RealFalconsAI commited on
Commit
4a7a2a8
·
verified ·
1 Parent(s): 5dc57bb

Delete falcondec_report.json

Browse files
Files changed (1) hide show
  1. falcondec_report.json +0 -900
falcondec_report.json DELETED
@@ -1,900 +0,0 @@
1
- {
2
- "version": "1.0.0",
3
- "notebook_version": "V2",
4
- "mode": "scratch",
5
- "parent_version": null,
6
- "lineage": [
7
- "jhu-clsp/ettin-encoder-150m"
8
- ],
9
- "config": {
10
- "MODE": "scratch",
11
- "FINETUNE_FROM": "latest",
12
- "MODEL_NAME": "FalconDec",
13
- "BACKBONE": "jhu-clsp/ettin-encoder-150m",
14
- "BACKBONE_INIT": "pretrained",
15
- "PRESET": "standard",
16
- "CUSTOM_DATA_JSONL": "",
17
- "CUSTOM_WEIGHT": 3.0,
18
- "INCLUDE_BUILDERS": [],
19
- "EXCLUDE_BUILDERS": [
20
- "mind2web"
21
- ],
22
- "MAX_LEN": 512,
23
- "LONG_MAX_LEN": 2048,
24
- "LONG_OPTS_THRESHOLD": 24,
25
- "HEAD_MAX_LEN": 192,
26
- "MAX_TOK_PER_OPT": 24,
27
- "MAX_OPTS_SINGLE_PASS": 96,
28
- "INTERACT_LAYERS": 2,
29
- "INTERACT_HEADS": 8,
30
- "BATCH_SIZE": 32,
31
- "TOKENS_PER_BATCH": 16384,
32
- "GRAD_ACCUM": 1,
33
- "LR_ENCODER": 4e-05,
34
- "LR_HEAD": 0.0003,
35
- "LLRD": 0.9,
36
- "WEIGHT_DECAY": 0.01,
37
- "WARMUP_FRAC": 0.06,
38
- "EPOCHS": 0,
39
- "EMA_DECAY": 0.999,
40
- "GRAD_CLIP": 1.0,
41
- "FINETUNE_LR_SCALE": 0.5,
42
- "SPHERICAL_W": 0.5,
43
- "BRIER_W": 0.0,
44
- "RPS_W": 1.0,
45
- "TASK_SAMPLING_ALPHA": 0.5,
46
- "NOTA_PROB": 0.08,
47
- "RLCD_EPOCHS": 0.0,
48
- "RLCD_SAMPLES": 4,
49
- "RLCD_SIGMA": 0.3,
50
- "RLCD_CE_ANCHOR": 0.5,
51
- "DEVICE": "cuda",
52
- "SEED": 42,
53
- "NUM_WORKERS": 0,
54
- "USE_COMPILE": false,
55
- "ATTN_IMPL": "auto",
56
- "OUTPUT_ROOT": "./falcondec_runs",
57
- "EXPORT_INT8": true,
58
- "SIZE_TARGET_MB": 250,
59
- "DEFER_THRESHOLD": 0.7,
60
- "EVAL_BASELINE_PROOF_V2": true,
61
- "BASELINE_MAX_PER_TASK": 120,
62
- "RUN_CPU_BENCH": true,
63
- "PUSH_TO_HUB": false,
64
- "HUB_REPO": ""
65
- },
66
- "epochs": 2,
67
- "best_epoch": 2,
68
- "rlcd_kept": false,
69
- "history": [
70
- {
71
- "epoch": 1,
72
- "train_loss": 0.8680336219771599,
73
- "train_acc": 0.6543994836099589,
74
- "val_macro": 0.731116463037862,
75
- "val_micro": 0.7382111347733495,
76
- "val_nll": 0.5654821096448589
77
- },
78
- {
79
- "epoch": 2,
80
- "train_loss": 0.5398828714239906,
81
- "train_acc": 0.8023448871064317,
82
- "val_macro": 0.7628908710347914,
83
- "val_micro": 0.7745664739884393,
84
- "val_nll": 0.5195555807120061
85
- }
86
- ],
87
- "train_minutes": 46.12855076789856,
88
- "temperatures": [
89
- [
90
- 1.7067416906356812,
91
- 1.3524072170257568,
92
- 1.4660307168960571,
93
- 1.3491928577423096
94
- ],
95
- [
96
- 1.4017767906188965,
97
- 1.4017767906188965,
98
- 1.4017767906188965,
99
- 1.4017767906188965
100
- ],
101
- [
102
- 1.4534460306167603,
103
- 1.4534460306167603,
104
- 1.4534460306167603,
105
- 1.4534460306167603
106
- ]
107
- ],
108
- "data_counts": {
109
- "train": 155747,
110
- "validation": 13148,
111
- "test": 17498
112
- },
113
- "skipped_builders": [],
114
- "test_overall": {
115
- "n": 17498.0,
116
- "micro_acc": 0.7249399931420734,
