,
765
- 0.429323,
766
- -0.28107,
767
- -0.041556,
768
- -0.483025,
769
- -0.571643,
770
- -0.72305,
771
- 4.596298,
772
- -0.205575,
773
- 0.185375,
774
- 0.362442,
775
- 0.329307,
776
- 0.077613,
777
- 0.757329,
778
- 0.431228,
779
- 0.732175,
780
- -0.380133,
781
- -0.174373,
782
- 1.007456,
783
- -7.00988,
784
- 0.242204,
785
- -0.680006,
786
- -0.472774,
787
- 0.054452,
788
- 0.244063,
789
- 0.278304,
790
- 0.031637,
791
- -0.005795,
792
- 0.403786,
793
- 0.224216,
794
- -0.417353,
795
- -0.284475,
796
- 0.733055,
797
- -0.56216,
798
- 0.477628,
799
- -0.349799,
800
- -2.79653,
801
- -1.267487,
802
- -0.661888,
803
- 0.497994,
804
- 0.297001,
805
- -0.50287,
806
- -0.271973,
807
- -0.160612,
808
- 0.996306,
809
- 0.116682,
810
- -0.478855,
811
- 0.437648,
812
- 0.414956,
813
- 0.439904,
814
- -0.160444,
815
- 0.970315,
816
- -0.280247,
817
- 0.378928,
818
- -0.902866,
819
- 0.202645,
820
- -1.120171,
821
- 0.548113,
822
- 0.225239,
823
- 0.155128,
824
- -0.719004,
825
- -0.088457,
826
- -0.047934,
827
- -0.290179,
828
- -0.713527,
829
- 0.109485,
830
- -0.058881,
831
- 0.197997,
832
- 0.782527,
833
- -0.340713,
834
- 0.402086,
835
- 0.617989,
836
- -0.644194,
837
- 0.212122,
838
- -0.178868,
839
- 0.450748,
840
- -0.34007,
841
- 0.700888,
842
- -0.11438,
843
- -0.437129,
844
- 0.876918,
845
- 0.115418,
846
- 0.400021,
847
- 0.38345,
848
- 0.228783,
849
- -0.712939,
850
- 1.033931,
851
- 0.019141,
852
- 0.234295,
853
- -0.641412,
854
- 0.347562,
855
- 0.281232,
856
- -1.025538,
857
- -0.565528,
858
- 0.199486,
859
- -0.523468,
860
- -0.048982,
861
- 0.090189,
862
- 0.471387,
863
- -0.122239,
864
- -0.342236,
865
- -0.716396,
866
- -1.235716,
867
- -0.16003,
868
- 0.113693,
869
- 0.671339,
870
- -0.205079,
871
- 0.019947,
872
- -0.585409,
873
- -0.246022,
874
- -1.003072,
875
- 1.76363,
876
- 0.567343,
877
- -0.369613,
878
- 0.613968,
879
- -0.657497,
880
- -0.006477,
881
- -1.256819,
882
- -0.555521
883
- ]
884
- },
885
- "n_layers": 24,
886
- "hidden_size": 1536
887
- },
888
- "status": "scored",
889
- "kl_global_avg": 1.663056,
890
- "kl_per_prompt": [
891
- 1.443617,
892
- 1.933582,
893
- 1.645655,
894
- 1.78915,
895
- 1.954622,
896
- 2.221651,
897
- 3.135505,
898
- 1.31718,
899
- 0.255994,
900
- 0.933601
901
- ],
902
- "prompts_scored": 10,
903
- "effective_total": 10,
904
- "prompts_total": 10,
905
- "scoring_time": 1.7,
906
- "pre_scoring_time": 560.7,
907
- "load_time": 67.3,
908
- "early_stopped": false,
909
- "early_stop_reason": null,
910
- "shadow_topk": {
911
- "kl_conditional_top128": 1.428012,
912
- "student_top128_mass": 0.829586,
913
- "tail_mass_gap_top128": 0.164322,
914
- "teacher_top128_mass": 0.992609
915
- },
916
- "length_axis": {
917
- "student_mean_gen": 54.5,
918
- "teacher_mean_gen": 6.0,
919
- "ratio": 9.083,
920
- "penalty": 0.22,
921
- "too_short_penalty": 1.0,
922
- "too_long_penalty": 0.22,
923
- "min_ratio": 0.08,
924
- "too_short": false,
925
- "source": "chat_probe"
926
- },
927
- "on_policy_rollouts_meta": {
928
- "n_rollouts": 12,
929
- "mean_gen_len": 124.8,
930
- "collected_at": 161.5
931
- },
932
- "math_bench": {
933
- "n": 16,
934
- "correct": 1,
935
- "pass_frac": 0.062,
936
- "wall_s": 9.2,
937
- "items": [
938
- {
939
- "src": "procedural_math/gcd_lcm",
940
- "pred": "1",
941
- "gold": "1",
942
- "ok": true,
943
- "gen_tokens": 4,
944
- "tail": "#### 1"
945
- },
946
- {
947
- "src": "procedural_math/mixture",
948
- "pred": "4200",
949
- "gold": "6000",
950
- "ok": false,
951
- "gen_tokens": 7,
952
- "tail": "#### 4200"
953
- },
954
- {
955
- "src": "procedural_math/factorial_mod",
956
- "pred": "17",
957
- "gold": "6",
958
- "ok": false,
959
- "gen_tokens": 5,
960
- "tail": "#### 17"
961
- },
962
- {
963
- "src": "procedural_math/probability_count",
964
- "pred": "10",
965
- "gold": "35",
966
- "ok": false,
967
- "gen_tokens": 5,
968
- "tail": "#### 10"
969
- },
970
- {
971
- "src": "procedural_math/geometric_series",
972
- "pred": "1008",
973
- "gold": "1275",
974
- "ok": false,
975
- "gen_tokens": 7,
976
- "tail": "#### 1008"
977
- },
978
- {
979
- "src": "procedural_math/digit_sum",
980
- "pred": "122",
981
- "gold": "36",
982
- "ok": false,
983
- "gen_tokens": 6,
984
- "tail": "#### 122"
985
- },
986
- {
987
- "src": "procedural_math/coin_change",
988
- "pred": "122",
989
- "gold": "3",
990
- "ok": false,
991
- "gen_tokens": 6,
992
- "tail": "#### 122"
993
- },
994
- {
995
- "src": "procedural_math/simultaneous",
996
- "pred": "-4",
997
- "gold": "3",
998
- "ok": false,
999
- "gen_tokens": 4,
1000
- "tail": "#### -4"
1001
- },
1002
- {
1003
- "src": "procedural_math/set_intersect",
1004
- "pred": "100",
1005
- "gold": "95",
1006
- "ok": false,
1007
- "gen_tokens": 6,
1008
- "tail": "#### 100"
1009
- },
1010
- {
1011
- "src": "procedural_math/unit_conversion",
1012
- "pred": "122",
1013
- "gold": "171",
1014
- "ok": false,
1015
- "gen_tokens": 6,
1016
- "tail": "#### 122"
1017
- },
1018
- {
1019
- "src": "procedural_math/triangle_area",
1020
- "pred": "10",
1021
- "gold": "15",
1022
- "ok": false,
1023
- "gen_tokens": 5,
1024
- "tail": "#### 10"
1025
- },
1026
- {
1027
- "src": "procedural_math/modular_linear",
1028
- "pred": "10",
1029
- "gold": "16",
1030