117
- "macro_acc": 0.7251494212975877,
118
- "nll": 0.6524463170584336,
119
- "brier": 0.35774529214224665,
120
- "ece": 0.024511970451033767,
121
- "aurc": 0.09680548556403747,
122
- "score_mae": 0.5717749670736633,
123
- "coverage@0.7": 0.5635501200137158,
124
- "acc_on_covered@0.7": 0.8963593955988236
125
- },
126
- "test_per_domain": {
127
- "agentic": 0.7535174993567496,
128
- "classification": 0.5841666666666667,
129
- "code": 0.8642874336991984,
130
- "guardrails": 0.7681783499031459,
131
- "intents": 0.7733333333333333,
132
- "policy": 0.6685185185185185,
133
- "reasoning": 0.6665164519404285,
134
- "support": 0.9966666666666666,
135
- "workflows": 0.689
136
- },
137
- "test_per_task": [
138
- {
139
- "task": "ag_news/topic",
140
- "n": 300,
141
- "acc": 0.8466666666666667,
142
- "chance": 0.25,
143
- "nll": 0.4110426408642282,
144
- "brier": 0.22030779087576163,
145
- "heldout": false,
146
- "domain": "classification",
147
- "ece": 0.05068659414847695,
148
- "proof_v2 (card)": NaN,
149
- "\u0394 vs proof_v2": NaN
150
- },
151
- {
152
- "task": "agentharm/refuse",
153
- "n": 416,
154
- "acc": 0.6538461538461539,
155
- "chance": 0.5,
156
- "nll": 0.6880831477453915,
157
- "brier": 0.4656698587049597,
158
- "heldout": true,
159
- "domain": "guardrails",
160
- "ece": 0.17823480728727117,
161
- "proof_v2 (card)": NaN,
162
- "\u0394 vs proof_v2": NaN
163
- },
164
- {
165
- "task": "anli/nli",
166
- "n": 300,
167
- "acc": 0.3933333333333333,
168
- "chance": 0.3333333333333333,
169
- "nll": 1.203708649550875,
170
- "brier": 0.7341421357408877,
171
- "heldout": false,
172
- "domain": "reasoning",
173
- "ece": 0.1847125640511513,
174
- "proof_v2 (card)": NaN,
175
- "\u0394 vs proof_v2": NaN
176
- },
177
- {
178
- "task": "aqua_rat/math",
179
- "n": 247,
180
- "acc": 0.2591093117408907,
181
- "chance": 0.2,
182
- "nll": 1.6161003136924403,
183
- "brier": 0.8032518029612505,
184
- "heldout": false,
185
- "domain": "reasoning",
186
- "ece": 0.0712899305077217,
187
- "proof_v2 (card)": NaN,
188
- "\u0394 vs proof_v2": NaN
189
- },
190
- {
191
- "task": "arc_challenge/mcq",
192
- "n": 300,
193
- "acc": 0.42333333333333334,
194
- "chance": 0.25,
195
- "nll": 1.2702170145014922,
196
- "brier": 0.690553698672778,
197
- "heldout": true,
198
- "domain": "reasoning",
199
- "ece": 0.0853065178791682,
200
- "proof_v2 (card)": 0.308,
201
- "\u0394 vs proof_v2": 0.11533333333333334
202
- },
203
- {
204
- "task": "arc_easy/mcq",
205
- "n": 300,
206
- "acc": 0.5533333333333333,
207
- "chance": 0.24966666666666668,
208
- "nll": 1.0077791693791127,
209
- "brier": 0.5445073279657635,
210
- "heldout": true,
211
- "domain": "reasoning",
212
- "ece": 0.05727233131726582,
213
- "proof_v2 (card)": 0.425,
214
- "\u0394 vs proof_v2": 0.12833333333333335
215
- },
216
- {
217
- "task": "banking77/intent",
218
- "n": 300,
219
- "acc": 0.9233333333333333,
220
- "chance": 0.3171111111111111,
221
- "nll": 0.23318792015245587,
222
- "brier": 0.11510283627855257,
223
- "heldout": true,
224
- "domain": "intents",
225
- "ece": 0.03749855528275173,
226
- "proof_v2 (card)": 0.883,
227
- "\u0394 vs proof_v2": 0.04033333333333333
228
- },
229
- {
230
- "task": "banking77/intent_77",
231
- "n": 300,
232
- "acc": 0.47,
233
- "chance": 0.012987012987012986,
234
- "nll": 2.278919235393405,
235
- "brier": 0.7254735049280885,
236
- "heldout": true,
237
- "domain": "intents",
238
- "ece": 0.20010005796949074,
239
- "proof_v2 (card)": NaN,
240
- "\u0394 vs proof_v2": NaN
241
- },
242
- {
243
- "task": "bigclonebench/clone",
244
- "n": 300,
245
- "acc": 0.8633333333333333,
246
- "chance": 0.5,
247
- "nll": 0.37696734445169566,
248
- "brier": 0.22472162603134804,
249
- "heldout": false,
250
- "domain": "code",
251
- "ece": 0.09757265766461692,
252
- "proof_v2 (card)": 0.383,
253
- "\u0394 vs proof_v2": 0.4803333333333333
254
- },
255
- {
256
- "task": "bitext/category",
257
- "n": 300,
258
- "acc": 0.9933333333333333,
259
- "chance": 0.17221560846560846,
260
- "nll": 0.012192320638529661,
261
- "brier": 0.0060866762539747805,
262
- "heldout": false,
263
- "domain": "support",
264
- "ece": 0.008061723212401064,
265
- "proof_v2 (card)": NaN,
266
- "\u0394 vs proof_v2": NaN
267
- },
268
- {
269
- "task": "bitext/route",
270
- "n": 300,
271
- "acc": 1.0,
272
- "chance": 0.19970466570466572,
273
- "nll": 0.002716773402629163,
274
- "brier": 0.000754853699874741,
275
- "heldout": false,
276
- "domain": "support",
277
- "ece": 0.0024759014447530356,
278
- "proof_v2 (card)": 0.958,
279
- "\u0394 vs proof_v2": 0.04200000000000004
280
- },
281
- {
282
- "task": "boolq/yes_no",
283
- "n": 300,
284
- "acc": 0.7833333333333333,
285
- "chance": 0.5,
286
- "nll": 0.5142986152755717,
287
- "brier": 0.33372815804815115,
288
- "heldout": false,
289
- "domain": "reasoning",
290
- "ece": 0.07064317325750984,
291
- "proof_v2 (card)": 0.717,
292
- "\u0394 vs proof_v2": 0.06633333333333336
293
- },
294
- {
295
- "task": "civil_comments/toxic",
296
- "n": 300,
297
- "acc": 0.8066666666666666,
298
- "chance": 0.5,
299
- "nll": 0.4095402888363848,
300
- "brier": 0.262189661845685,
301
- "heldout": false,
302
- "domain": "guardrails",
303
- "ece": 0.05080057859420778,
304
- "proof_v2 (card)": NaN,
305
- "\u0394 vs proof_v2": NaN
306
- },
307
- {
308
- "task": "clinc150/intent",
309
- "n": 300,
310
- "acc": 0.7933333333333333,
311
- "chance": 0.12188708599454368,
312
- "nll": 0.700936573866705,
313
- "brier": 0.2769989036180038,
314
- "heldout": false,
315
- "domain": "intents",
316
- "ece": 0.08069658771157259,
317
- "proof_v2 (card)": 0.85,
318
- "\u0394 vs proof_v2": -0.05666666666666664
319
- },
320
- {
321
- "task": "codexglue/code_to_doc",
322
- "n": 300,
323
- "acc": 0.9766666666666667,
324
- "chance": 0.2563333333333333,
325
- "nll": 0.04927768305185358,
326
- "brier": 0.0304432559207717,
327
- "heldout": false,
328
- "domain": "code",
329
- "ece": 0.016067063510417912,
330
- "proof_v2 (card)": 0.969,
331
- "\u0394 vs proof_v2": 0.0076666666666667105
332
- },
333
- {
334
- "task": "codexglue/doc_to_code",
335
- "n": 300,
336