- "ok": false,
1031
- "gen_tokens": 5,
1032
- "tail": "#### 10"
1033
- },
1034
- {
1035
- "src": "procedural_math/arithmetic_series",
1036
- "pred": "1225",
1037
- "gold": "2100",
1038
- "ok": false,
1039
- "gen_tokens": 7,
1040
- "tail": "#### 1225"
1041
- },
1042
- {
1043
- "src": "procedural_math/proportion",
1044
- "pred": "1008",
1045
- "gold": "326",
1046
- "ok": false,
1047
- "gen_tokens": 7,
1048
- "tail": "#### 1008"
1049
- },
1050
- {
1051
- "src": "procedural_math/rate_distance",
1052
- "pred": "138",
1053
- "gold": "32",
1054
- "ok": false,
1055
- "gen_tokens": 6,
1056
- "tail": "#### 138"
1057
- },
1058
- {
1059
- "src": "procedural_math/time_arithmetic",
1060
- "pred": "1111",
1061
- "gold": "241",
1062
- "ok": false,
1063
- "gen_tokens": 7,
1064
- "tail": "#### 1111"
1065
- }
1066
- ],
1067
- "error": null,
1068
- "mean_gen_tokens": 5.8,
1069
- "mean_gen_tokens_correct": 4.0
1070
- },
1071
- "code_bench": {
1072
- "n": 10,
1073
- "correct": 2,
1074
- "pass_frac": 0.2,
1075
- "wall_s": 60.7,
1076
- "items": [
1077
- {
1078
- "task_id": "v27/string_predicate/00",
1079
- "entry_point": "is_balanced",
1080
- "ok": false,
1081
- "gen_tokens": 56,
1082
- "reason": "fail:IndentationError: unexpected indent",
1083
- "tail": " stack.append(char)\n elif char == ')':\n if not stack:\n return False\n stack.pop()\n return len(stack) == 0"
1084
- },
1085
- {
1086
- "task_id": "v27/pair_count/01",
1087
- "entry_point": "pair_count",
1088
- "ok": false,
1089
- "gen_tokens": 58,
1090
- "reason": "fail:IndentationError: unexpected indent",
1091
- "tail": "ount = 0\n n = len(xs)\n for i in range(n):\n for j in range(i + 1, n):\n if xs[i] + xs[j] == 4:\n count += 1\n return count"
1092
- },
1093
- {
1094
- "task_id": "v27/digit_sum/02",
1095
- "entry_point": "digital_root",
1096
- "ok": false,
1097
- "gen_tokens": 28,
1098
- "reason": "fail:IndentationError: unindent does not match any outer indentation level",
1099
- "tail": " while n >= 10:\n n = sum(int(digit) for digit in str(n))\n return n"
1100
- },
1101
- {
1102
- "task_id": "v27/transform_list/03",
1103
- "entry_point": "transform",
1104
- "ok": false,
1105
- "gen_tokens": 13,
1106
- "reason": "fail:SyntaxError: 'return' outside function",
1107
- "tail": " return [x + 3 for x in xs]"
1108
- },
1109
- {
1110
- "task_id": "v27/aggregate_list/04",
1111
- "entry_point": "aggregate",
1112
- "ok": false,
1113
- "gen_tokens": 16,
1114
- "reason": "fail:SyntaxError: 'return' outside function",
1115
- "tail": " return sum(1 for x in xs if x > 0)"
1116
- },
1117
- {
1118
- "task_id": "v27/string_transform/05",
1119
- "entry_point": "alternate_case",
1120
- "ok": true,
1121
- "gen_tokens": 74,
1122
- "reason": "pass",
1123
- "tail": "sult.append(char.upper())\n else:\n result.append(char.lower())\n else:\n result.append(char)\n return ''.join(result)"
1124
- },
1125
- {
1126
- "task_id": "v27/window_sum/06",
1127
- "entry_point": "window_max_sum",
1128
- "ok": true,
1129
- "gen_tokens": 146,
1130
- "reason": "pass",
1131
- "tail": "for i in range(k, len(xs)):\n current_sum += xs[i] - xs[i - k]\n if current_sum > max_sum:\n max_sum = current_sum\n \n return max_sum"
1132
- },
1133
- {
1134
- "task_id": "v27/run_length/07",
1135
- "entry_point": "rle",
1136
- "ok": false,
1137
- "gen_tokens": 94,
1138
- "reason": "fail:IndentationError: unindent does not match any outer indentation level",
1139
- "tail": " result.append((current_char, count))\n current_char = char\n count = 1\n result.append((current_char, count))\n return result"
1140
- },
1141
- {
1142
- "task_id": "v27/string_predicate/08",
1143
- "entry_point": "is_balanced",
1144
- "ok": false,
1145
- "gen_tokens": 56,
1146
- "reason": "fail:IndentationError: unexpected indent",
1147
- "tail": " stack.append(char)\n elif char == ')':\n if not stack:\n return False\n stack.pop()\n return len(stack) == 0"
1148
- },
1149
- {
1150
- "task_id": "v27/pair_count/09",
1151
- "entry_point": "pair_count",
1152
- "ok": false,
1153
- "gen_tokens": 58,
1154
- "reason": "fail:IndentationError: unexpected indent",
1155
- "tail": "ount = 0\n n = len(xs)\n for i in range(n):\n for j in range(i + 1, n):\n if xs[i] + xs[j] == 9:\n count += 1\n return count"
1156
- }
1157
- ],
1158
- "error": null,
1159
- "mean_gen_tokens": 59.9,
1160
- "mean_gen_tokens_correct": 110.0
1161
- },
1162
- "reasoning_bench": {
1163
- "n": 12,
1164
- "correct": 6,
1165
- "pass_frac": 0.5,
1166
- "wall_s": 3.8,
1167
- "items": [
1168
- {
1169
- "src": "procedural_reasoning/sequence_next",
1170
- "pred": "(B)",
1171
- "gold": "(B)",
1172
- "ok": true,
1173
- "gen_tokens": 3,
1174
- "tail": "(B)"
1175
- },
1176
- {
1177
- "src": "procedural_reasoning/boolean_eval",
1178
- "pred": "(A)",
1179
- "gold": "(A)",
1180
- "ok": true,
1181
- "gen_tokens": 3,
1182
- "tail": "(A)"
1183
- },
1184
- {
1185
- "src": "procedural_reasoning/deduction",
1186
- "pred": "(B)",
1187
- "gold": "(A)",
1188
- "ok": false,
1189
- "gen_tokens": 3,
1190
- "tail": "(B)"
1191
- },
1192
- {
1193
- "src": "procedural_reasoning/analogy_letter",
1194
- "pred": "(B)",
1195
- "gold": "(B)",
1196
- "ok": true,
1197
- "gen_tokens": 3,
1198
- "tail": "(B)"
1199
- },
1200
- {
1201