- "acc": 0.9733333333333334,
337
- "chance": 0.27416666666666667,
338
- "nll": 0.07633179273092537,
339
- "brier": 0.03791739648635504,
340
- "heldout": false,
341
- "domain": "code",
342
- "ece": 0.021707421441872998,
343
- "proof_v2 (card)": 0.961,
344
- "\u0394 vs proof_v2": 0.012333333333333418
345
- },
346
- {
347
- "task": "codexglue/func_name",
348
- "n": 275,
349
- "acc": 0.9381818181818182,
350
- "chance": 0.2627878787878788,
351
- "nll": 0.17262806540259837,
352
- "brier": 0.08739935554630737,
353
- "heldout": false,
354
- "domain": "code",
355
- "ece": 0.021064434268257853,
356
- "proof_v2 (card)": 0.901,
357
- "\u0394 vs proof_v2": 0.037181818181818205
358
- },
359
- {
360
- "task": "codexglue/lang_id",
361
- "n": 300,
362
- "acc": 1.0,
363
- "chance": 0.23505555555555555,
364
- "nll": 0.004893879278594492,
365
- "brier": 0.0018880184673689814,
366
- "heldout": false,
367
- "domain": "code",
368
- "ece": 0.004261451959609995,
369
- "proof_v2 (card)": 0.997,
370
- "\u0394 vs proof_v2": 0.0030000000000000027
371
- },
372
- {
373
- "task": "commonsense_qa/mcq",
374
- "n": 296,
375
- "acc": 0.6114864864864865,
376
- "chance": 0.19999999999999998,
377
- "nll": 0.9976526975466137,
378
- "brier": 0.5097789520615463,
379
- "heldout": false,
380
- "domain": "reasoning",
381
- "ece": 0.058095945800478374,
382
- "proof_v2 (card)": 0.442,
383
- "\u0394 vs proof_v2": 0.1694864864864865
384
- },
385
- {
386
- "task": "counsel/critique_quality",
387
- "n": 201,
388
- "acc": 0.5572139303482587,
389
- "chance": 0.3333333333333333,
390
- "nll": 0.947643574197494,
391
- "brier": 0.5637741453958017,
392
- "heldout": false,
393
- "domain": "agentic",
394
- "ece": 0.09350576225797928,
395
- "proof_v2 (card)": NaN,
396
- "\u0394 vs proof_v2": NaN
397
- },
398
- {
399
- "task": "counsel/step_has_error",
400
- "n": 201,
401
- "acc": 0.7910447761194029,
402
- "chance": 0.5,
403
- "nll": 0.48369112497413375,
404
- "brier": 0.3008154577122218,
405
- "heldout": false,
406
- "domain": "agentic",
407
- "ece": 0.0929471070493632,
408
- "proof_v2 (card)": NaN,
409
- "\u0394 vs proof_v2": NaN
410
- },
411
- {
412
- "task": "devign/vulnerability",
413
- "n": 300,
414
- "acc": 0.5533333333333333,
415
- "chance": 0.5,
416
- "nll": 0.6878520819544792,
417
- "brier": 0.49477992541143456,
418
- "heldout": false,
419
- "domain": "code",
420
- "ece": 0.026096095244089852,
421
- "proof_v2 (card)": 0.542,
422
- "\u0394 vs proof_v2": 0.011333333333333306
423
- },
424
- {
425
- "task": "emotion/6way",
426
- "n": 300,
427
- "acc": 0.48,
428
- "chance": 0.16666666666666666,
429
- "nll": 1.29548427477479,
430
- "brier": 0.6518899203560358,
431
- "heldout": true,
432
- "domain": "classification",
433
- "ece": 0.04883478035529455,
434
- "proof_v2 (card)": NaN,
435
- "\u0394 vs proof_v2": NaN
436
- },
437
- {
438
- "task": "gsm8k/math",
439
- "n": 300,
440
- "acc": 0.6366666666666667,
441
- "chance": 0.25,
442
- "nll": 0.8585682193624476,
443
- "brier": 0.4891388351483126,
444
- "heldout": false,
445
- "domain": "reasoning",
446
- "ece": 0.05008313804864885,
447
- "proof_v2 (card)": 0.275,
448
- "\u0394 vs proof_v2": 0.3616666666666667
449
- },
450
- {
451
- "task": "hellaswag/continuation",
452
- "n": 300,
453
- "acc": 0.38333333333333336,
454
- "chance": 0.25,
455
- "nll": 1.4718584705268343,
456
- "brier": 0.7740508552747831,
457
- "heldout": false,
458
- "domain": "reasoning",
459
- "ece": 0.17061606506506602,
460
- "proof_v2 (card)": NaN,
461
- "\u0394 vs proof_v2": NaN
462
- },
463
- {
464
- "task": "hotpotqa/comparison_yes_no",
465
- "n": 17,
466
- "acc": 0.8235294117647058,
467
- "chance": 0.5,
468
- "nll": 0.38011277554666295,
469
- "brier": 0.2316522594238636,
470
- "heldout": false,
471
- "domain": "agentic",
472
- "ece": 0.1851463878855986,
473
- "proof_v2 (card)": NaN,
474
- "\u0394 vs proof_v2": NaN
475
- },
476
- {
477
- "task": "hotpotqa/retrieve",
478
- "n": 298,
479
- "acc": 0.8422818791946308,
480
- "chance": 0.16762410780867157,
481
- "nll": 0.5295110404664876,
482
- "brier": 0.2438292407339771,
483
- "heldout": false,
484
- "domain": "agentic",
485
- "ece": 0.07827200480555527,
486
- "proof_v2 (card)": NaN,
487
- "\u0394 vs proof_v2": NaN
488
- },
489
- {
490
- "task": "humaneval/completion",
491
- "n": 119,
492
- "acc": 0.7563025210084033,
493
- "chance": 0.39355742296918783,
494
- "nll": 0.6370426528781903,
495
- "brier": 0.3969878365389353,
496
- "heldout": true,
497
- "domain": "code",
498
- "ece": 0.15178211521701646,
499
- "proof_v2 (card)": 0.575,
500
- "\u0394 vs proof_v2": 0.18130252100840338
501
- },
502
- {
503
- "task": "jailbreak/detect",
504
- "n": 262,
505
- "acc": 0.9656488549618321,
506
- "chance": 0.5,
507
- "nll": 0.08600628703664003,
508
- "brier": 0.04878715579412294,
509
- "heldout": false,
510
- "domain": "guardrails",
511
- "ece": 0.01832273079238785,
512
- "proof_v2 (card)": NaN,
513
- "\u0394 vs proof_v2": NaN
514
- },
515
- {
516
- "task": "massive_en/intent",
517
- "n": 300,
518
- "acc": 0.9066666666666666,
519
- "chance": 0.12198158507543068,
520
- "nll": 0.3105201367917834,
521
- "brier": 0.14457025344814517,
522
- "heldout": false,
523
- "domain": "intents",
524
- "ece": 0.046070452282826156,
525
- "proof_v2 (card)": NaN,
526
- "\u0394 vs proof_v2": NaN
527
- },
528
- {
529
- "task": "mbpp/bugspot",
530
- "n": 156,
531
- "acc": 0.7307692307692307,
532
- "chance": 0.41346153846153844,
533
- "nll": 0.5468092209865483,
534
- "brier": 0.36555076664459357,
535
- "heldout": false,
536
- "domain": "code",
537
- "ece": 0.08602869434234424,
538
- "proof_v2 (card)": 0.475,
539
- "\u0394 vs proof_v2": 0.25576923076923075
540
- },
541
- {
542
- "task": "mbpp/solution",
543
- "n": 300,
544
- "acc": 0.9866666666666667,
545
- "chance": 0.25,
546
- "nll": 0.039979492869557966,
547
- "brier": 0.019202444049521386,
548
- "heldout": false,
549
- "domain": "code",
550
- "ece": 0.014268100659052535,
551
- "proof_v2 (card)": 0.992,
552
- "\u0394 vs proof_v2": -0.005333333333333301
553
- },
554
- {
555
- "task": "mmlu/mcq",
556
- "n": 300,
557
- "acc": 0.3933333333333333,