- "src": "procedural_reasoning/odd_one_out",
1202
- "pred": "(B)",
1203
- "gold": "(A)",
1204
- "ok": false,
1205
- "gen_tokens": 3,
1206
- "tail": "(B)"
1207
- },
1208
- {
1209
- "src": "procedural_reasoning/ordering",
1210
- "pred": "(C)",
1211
- "gold": "(D)",
1212
- "ok": false,
1213
- "gen_tokens": 3,
1214
- "tail": "(C)"
1215
- },
1216
- {
1217
- "src": "procedural_reasoning/sequence_next",
1218
- "pred": "(B)",
1219
- "gold": "(A)",
1220
- "ok": false,
1221
- "gen_tokens": 3,
1222
- "tail": "(B)"
1223
- },
1224
- {
1225
- "src": "procedural_reasoning/boolean_eval",
1226
- "pred": "(A)",
1227
- "gold": "(A)",
1228
- "ok": true,
1229
- "gen_tokens": 3,
1230
- "tail": "(A)"
1231
- },
1232
- {
1233
- "src": "procedural_reasoning/deduction",
1234
- "pred": "(B)",
1235
- "gold": "(B)",
1236
- "ok": true,
1237
- "gen_tokens": 3,
1238
- "tail": "(B)"
1239
- },
1240
- {
1241
- "src": "procedural_reasoning/analogy_letter",
1242
- "pred": "(A)",
1243
- "gold": "(D)",
1244
- "ok": false,
1245
- "gen_tokens": 3,
1246
- "tail": "(A)"
1247
- },
1248
- {
1249
- "src": "procedural_reasoning/odd_one_out",
1250
- "pred": "(A)",
1251
- "gold": "(C)",
1252
- "ok": false,
1253
- "gen_tokens": 3,
1254
- "tail": "(A)"
1255
- },
1256
- {
1257
- "src": "procedural_reasoning/ordering",
1258
- "pred": "(B)",
1259
- "gold": "(B)",
1260
- "ok": true,
1261
- "gen_tokens": 3,
1262
- "tail": "(B)"
1263
- }
1264
- ],
1265
- "error": null,
1266
- "mean_gen_tokens": 3.0,
1267
- "mean_gen_tokens_correct": 3.0
1268
- },
1269
- "knowledge_bench": {
1270
- "n": 0,
1271
- "correct": 0,
1272
- "pass_frac": 0.0,
1273
- "wall_s": 0.0,
1274
- "items": [],
1275
- "error": null,
1276
- "mean_gen_tokens": 0.0,
1277
- "mean_gen_tokens_correct": 0.0
1278
- },
1279
- "ifeval_bench": {
1280
- "n": 10,
1281
- "correct": 7,
1282
- "pass_frac": 0.7,
1283
- "wall_s": 106.2,
1284
- "items": [
1285
- {
1286
- "src": "procedural_ifeval/json_format",
1287
- "instruction_ids": [
1288
- "detectable_format:json_format"
1289
- ],
1290
- "per_instruction": [
1291
- true
1292
- ],
1293
- "ok": true,
1294
- "gen_tokens": 96,
1295
- "tail": "y spaces.\",\n \"keywords\": [\n \"library\",\n \"renovation\",\n \"reopening\",\n \"community\",\n \"modernization\"\n ]\n}"
1296
- },
1297
- {
1298
- "src": "procedural_ifeval/max_words",
1299
- "instruction_ids": [
1300
- "length_constraints:number_words"
1301
- ],
1302
- "per_instruction": [
1303
- true
1304
- ],
1305
- "ok": true,
1306
- "gen_tokens": 17,
1307
- "tail": "Urban gardening brings fresh food, stress relief, and community connection to city living."
1308
- },
1309
- {
1310
- "src": "procedural_ifeval/min_words",
1311
- "instruction_ids": [
1312
- "length_constraints:number_words"
1313
- ],
1314
- "per_instruction": [
1315
- false
1316
- ],
1317
- "ok": false,
1318
- "gen_tokens": 94,
1319
- "tail": "rant hub for learning and connection, ensuring it remains a cherished cornerstone of the neighborhood for years to come."
1320
- },
1321
- {
1322
- "src": "procedural_ifeval/exact_sentences",
1323
- "instruction_ids": [
1324
- "length_constraints:number_sentences"
1325
- ],
1326
- "per_instruction": [
1327
- true
1328
- ],
1329
- "ok": true,
1330
- "gen_tokens": 31,
1331
- "tail": "raffic through treacherous waters. Their isolation demanded immense resilience and a deep dedication to maritime safety."
1332
- },
1333
- {
1334
- "src": "procedural_ifeval/ends_with_phrase",
1335
- "instruction_ids": [
1336
- "startend:end_checker"
1337
- ],
1338
- "per_instruction": [
1339
- false
1340
- ],
1341
- "ok": false,
1342
- "gen_tokens": 86,
1343
- "tail": "requently associate with a single generation of genetic memory and environmental cues rather than individual experience."
1344
- },
1345
- {
1346
- "src": "procedural_ifeval/all_lowercase",
1347
- "instruction_ids": [
1348
- "change_case:english_lowercase"
1349
- ],
1350
- "per_instruction": [
1351
- true
1352
- ],
1353
- "ok": true,
1354
- "gen_tokens": 53,
1355
- "tail": "ures are meticulously designed to withstand harsh weather conditions and seismic activity in high-altitude environments."
1356
- },
1357
- {
1358
- "src": "procedural_ifeval/bullet_list",
1359
- "instruction_ids": [
1360
- "detectable_format:number_bullet_lists"
1361
- ],
1362
- "per_instruction": [
1363
- false
1364
- ],
1365
- "ok": false,
1366
- "gen_tokens": 238,
1367
- "tail": " of the sun, Earth's magnetic field, and visual landmarks to find their way to the same specific groves year after year."
1368
- },
1369
- {
1370
- "src": "procedural_ifeval/no_comma",
1371
- "instruction_ids": [
1372
- "punctuation:no_comma"
1373
- ],
1374
- "per_instruction": [
1375
- true
1376
- ],
1377
- "ok": true,
1378
- "gen_tokens": 47,
1379
- "tail": "es mental health. Communities bond over shared green spaces. This hobby connects people with nature in concrete jungles."
1380
- },
1381
- {
1382
- "src": "procedural_ifeval/include_keyword",
1383
- "instruction_ids": [
1384
- "keywords:frequency"
1385
- ],
1386
- "per_instruction": [
1387
- true
1388
- ],
1389
- "ok": true,
1390
- "gen_tokens": 133,
1391
- "tail": "preserving this delicate ecological **compass**, ensuring that the ancient pathways remain intact for future migrations."