558
- "chance": 0.25,
559
- "nll": 1.3269313552727302,
560
- "brier": 0.7154630336922222,
561
- "heldout": true,
562
- "domain": "reasoning",
563
- "ece": 0.0942566188176473,
564
- "proof_v2 (card)": NaN,
565
- "\u0394 vs proof_v2": NaN
566
- },
567
- {
568
- "task": "mnli/claim",
569
- "n": 300,
570
- "acc": 0.8066666666666666,
571
- "chance": 0.3333333333333333,
572
- "nll": 0.49059042249961443,
573
- "brier": 0.2750930787976646,
574
- "heldout": false,
575
- "domain": "reasoning",
576
- "ece": 0.07203083296616872,
577
- "proof_v2 (card)": 0.492,
578
- "\u0394 vs proof_v2": 0.31466666666666665
579
- },
580
- {
581
- "task": "openbookqa/mcq",
582
- "n": 300,
583
- "acc": 0.55,
584
- "chance": 0.25,
585
- "nll": 1.0744397868712743,
586
- "brier": 0.5694824859262314,
587
- "heldout": false,
588
- "domain": "reasoning",
589
- "ece": 0.06834218502044676,
590
- "proof_v2 (card)": 0.292,
591
- "\u0394 vs proof_v2": 0.25800000000000006
592
- },
593
- {
594
- "task": "policy/access_control_transfer",
595
- "n": 300,
596
- "acc": 1.0,
597
- "chance": 0.3333333333333333,
598
- "nll": 0.001206557510455847,
599
- "brier": 1.9713182036255528e-05,
600
- "heldout": false,
601
- "domain": "policy",
602
- "ece": 0.0012009245157241777,
603
- "proof_v2 (card)": NaN,
604
- "\u0394 vs proof_v2": NaN
605
- },
606
- {
607
- "task": "policy/count_threshold_transfer",
608
- "n": 300,
609
- "acc": 0.5233333333333333,
610
- "chance": 0.1789126984126984,
611
- "nll": 1.0874939475084344,
612
- "brier": 0.5886331220896244,
613
- "heldout": false,
614
- "domain": "policy",
615
- "ece": 0.08493562882145246,
616
- "proof_v2 (card)": NaN,
617
- "\u0394 vs proof_v2": NaN
618
- },
619
- {
620
- "task": "policy/free_shipping_transfer",
621
- "n": 300,
622
- "acc": 0.85,
623
- "chance": 0.5,
624
- "nll": 0.32513019680637323,
625
- "brier": 0.21239506877440237,
626
- "heldout": false,
627
- "domain": "policy",
628
- "ece": 0.06672031641006476,
629
- "proof_v2 (card)": NaN,
630
- "\u0394 vs proof_v2": NaN
631
- },
632
- {
633
- "task": "policy/invoice_overdue_transfer",
634
- "n": 300,
635
- "acc": 0.4066666666666667,
636
- "chance": 0.5,
637
- "nll": 0.986925971681873,
638
- "brier": 0.7342384205663931,
639
- "heldout": false,
640
- "domain": "policy",
641
- "ece": 0.36398835837841026,
642
- "proof_v2 (card)": NaN,
643
- "\u0394 vs proof_v2": NaN
644
- },
645
- {
646
- "task": "policy/invoice_total_transfer",
647
- "n": 300,
648
- "acc": 0.5233333333333333,
649
- "chance": 0.5,
650
- "nll": 0.6931114653746288,
651
- "brier": 0.4999460506940947,
652
- "heldout": false,
653
- "domain": "policy",
654
- "ece": 0.020025810400644932,
655
- "proof_v2 (card)": NaN,
656
- "\u0394 vs proof_v2": NaN
657
- },
658
- {
659
- "task": "policy/refund_approval_transfer",
660
- "n": 300,
661
- "acc": 0.71,
662
- "chance": 0.3333333333333333,
663
- "nll": 1.0124957078392618,
664
- "brier": 0.5042078571413049,
665
- "heldout": false,
666
- "domain": "policy",
667
- "ece": 0.2186933029691378,
668
- "proof_v2 (card)": NaN,
669
- "\u0394 vs proof_v2": NaN