1392
- },
1393
- {
1394
- "src": "procedural_ifeval/all_uppercase",
1395
- "instruction_ids": [
1396
- "change_case:english_capital"
1397
- ],
1398
- "per_instruction": [
1399
- true
1400
- ],
1401
- "ok": true,
1402
- "gen_tokens": 256,
1403
- "tail": "GISTS ANALOGEOROLOGISTS ANALOGEOROLOGISTS ANALOGEOROLOGISTS ANALOGEOROLOGISTS ANALOGEOROLOGISTS ANALOGEOROLOGISTS ANALOG"
1404
- }
1405
- ],
1406
- "error": null,
1407
- "mean_gen_tokens": 105.1,
1408
- "mean_gen_tokens_correct": 90.4
1409
- },
1410
- "aime_bench": {
1411
- "n": 10,
1412
- "correct": 0,
1413
- "pass_frac": 0.0,
1414
- "wall_s": 5.2,
1415
- "items": [
1416
- {
1417
- "src": "procedural_aime/factor_chain",
1418
- "pred": "44",
1419
- "gold": "37",
1420
- "ok": false,
1421
- "gen_tokens": 7,
1422
- "tail": "\\boxed{44}"
1423
- },
1424
- {
1425
- "src": "procedural_aime/chained_modular",
1426
- "pred": "10",
1427
- "gold": "22",
1428
- "ok": false,
1429
- "gen_tokens": 7,
1430
- "tail": "\\boxed{10}"
1431
- },
1432
- {
1433
- "src": "procedural_aime/lcm_residue",
1434
- "pred": "105",
1435
- "gold": "65",
1436
- "ok": false,
1437
- "gen_tokens": 8,
1438
- "tail": "\\boxed{105}"
1439
- },
1440
- {
1441
- "src": "procedural_aime/iterated_digit",
1442
- "pred": "108",
1443
- "gold": "59",
1444
- "ok": false,
1445
- "gen_tokens": 8,
1446
- "tail": "\\boxed{108}"
1447
- },
1448
- {
1449
- "src": "procedural_aime/diophantine",
1450
- "pred": "10",
1451
- "gold": "8",
1452
- "ok": false,
1453
- "gen_tokens": 7,
1454
- "tail": "\\boxed{10}"
1455
- },
1456
- {
1457
- "src": "procedural_aime/factor_chain",
1458
- "pred": "44",
1459
- "gold": "45",
1460
- "ok": false,
1461
- "gen_tokens": 7,
1462
- "tail": "\\boxed{44}"
1463
- },
1464
- {
1465
- "src": "procedural_aime/chained_modular",
1466
- "pred": "10",
1467
- "gold": "50",
1468
- "ok": false,
1469
- "gen_tokens": 7,
1470
- "tail": "\\boxed{10}"
1471
- },
1472
- {
1473
- "src": "procedural_aime/lcm_residue",
1474
- "pred": "105",
1475
- "gold": "83",
1476
- "ok": false,
1477
- "gen_tokens": 8,
1478
- "tail": "\\boxed{105}"
1479
- },
1480
- {
1481
- "src": "procedural_aime/iterated_digit",
1482
- "pred": "108",
1483
- "gold": "73",
1484
- "ok": false,
1485
- "gen_tokens": 8,
1486
- "tail": "\\boxed{108}"
1487
- },
1488
- {
1489
- "src": "procedural_aime/diophantine",
1490
- "pred": "10",
1491
- "gold": "202",
1492
- "ok": false,
1493
- "gen_tokens": 7,
1494
- "tail": "\\boxed{10}"
1495
- }
1496
- ],
1497
- "error": null,
1498
- "mean_gen_tokens": 7.4,
1499
- "mean_gen_tokens_correct": 0.0
1500
- },
1501
- "mbpp_bench": {
1502
- "n": 10,
1503
- "correct": 10,
1504
- "pass_frac": 1.0,
1505
- "wall_s": 90.9,
1506
- "items": [
1507
- {
1508
- "task_id": "v27/string_predicate/00",
1509
- "entry_point": "is_anagram",
1510
- "ok": true,
1511
- "gen_tokens": 64,
1512
- "reason": "pass",
1513
- "tail": "nagram of string ``b`` (case-sensitive). Different lengths return False.\n \"\"\"\n if len(a) != len(b):\n return False\n return sorted(a) == sorted(b)"
1514
- },
1515
- {
1516
- "task_id": "v27/string_transform/01",
1517
- "entry_point": "swap_pairs",
1518
- "ok": true,
1519
- "gen_tokens": 106,
1520
- "reason": "pass",
1521
- "tail": "n place.\n \"\"\"\n chars = list(s)\n for i in range(0, len(chars) - 1, 2):\n chars[i], chars[i + 1] = chars[i + 1], chars[i]\n return ''.join(chars)"
1522
- },
1523
- {
1524
- "task_id": "v27/aggregate_list/02",
1525
- "entry_point": "aggregate",
1526
- "ok": true,
1527
- "gen_tokens": 45,
1528
- "reason": "pass",
1529
- "tail": " \"\"\"Return the difference between the largest and the smallest value in ``xs``. The list always has at least one element.\n \"\"\"\n return max(xs) - min(xs)"
1530
- },
1531
- {
1532
- "task_id": "v27/run_length/03",
1533
- "entry_point": "rle",
1534
- "ok": true,
1535
- "gen_tokens": 100,
1536
- "reason": "pass",
1537
- "tail": " result.append((current_char, count))\n current_char = char\n count = 1\n result.append((current_char, count))\n return result"
1538
- },
1539
- {
1540
- "task_id": "v27/transform_list/04",
1541
- "entry_point": "transform",
1542
- "ok": true,
1543
- "gen_tokens": 18,
1544
- "reason": "pass",
1545
- "tail": "def transform(xs):\n return [x + 3 for x in xs]"
1546
- },
1547
- {
1548
- "task_id": "v27/digit_sum/05",
1549
- "entry_point": "count_divisors",
1550
- "ok": true,
1551
- "gen_tokens": 103,
1552
- "reason": "pass",
1553
- "tail": " <= n:\n if n % i == 0:\n if i * i == n:\n count += 1\n else:\n count += 2\n i += 1\n return count"
1554
- },
1555
- {
1556
- "task_id": "v27/window_sum/06",
1557
- "entry_point": "window_max_sum",
1558
- "ok": true,
1559
- "gen_tokens": 191,
1560
- "reason": "pass",
1561
- "tail": "):\n current_sum = current_sum - xs[i - 1] + xs[i + k - 1]\n if current_sum > max_sum:\n max_sum = current_sum\n \n return max_sum\n```"
1562
- },
1563
- {
1564
- "task_id": "v27/pair_count/07",
1565
- "entry_point": "pair_count",
1566
- "ok": true,
1567
- "gen_tokens": 100,
1568
- "reason": "pass",
1569
- "tail": "ount = 0\n n = len(xs)\n for i in range(n):\n for j in range(i + 1, n):\n if xs[i] + xs[j] == 4:\n count += 1\n return count"
1570
- },
1571
- {
1572
- "task_id": "v27/string_predicate/08",
1573
- "entry_point": "is_anagram",
1574
- "ok": true,
1575
- "gen_tokens": 64,
1576
- "reason": "pass",
1577
- "tail": "nagram of string ``b`` (case-sensitive). Different lengths return False.\n \"\"\"\n if len(a) != len(b):\n return False\n return sorted(a) == sorted(b)"
1578
- },
1579
- {
1580
- "task_id": "v27/string_transform/09",
1581
- "entry_point": "swap_pairs",
1582
- "ok": true,
1583
- "gen_tokens": 106,
1584
- "reason": "pass",
1585
- "tail": "n place.\n \"\"\"\n chars = list(s)\n for i in range(0, len(chars) - 1, 2):\n chars[i], chars[i + 1] = chars[i + 1], chars[i]\n return ''.join(chars)"
1586
- }
1587
- ],
1588
- "error": null,
1589
- "mean_gen_tokens": 89.7,
1590
- "mean_gen_tokens_correct": 89.7
1591
- },
1592
- "tool_use_bench": {
1593
- "n": 8,
1594
- "correct": 0,
1595
- "pass_frac": 0.0,
1596
- "wall_s": 6.2,
1597
- "items": [
1598
- {
1599
- "src": "procedural_math/probability_count",
1600
- "tool_used": false,
1601
- "tool_result": null,
1602
- "pred": "10",
1603
- "gold": "16",
1604
- "ok": false,
1605
- "gen_tokens": 7,
1606
- "tail": "\\boxed{10}"
1607
- },
1608
- {
1609
- "src": "procedural_math/triangle_area",
1610
- "tool_used": false,
1611
- "tool_result": null,
1612
- "pred": "155",
1613
- "gold": "45",
1614
- "ok": false,
1615
- "gen_tokens": 8,
1616
- "tail": "\\boxed{155}"
1617
- },
1618
- {
1619
- "src": "procedural_math/unit_conversion",
1620
- "tool_used": false,
1621
- "tool_result": null,
1622
- "pred": "288",
1623
- "gold": "208",
1624
- "ok": false,
1625
- "gen_tokens": 8,
1626
- "tail": "\\boxed{288}"
1627
- },
1628
- {
1629
- "src": "procedural_math/factorial_mod",
1630
- "tool_used": false,
1631
- "tool_result": null,
1632
- "pred": "13",
1633
- "gold": "6",
1634
- "ok": false,
1635
- "gen_tokens": 7,
1636
- "tail": "\\boxed{13}"
1637
- },
1638
- {
1639
- "src": "procedural_math/modular_linear",
1640
- "tool_used": false,
1641
- "tool_result": null,
1642
- "pred": "1",
1643
- "gold": "11",
1644
- "ok": false,
1645
- "gen_tokens": 6,
1646
- "tail": "\\boxed{1}"
1647
- },
1648
- {
1649
- "src": "procedural_math/coin_change",
1650
- "tool_used": false,
1651
- "tool_result": null,
1652
- "pred": "105",
1653
- "gold": "6",
1654
- "ok": false,
1655
- "gen_tokens": 8,
1656
- "tail": "\\boxed{105}"
1657
- },
1658
- {
1659
- "src": "procedural_math/mixture",
1660
- "tool_used": false,
1661
- "tool_result": null,
1662
- "pred": "3250",
1663
- "gold": "4980",
1664
- "ok": false,
1665
- "gen_tokens": 9,
1666
- "tail": "\\boxed{3250}"
1667
- },
1668
- {
1669
- "src": "procedural_math/time_arithmetic",
1670
- "tool_used": false,
1671
- "tool_result": null,
1672
- "pred": "101",
1673
- "gold": "609",
1674
- "ok": false,
1675
- "gen_tokens": 8,
1676
- "tail": "\\boxed{101}"
1677
- }
1678
- ],
1679
- "error": null,
1680
- "mean_gen_tokens": 7.6,
1681
- "mean_gen_tokens_correct": 0.0
1682
- },
1683
- "self_consistency_bench": {
1684
- "n": 0,
1685
- "correct": 0,
1686
- "pass_frac": 0.0,
1687
- "wall_s": 0.0,
1688
- "items": [],
1689
- "error": null,
1690
- "mean_gen_tokens": 0.0,
1691
- "mean_gen_tokens_correct": 0.0
1692
- },
1693
- "arc_bench": {
1694
- "n": 0,
1695
- "correct": 0,
1696
- "pass_frac": 0.0,
1697
- "wall_s": 0.0,
1698
- "items": [],
1699
- "error": null,
1700
- "mean_gen_tokens": 0.0,
1701
- "mean_gen_tokens_correct": 0.0
1702
- },
1703
- "truthful_bench": {
1704
- "n": 0,
1705
- "correct": 0,
1706
- "pass_frac": 0.0,
1707
- "wall_s": 0.0,
1708
- "items": [],
1709
- "error": null,
1710
- "mean_gen_tokens": 0.0,
1711
- "mean_gen_tokens_correct": 0.0
1712
- },
1713
- "long_context_bench": {
1714
- "n": 6,
1715
- "correct": 0,
1716
- "pass_frac": 0.0,
1717
- "wall_s": 3.1,
1718
- "items": [
1719
- {
1720
- "src": "long_context",
1721
- "gold": "XWMBUM7",
1722
- "pred_tail": "D3K64",
1723
- "ok": false,
1724
- "gen_tokens": 6,
1725
- "needle_position": 19,
1726
- "confuser_positions": [
1727
- 1,
1728
- 42,
1729
- 23
1730
- ],
1731
- "confuser_hit": false,
1732
- "gold_in_pred": false
1733
- },
1734
- {
1735
- "src": "long_context",
1736
- "gold": "N2NERPB",
1737
- "pred_tail": "256J4U",
1738
- "ok": false,
1739
- "gen_tokens": 7,
1740
- "needle_position": 13,
1741
- "confuser_positions": [
1742
- 9,
1743
- 35,
1744
- 30
1745
- ],
1746
- "confuser_hit": false,
1747
- "gold_in_pred": false
1748
- },
1749
- {
1750
- "src": "long_context",
1751
- "gold": "V8C3MQG",
1752
- "pred_tail": "7YQMBB",
1753
- "ok": false,
1754
- "gen_tokens": 6,
1755
- "needle_position": 15,
1756
- "confuser_positions": [
1757
- 42,
1758
- 7,
1759
- 18
1760
- ],
1761
- "confuser_hit": false,
1762
- "gold_in_pred": false
1763
- },
1764
- {
1765
- "src": "long_context",
1766
- "gold": "7PXTYV2",
1767
- "pred_tail": "7S2P2B",
1768
- "ok": false,
1769
- "gen_tokens": 7,
1770
- "needle_position": 26,
1771
- "confuser_positions": [
1772
- 1,
1773
- 15,
1774
- 29
1775
- ],
1776
- "confuser_hit": false,
1777
- "gold_in_pred": false
1778
- },
1779
- {
1780
- "src": "long_context",
1781
- "gold": "K4USQ2C",
1782
- "pred_tail": "9RZC62",
1783
- "ok": false,
1784
- "gen_tokens": 7,
1785
- "needle_position": 13,
1786
- "confuser_positions": [
1787
- 3,
1788
- 33,
1789
- 20
1790
- ],
1791
- "confuser_hit": false,
1792
- "gold_in_pred": false
1793
- },
1794