670
- },
671
- {
672
- "task": "policy/return_window_transfer",
673
- "n": 300,
674
- "acc": 1.0,
675
- "chance": 0.3333333333333333,
676
- "nll": 0.020440605763991093,
677
- "brier": 0.004461685693040299,
678
- "heldout": false,
679
- "domain": "policy",
680
- "ece": 0.019133159518241915,
681
- "proof_v2 (card)": NaN,
682
- "\u0394 vs proof_v2": NaN
683
- },
684
- {
685
- "task": "policy/sla_urgency_transfer",
686
- "n": 300,
687
- "acc": 0.7133333333333334,
688
- "chance": 0.25,
689
- "nll": 1.1532283271973331,
690
- "brier": 0.5171943386039836,
691
- "heldout": false,
692
- "domain": "policy",
693
- "ece": 0.21417010804017383,
694
- "proof_v2 (card)": NaN,
695
- "\u0394 vs proof_v2": NaN
696
- },
697
- {
698
- "task": "policy/table_extreme_transfer",
699
- "n": 300,
700
- "acc": 0.29,
701
- "chance": 0.23983333333333334,
702
- "nll": 1.456577208439509,
703
- "brier": 0.7586071370001425,
704
- "heldout": false,
705
- "domain": "policy",
706
- "ece": 0.03645655060807866,
707
- "proof_v2 (card)": NaN,
708
- "\u0394 vs proof_v2": NaN
709
- },
710
- {
711
- "task": "prompt_injections/detect",
712
- "n": 116,
713
- "acc": 0.646551724137931,
714
- "chance": 0.5,
715
- "nll": 1.053943488778996,
716
- "brier": 0.6090117318454916,
717
- "heldout": true,
718
- "domain": "guardrails",
719
- "ece": 0.2722245978898016,
720
- "proof_v2 (card)": NaN,
721
- "\u0394 vs proof_v2": NaN
722
- },
723
- {
724
- "task": "qasc/mcq",
725
- "n": 300,
726
- "acc": 0.9833333333333333,
727
- "chance": 0.125,
728
- "nll": 0.05723339068814918,
729
- "brier": 0.026824159004342575,
730
- "heldout": false,
731
- "domain": "reasoning",
732
- "ece": 0.006817154586315196,
733
- "proof_v2 (card)": NaN,
734
- "\u0394 vs proof_v2": NaN
735
- },
736
- {
737
- "task": "sciq/mcq",
738
- "n": 300,
739
- "acc": 0.95,
740
- "chance": 0.25,
741
- "nll": 0.11919435005791457,
742
- "brier": 0.06139228769287704,
743
- "heldout": false,
744
- "domain": "reasoning",
745
- "ece": 0.022936838169892584,
746
- "proof_v2 (card)": 0.692,
747
- "\u0394 vs proof_v2": 0.258
748
- },
749
- {
750
- "task": "scitail/support",
751
- "n": 300,
752
- "acc": 0.9566666666666667,
753
- "chance": 0.5,
754
- "nll": 0.1297914838930592,
755
- "brier": 0.06999127560996722,
756
- "heldout": false,
757
- "domain": "reasoning",
758
- "ece": 0.030403847098350548,
759
- "proof_v2 (card)": NaN,
760
- "\u0394 vs proof_v2": NaN
761
- },
762
- {
763
- "task": "snli/contradicts",
764
- "n": 300,
765
- "acc": 0.9666666666666667,
766
- "chance": 0.3333333333333333,
767
- "nll": 0.09257454760700057,
768
- "brier": 0.051637771503957354,
769
- "heldout": false,
770
- "domain": "reasoning",
771
- "ece": 0.03507927517096204,
772
- "proof_v2 (card)": 0.892,
773
- "\u0394 vs proof_v2": 0.07466666666666666
774
- },
775
- {
776
- "task": "snli/must_be_true",
777
- "n": 300,
778
- "acc": 0.97,
779
- "chance": 0.3333333333333333,
780
- "nll": 0.0936215610774525,
781
- "brier": 0.052880273017698834,
782
- "heldout": false,
783
- "domain": "reasoning",
784