- {
1795
- "src": "long_context",
1796
- "gold": "ERYJ6SQ",
1797
- "pred_tail": "FJNBK6",
1798
- "ok": false,
1799
- "gen_tokens": 6,
1800
- "needle_position": 12,
1801
- "confuser_positions": [
1802
- 19,
1803
- 22,
1804
- 25
1805
- ],
1806
- "confuser_hit": false,
1807
- "gold_in_pred": false
1808
- }
1809
- ],
1810
- "error": null,
1811
- "mean_gen_tokens": 6.5,
1812
- "mean_gen_tokens_correct": 0.0
1813
- },
1814
- "procedural_bench": {
1815
- "n": 0,
1816
- "correct": 0,
1817
- "pass_frac": 0.0,
1818
- "wall_s": 0.0,
1819
- "items": [],
1820
- "error": null,
1821
- "mean_gen_tokens": 0.0,
1822
- "mean_gen_tokens_correct": 0.0
1823
- },
1824
- "robustness_bench": {
1825
- "n": 16,
1826
- "correct": 0,
1827
- "pass_frac": 0.0,
1828
- "wall_s": 11.2,
1829
- "items": [
1830
- {
1831
- "src": "procedural_math/geometric_series",
1832
- "perturbation": "instruction_synonym",
1833
- "pred": "100",
1834
- "gold": "75",
1835
- "ok": false,
1836
- "gen_tokens": 6
1837
- },
1838
- {
1839
- "src": "procedural_math/geometric_series",
1840
- "perturbation": "context_prefix",
1841
- "pred": "100",
1842
- "gold": "75",
1843
- "ok": false,
1844
- "gen_tokens": 6
1845
- },
1846
- {
1847
- "src": "procedural_math/mixture",
1848
- "perturbation": "instruction_synonym",
1849
- "pred": "4200",
1850
- "gold": "5590",
1851
- "ok": false,
1852
- "gen_tokens": 7
1853
- },
1854
- {
1855
- "src": "procedural_math/mixture",
1856
- "perturbation": "context_prefix",
1857
- "pred": "4200",
1858
- "gold": "5590",
1859
- "ok": false,
1860
- "gen_tokens": 7
1861
- },
1862
- {
1863
- "src": "procedural_math/probability_count",
1864
- "perturbation": "instruction_synonym",
1865
- "pred": "10",
1866
- "gold": "25",
1867
- "ok": false,
1868
- "gen_tokens": 5
1869
- },
1870
- {
1871
- "src": "procedural_math/probability_count",
1872
- "perturbation": "context_prefix",
1873
- "pred": "10",
1874
- "gold": "25",
1875
- "ok": false,
1876
- "gen_tokens": 5
1877
- },
1878
- {
1879
- "src": "procedural_math/polynomial_eval",
1880
- "perturbation": "instruction_synonym",
1881
- "pred": "1",
1882
- "gold": "28",
1883
- "ok": false,
1884
- "gen_tokens": 18
1885
- },
1886
- {
1887
- "src": "procedural_math/polynomial_eval",
1888
- "perturbation": "context_prefix",
1889
- "pred": "10",
1890
- "gold": "28",
1891
- "ok": false,
1892
- "gen_tokens": 5
1893
- },
1894
- {
1895
- "src": "procedural_math/proportion",
1896
- "perturbation": "instruction_synonym",
1897
- "pred": "1084",
1898
- "gold": "1287",
1899
- "ok": false,
1900
- "gen_tokens": 7
1901
- },
1902
- {
1903
- "src": "procedural_math/proportion",
1904
- "perturbation": "context_prefix",
1905
- "pred": "1021",
1906
- "gold": "1287",
1907
- "ok": false,
1908
- "gen_tokens": 7
1909
- },
1910
- {
1911
- "src": "procedural_math/set_intersect",
1912
- "perturbation": "instruction_synonym",
1913
- "pred": "100",
1914
- "gold": "103",
1915
- "ok": false,
1916
- "gen_tokens": 6
1917
- },
1918
- {
1919
- "src": "procedural_math/set_intersect",
1920
- "perturbation": "context_prefix",
1921
- "pred": "100",
1922
- "gold": "103",
1923
- "ok": false,
1924
- "gen_tokens": 6
1925
- },
1926
- {
1927
- "src": "procedural_math/rate_distance",
1928
- "perturbation": "instruction_synonym",
1929
- "pred": "102",
1930
- "gold": "28",
1931
- "ok": false,
1932
- "gen_tokens": 6
1933
- },
1934
- {
1935
- "src": "procedural_math/rate_distance",
1936
- "perturbation": "context_prefix",
1937
- "pred": "102",
1938
- "gold": "28",
1939
- "ok": false,
1940
- "gen_tokens": 6
1941
- },
1942
- {
1943
- "src": "procedural_math/digit_sum",
1944
- "perturbation": "instruction_synonym",
1945
- "pred": "122",
1946
- "gold": "167",
1947
- "ok": false,
1948
- "gen_tokens": 6
1949
- },
1950
- {
1951
- "src": "procedural_math/digit_sum",
1952
- "perturbation": "context_prefix",
1953
- "pred": "113",
1954
- "gold": "167",
1955
- "ok": false,
1956
- "gen_tokens": 6
1957
- }
1958
- ],
1959
- "error": null,
1960
- "mean_gen_tokens": 6.8,
1961
- "mean_gen_tokens_correct": 0.0
1962
- },
1963
- "noise_resistance_bench": {
1964
- "n": 0,
1965
- "correct": 0,
1966
- "pass_frac": 0.0,
1967
- "wall_s": 0.0,
1968
- "items": [],
1969
- "error": null,
1970
- "mean_gen_tokens": 0.0,
1971
- "mean_gen_tokens_correct": 0.0
1972
- },
1973
- "bench_gate": {
1974
- "run": true,
1975
- "reason": "no_king_kl",
1976
- "only_kl_candidates": true,
1977
- "kl_candidate_slack": 0.0,
1978
- "student_kl": 1.663056,
1979
- "king_kl": null,
1980
- "wall_s": 296.5
1981
- },
1982
- "on_policy_rkl": {
1983
- "n_rollouts": 12,
1984
- "tokens": 1497,
1985
- "mean_rkl": 0.520277,
1986
- "mean_fkl": 1.152991,
1987
- "mean_skl": 0.275421,
1988
- "mean_sampled_gap": 0.594948,
1989
- "mean_gen_len": 124.8,
1990
- "skew_alpha": 0.1,
1991
- "top_k": 128,
1992
- "per_rollout": [
1993
- {
1994
- "prompt": "Who painted the Mona Lisa?",
1995
- "gen_len": 125,
1996
- "tokens_scored": 125,
1997
- "rkl": 0.409653,
1998
- "fkl": 0.928422,
1999
- "skl": 0.210175,
2000
- "sampled_gap": 0.44171,
2001
- "tail": " period: Early 16th century (around 1503\u20131506, possibly continued until 1517).\n * Current location: The Louvre Museum, Paris.\n\n3. **Formulate the answer:** Directly state the name of the artist."
2002
- },
2003
- {
2004
- "prompt": "How many continents are there?",
2005
- "gen_len": 125,
2006
- "tokens_scored": 125,
2007
- "rkl": 0.582701,
2008
- "fkl": 1.658169,
2009
- "skl": 0.383538,
2010
- "sampled_gap": 0.757408,
2011
- "tail": "frica, Antarctica, Asia, Europe, North America, Oceania/Australia, South America).\n * **Alternative Answers (6 continents):** The \"6-continent model\" often merges Europe and Asia into Eurasia, or"
2012
- },
2013
- {
2014
- "prompt": "Explain a sunset using exactly three adjectives.",
2015
- "gen_len": 125,
2016
- "tokens_scored": 125,
2017
- "rkl": 0.621109,
2018
- "fkl": 1.525348,
2019
- "skl": 0.362313,
2020
- "sampled_gap": 0.718639,
2021
- "tail": "m, soft.\n * Mood/Quality: Beautiful, breathtaking, dramatic, serene, fading, fading, brilliant, melancholic.\n * State: Setting, dying, sinking, glowing.\n\n3. **Drafting Sentences (aiming for"
2022
- },
2023
- {
2024
- "prompt": "What is the square root of 144?",
2025
- "gen_len": 125,
2026
- "tokens_scored": 125,
2027
- "rkl": 0.429711,
2028
- "fkl": 0.696605,
2029
- "skl": 0.179877,
2030
- "sampled_gap": 0.384291,
2031
- "tail": "eed to find a number that, when multiplied by itself, equals 144.\n3. **Calculate:**\n * $10 \\times 10 = 100$ (too small)\n * $12 \\times 12 = 144$ (correct)\n * Also, $\\sqrt{144} = 12$.\n4."
2032
- },
2033
- {
2034
- "prompt": "Translate to French: The cat sat on the mat.",
2035
- "gen_len": 125,
2036
- "tokens_scored": 125,
2037
- "rkl": 0.611311,
2038
- "fkl": 1.10299,
2039
- "skl": 0.276324,
2040
- "sampled_gap": 0.677515,
2041
- "tail": "e chat\"\n * \"sat\" -> \"s'\u00e9tait assis\" (pass\u00e9 compos\u00e9 would be \"s'est assis\", but simple past \"pass\u00e9 simple\" is \"s'assit\", though in modern French narrative, \"\u00e9tait assis\" or \"s'est assis\" is common"
2042
- },
2043
- {
2044
- "prompt": "In plain language, what is an operating system?",
2045
- "gen_len": 125,
2046
- "tokens_scored": 125,
2047
- "rkl": 0.552007,
2048
- "fkl": 1.453894,
2049
- "skl": 0.328995,
2050
- "sampled_gap": 0.621598,
2051
- "tail": "fy Key Concepts:**\n * What does an OS do? It manages hardware and software.\n * What does the user see? Windows, macOS, iOS, Android icons.\n * What does the computer do? Processes tasks,"
2052
- },
2053
- {
2054
- "prompt": "In exactly one sentence, outline what courage is.",
2055
- "gen_len": 125,
2056
- "tokens_scored": 125,
2057
- "rkl": 0.690625,
2058
- "fkl": 1.588047,
2059
- "skl": 0.396777,
2060
- "sampled_gap": 0.723275,
2061
- "tail": "olved. It's about acting in spite of fear. It involves strength, bravery, moral fortitude, and taking action.\n\n3. **Drafting Sentences:**\n * *Draft 1:* Courage is the ability to face fear and do"
2062
- },
2063
- {
2064
- "prompt": "Outline what a compiler does, in one paragraph.",
2065
- "gen_len": 125,
2066
- "tokens_scored": 125,
2067
- "rkl": 0.454211,
2068
- "fkl": 1.035425,
2069
- "skl": 0.244046,
2070
- "sampled_gap": 0.524094,
2071
- "tail": "binary, machine-readable, low-level).\n * Process: Translation, parsing, optimization, code generation.\n * Timing: Before execution (static translation).\n * Purpose: Bridge between human"
2072
- },
2073
- {
2074
- "prompt": "What is a linked enumerate in computer science? Two sentences.",
2075
- "gen_len": 125,
2076
- "tokens_scored": 125,
2077
- "rkl": 0.589372,
2078
- "fkl": 1.265752,
2079
- "skl": 0.276325,
2080
- "sampled_gap": 0.690847,
2081
- "tail": "ist, Enumeration (enum), Enumerate (function), Linked enumerate?\n * Hypothesis: It's likely a confusion or a non-standard term.\n * Possibility A: \"Linked List\" + \"Enumeration\".\n *"
2082
- },
2083
- {
2084
- "prompt": "Give three synonyms of 'happy', one per line.",
2085
- "gen_len": 123,
2086
- "tokens_scored": 123,
2087
- "rkl": 0.326833,
2088
- "fkl": 0.623001,
2089
- "skl": 0.162384,
2090
- "sampled_gap": 0.470018,
2091
- "tail": " * Joyful\n * Glad\n * Content\n * Cheery\n * Blissful\n * Pleased\n * Delighted\n * Cheerful\n\n3. **Select Three Distinct Synonyms:**\n * Joyful\n * Glad\n *"
2092
- },
2093
- {
2094
- "prompt": "Translate to Dutch: The red house is on the left.",
2095
- "gen_len": 125,
2096
- "tokens_scored": 125,
2097
- "rkl": 0.430915,
2098
- "fkl": 0.787817,
2099
- "skl": 0.202737,
2100
- "sampled_gap": 0.507985,
2101
- "tail": " \"The\": Definite article (singular, neuter/feminine/masculine depending on noun).\n * \"red\": Adjective (color).\n * \"house\": Noun (gender: het huis).\n * \"is\": Verb (to be, present tense,"
2102
- },
2103
- {
2104
- "prompt": "Define machine learning in one sentence.",
2105
- "gen_len": 124,
2106
- "tokens_scored": 124,
2107
- "rkl": 0.541953,
2108
- "fkl": 1.162009,
2109
- "skl": 0.279784,
2110
- "sampled_gap": 0.620198,
2111
- "tail": "rogramming for every task.\n * Improvement over time (usually).\n\n3. **Draft Initial Definitions:**\n * Machine learning is a subset of artificial intelligence where computers learn from data."