- "ece": 0.029101738135019904,
785
- "proof_v2 (card)": 0.908,
786
- "\u0394 vs proof_v2": 0.061999999999999944
787
- },
788
- {
789
- "task": "snli/nli",
790
- "n": 594,
791
- "acc": 0.8367003367003367,
792
- "chance": 0.3333333333333333,
793
- "nll": 0.4089164471626909,
794
- "brier": 0.23002962736941795,
795
- "heldout": false,
796
- "domain": "reasoning",
797
- "ece": 0.0345901768115233,
798
- "proof_v2 (card)": NaN,
799
- "\u0394 vs proof_v2": NaN
800
- },
801
- {
802
- "task": "sst5/score",
803
- "n": 300,
804
- "acc": 0.37,
805
- "chance": 0.19999999999999998,
806
- "nll": 1.3181062419215839,
807
- "brier": 0.7127016977274074,
808
- "heldout": true,
809
- "domain": "classification",
810
- "ece": 0.08605750997861225,
811
- "proof_v2 (card)": NaN,
812
- "\u0394 vs proof_v2": NaN
813
- },
814
- {
815
- "task": "typed_decisions/agent_trace_observability",
816
- "n": 500,
817
- "acc": 0.696,
818
- "chance": 0.3,
819
- "nll": 0.7389556468054652,
820
- "brier": 0.4281462582665551,
821
- "heldout": false,
822
- "domain": "workflows",
823
- "ece": 0.10615407174825665,
824
- "proof_v2 (card)": NaN,
825
- "\u0394 vs proof_v2": NaN
826
- },
827
- {
828
- "task": "typed_decisions/customer_service",
829
- "n": 500,
830
- "acc": 0.72,
831
- "chance": 0.28,
832
- "nll": 0.6959334943462163,
833
- "brier": 0.3951960258685939,
834
- "heldout": false,
835
- "domain": "workflows",
836
- "ece": 0.09901113402843474,
837
- "proof_v2 (card)": NaN,
838
- "\u0394 vs proof_v2": NaN
839
- },
840
- {
841
- "task": "typed_decisions/invoice_processing",
842
- "n": 500,
843
- "acc": 0.628,
844
- "chance": 0.35,
845
- "nll": 0.8374121308475733,
846
- "brier": 0.49094015332838314,
847
- "heldout": false,
848
- "domain": "workflows",
849
- "ece": 0.10137240266799925,
850
- "proof_v2 (card)": NaN,
851
- "\u0394 vs proof_v2": NaN
852
- },
853
- {
854
- "task": "typed_decisions/security_incidents",
855
- "n": 500,
856
- "acc": 0.712,
857
- "chance": 0.34,
858
- "nll": 0.7715732423365116,
859
- "brier": 0.44351636558882973,
860
- "heldout": false,
861
- "domain": "workflows",
862
- "ece": 0.14024459588527682,
863
- "proof_v2 (card)": NaN,
864
- "\u0394 vs proof_v2": NaN
865
- },
866
- {
867
- "task": "winogrande/blank",
868
- "n": 300,
869
- "acc": 0.54,
870
- "chance": 0.5,
871
- "nll": 0.702468536173304,
872
- "brier": 0.5075546648189806,
873
- "heldout": false,
874
- "domain": "reasoning",
875
- "ece": 0.08909334023793541,
876
- "proof_v2 (card)": NaN,
877
- "\u0394 vs proof_v2": NaN
878
- },
879
- {
880
- "task": "yelp/score",
881
- "n": 300,
882
- "acc": 0.64,
883
- "chance": 0.19999999999999998,
884
- "nll": 0.8608952843522032,
885
- "brier": 0.4893681077847031,
886
- "heldout": false,
887
- "domain": "classification",
888
- "ece": 0.06891959726810454,
889
- "proof_v2 (card)": NaN,
890
- "\u0394 vs proof_v2": NaN
891
- }
892
- ],
893
- "head_to_head": null,
894
- "env": {
895
- "torch": "2.11.0+cu128",
896
- "transformers": "5.17.0",
897
- "python": "3.14.4",
898
- "gpu": "NVIDIA GeForce RTX 5090 Laptop GPU"
899
- }
900
- }