2112
- }
2113
- ],
2114
- "scoring_time": 5.0
2115
- },
2116
- "judge_probe": {
2117
- "n": 16,
2118
- "n_valid": 16,
2119
- "mean_score": 4.875,
2120
- "normalized": 0.9688,
2121
- "per_prompt": [
2122
- {
2123
- "prompt": "Name one thing you should check before submitting a pull request. Single-line answer.",
2124
- "response_preview": "Run the project's tests to ensure nothing is broken.",
2125
- "raw": "5",
2126
- "score": 5
2127
- },
2128
- {
2129
- "prompt": "In one short paragraph, explain why you would or wouldn't suggest ice baths for sore muscles after exercise.",
2130
- "response_preview": "I would suggest ice baths with caution, as they are effective at rapidly reducing inflammation and pain immediately afte",
2131
- "raw": "5",
2132
- "score": 5
2133
- },
2134
- {
2135
- "prompt": "You have a 3-liter jug and a 5-liter jug and a water source. Outline briefly how to measure exactly 4 liters.",
2136
- "response_preview": "Here is the step-by-step process to measure exactly 4 liters using a 3-liter jug and a 5-liter jug:\n\n1. **Fill** the 3-",
2137
- "raw": "4",
2138
- "score": 4
2139
- },
2140
- {
2141
- "prompt": "Offer one creative metaphor comparing a commit history to something from everyday life.",
2142
- "response_preview": "A commit history is like the **layered strata of a geological formation**, where each commit is a distinct sediment depo",
2143
- "raw": "5",
2144
- "score": 5
2145
- },
2146
- {
2147
- "prompt": "If it is 2:30 PM now, what time will it be 250 minutes from now? Demonstrate your work.",
2148
- "response_preview": "To determine the time 250 minutes from 2:30 PM, we will break the calculation down into hours and minutes, then add them",
2149
- "raw": "4",
2150
- "score": 4
2151
- },
2152
- {
2153
- "prompt": "What is the output of `print(list(range(3, 10, 2)))` in Python? Just the output.",
2154
- "response_preview": "[3, 5, 7, 9]",
2155
- "raw": "5",
2156
- "score": 5
2157
- },
2158
- {
2159
- "prompt": "Produce a haiku about winter. The only output should be three lines of the haiku \u2014 no intro, no explanation.",
2160
- "response_preview": "Silent snow descends,\nBare branches hold the frozen air,\nSpring waits in the cold.",
2161
- "raw": "5",
2162
- "score": 5
2163
- },
2164
- {
2165
- "prompt": "Outline in one sentence why running database migrations in transactions is usually a good idea.",
2166
- "response_preview": "Running database migrations in transactions is usually a good idea because it ensures atomicity, guaranteeing that eithe",
2167
- "raw": "5",
2168
- "score": 5
2169
- },
2170
- {
2171
- "prompt": "Write a SQL query that returns the top 5 customers by total `amount` from a table `orders(customer_id, amount)`. Just the query.",
2172
- "response_preview": "```sql\nSELECT customer_id, SUM(amount) AS total_amount\nFROM orders\nGROUP BY customer_id\nORDER BY total_amount DESC\nLIMIT",
2173
- "raw": "5",
2174
- "score": 5
2175
- },
2176
- {
2177
- "prompt": "Concisely: what does a cache do in a CPU?",
2178
- "response_preview": "A CPU cache is a small, ultra-fast memory located directly on the processor that stores frequently accessed data and ins",
2179
- "raw": "5",
2180
- "score": 5
2181
- },
2182
- {
2183
- "prompt": "Demonstrate a simple Bash loop that prints numbers 1 through 5, one per line.",
2184
- "response_preview": "Here is a simple Bash loop that prints numbers from 1 to 5:\n\n```bash\nfor i in {1..5}; do\n echo \"$i\"\ndone\n```\n\n**How i",
2185
- "raw": "5",
2186
- "score": 5
2187
- },
2188
- {
2189
- "prompt": "When is it appropriate to ask for clarification when given an ambiguous task? One sentence.",
2190
- "response_preview": "It is appropriate to ask for clarification immediately upon recognizing ambiguity, as early intervention prevents wasted",
2191
- "raw": "5",
2192
- "score": 5
2193
- },
2194
- {
2195
- "prompt": "Write the opening two sentences of a short story set on a lighthouse during a storm.",
2196
- "response_preview": "The wind howled like a starving beast, tearing at the iron grating of the lantern room and threatening to unseat the gre",
2197
- "raw": "5",
2198
- "score": 5
2199
- },
2200
- {
2201
- "prompt": "Write a one-line Python enumerate comprehension that returns the squares of the even numbers from 0 to 20 inclusive.",
2202
- "response_preview": "`[x**2 for i, x in enumerate(range(21)) if x % 2 == 0]`",
2203
- "raw": "5",
2204
- "score": 5
2205
- },
2206
- {
2207
- "prompt": "What's the difference between correlation and causation? Provide one example in 2 sentences.",
2208
- "response_preview": "Correlation describes a statistical relationship where two variables change together, while causation indicates that one",
2209
- "raw": "5",
2210
- "score": 5
2211
- },
2212
- {
2213
- "prompt": "Why does water expand when it freezes? Offer a concise explanation.",
2214
- "response_preview": "Water expands when it freezes primarily due to the unique behavior of **hydrogen bonds** between its molecules.\n\nIn liqu",
2215
- "raw": "5",
2216
- "score": 5
2217
- }
2218
- ],
2219
- "scoring_time": 27.4,
2220
- "in_composite": true,
2221
- "version": 1
2222
- },
2223
- "chat_turns_probe": {
2224
- "n": 6,
2225
- "n_valid": 6,
2226
- "n_turns": 3,
2227
- "mean_score": 4.667,
2228
- "normalized": 0.9167,
2229
- "per_convo": [
2230
- {
2231
- "seed": "I'm planning a small dinner party for 6 people. Can you help me pick a theme?",
2232
- "raw": "3",
2233
- "score": 3
2234
- },
2235
- {
2236
- "seed": "Outline the difference between `let`, `const`, and `var` in JavaScript.",
2237
- "raw": "5",
2238
- "score": 5
2239
- },
2240
- {
2241
- "seed": "My tomato plants have yellow leaves. What could be wrong?",
2242
- "raw": "5",
2243
- "score": 5
2244
- },
2245
- {
2246
- "seed": "Offer me a simple recipe for chocolate chip cookies.",
2247
- "raw": "5",
2248
- "score": 5
2249
- },
2250
- {
2251
- "seed": "Help me debug: my Python script prints 'Hello' twice even though I only called print once.",
2252
- "raw": "5",
2253
- "score": 5
2254
- },
2255
- {
2256
- "seed": "I want to meditate but can't focus. Any practical starting advice?",
2257
- "raw": "5",
2258
- "score": 5
2259
- }
2260
- ],
2261
- "scoring_time": 10.5,
2262
- "in_composite": true,
2263
- "version": 1
2264
- }
2265
- },
2266
- "Qwen/Qwen3.5-4B": {
2267
- "status": "teacher_sanity",
2268
- "capability": {
2269
- "n": 36,
2270
- "correct": 32,
2271
- "pass_frac": 0.889,
2272
- "teacher_pass_frac": 0.889,
2273
- "items": [],
2274
- "source": "teacher_sanity_probe"
2275
- }
2276
- }
2277
- },
2278
- "timings": {
2279
- "vllm_startup": 96.7,
2280
- "vllm_generation": 7.5,
2281
- "teacher_probe_refs": 8.1,
2282
- "teacher_logits_pass": 0.0,
2283
- "teacher_hf_load": 0.0,
2284
- "on_policy_rkl": 5.0,
2285
- "judge_probe": 27.4,
2286
- "chat_turns_probe": 10.5,
2287
- "phase_b_total": 45.9
2288
- }
2289
- }