bogdanraduta commited on
Commit
48cd36c
·
verified ·
1 Parent(s): f8eaa36

Upload eval/injection_eval.json with huggingface_hub

Browse files
Files changed (1) hide show
  1. eval/injection_eval.json +527 -507
eval/injection_eval.json CHANGED
@@ -1,296 +1,296 @@
1
  {
2
  "task": "injection",
3
- "threshold": 0.19,
4
  "per_language": {
5
  "az": {
6
- "support": 14,
7
  "precision": 1.0,
8
- "recall": 0.9286,
9
- "f1": 0.963,
10
  "fpr": 0.0,
11
- "accuracy": 0.8636,
12
- "n": 44
13
  },
14
  "bg": {
15
- "support": 13,
16
  "precision": 1.0,
17
  "recall": 1.0,
18
  "f1": 1.0,
19
  "fpr": 0.0,
20
- "accuracy": 0.8372,
21
- "n": 43
22
  },
23
  "cs": {
24
- "support": 13,
25
- "precision": 1.0,
26
- "recall": 0.9231,
27
- "f1": 0.96,
28
- "fpr": 0.0,
29
- "accuracy": 0.8372,
30
- "n": 43
31
  },
32
  "da": {
33
- "support": 14,
34
- "precision": 0.9333,
35
  "recall": 1.0,
36
- "f1": 0.9655,
37
- "fpr": 0.0333,
38
- "accuracy": 0.8409,
39
- "n": 44
40
  },
41
  "de": {
42
- "support": 14,
43
- "precision": 0.9333,
44
  "recall": 1.0,
45
- "f1": 0.9655,
46
- "fpr": 0.0333,
47
- "accuracy": 0.7955,
48
- "n": 44
49
  },
50
  "el": {
51
- "support": 14,
52
  "precision": 1.0,
53
  "recall": 1.0,
54
  "f1": 1.0,
55
  "fpr": 0.0,
56
- "accuracy": 0.8409,
57
- "n": 44
58
  },
59
  "en": {
60
- "support": 14,
61
- "precision": 0.9333,
62
  "recall": 1.0,
63
- "f1": 0.9655,
64
- "fpr": 0.0333,
65
- "accuracy": 0.8409,
66
- "n": 44
67
  },
68
  "es": {
69
- "support": 12,
70
  "precision": 1.0,
71
  "recall": 1.0,
72
  "f1": 1.0,
73
  "fpr": 0.0,
74
- "accuracy": 0.881,
75
- "n": 42
76
  },
77
  "et": {
78
- "support": 14,
79
- "precision": 0.9333,
80
  "recall": 1.0,
81
- "f1": 0.9655,
82
- "fpr": 0.0333,
83
- "accuracy": 0.8636,
84
- "n": 44
85
  },
86
  "fi": {
87
- "support": 14,
88
- "precision": 0.9333,
89
  "recall": 1.0,
90
- "f1": 0.9655,
91
- "fpr": 0.0333,
92
- "accuracy": 0.8864,
93
- "n": 44
94
  },
95
  "fr": {
96
- "support": 13,
97
- "precision": 0.9231,
98
- "recall": 0.9231,
99
- "f1": 0.9231,
100
- "fpr": 0.0333,
101
- "accuracy": 0.8605,
102
- "n": 43
103
- },
104
- "ga": {
105
- "support": 14,
106
  "precision": 1.0,
107
- "recall": 1.0,
108
- "f1": 1.0,
109
  "fpr": 0.0,
110
- "accuracy": 0.9091,
111
- "n": 44
 
 
 
 
 
 
 
 
 
112
  },
113
  "hr": {
114
- "support": 13,
115
- "precision": 1.0,
116
  "recall": 1.0,
117
- "f1": 1.0,
118
- "fpr": 0.0,
119
- "accuracy": 0.8605,
120
- "n": 43
121
  },
122
  "hu": {
123
- "support": 14,
124
- "precision": 1.0,
125
  "recall": 1.0,
126
- "f1": 1.0,
127
- "fpr": 0.0,
128
- "accuracy": 0.8864,
129
- "n": 44
130
  },
131
  "it": {
132
- "support": 14,
133
  "precision": 1.0,
134
  "recall": 1.0,
135
  "f1": 1.0,
136
  "fpr": 0.0,
137
- "accuracy": 0.8409,
138
- "n": 44
139
  },
140
  "lt": {
141
- "support": 13,
142
  "precision": 1.0,
143
  "recall": 1.0,
144
  "f1": 1.0,
145
  "fpr": 0.0,
146
- "accuracy": 0.8605,
147
- "n": 43
148
  },
149
  "lv": {
150
- "support": 14,
151
- "precision": 1.0,
152
  "recall": 1.0,
153
- "f1": 1.0,
154
- "fpr": 0.0,
155
- "accuracy": 0.8864,
156
- "n": 44
157
  },
158
  "mt": {
159
- "support": 14,
160
- "precision": 1.0,
161
- "recall": 0.5714,
162
- "f1": 0.7273,
163
- "fpr": 0.0,
164
- "accuracy": 0.7955,
165
- "n": 44
166
  },
167
  "nl": {
168
- "support": 14,
169
- "precision": 1.0,
170
  "recall": 1.0,
171
- "f1": 1.0,
172
- "fpr": 0.0,
173
- "accuracy": 0.8864,
174
- "n": 44
175
  },
176
  "pl": {
177
- "support": 14,
178
  "precision": 1.0,
179
  "recall": 1.0,
180
  "f1": 1.0,
181
  "fpr": 0.0,
182
- "accuracy": 0.8409,
183
- "n": 44
184
  },
185
  "pt": {
186
- "support": 14,
187
- "precision": 0.875,
188
  "recall": 1.0,
189
- "f1": 0.9333,
190
- "fpr": 0.0667,
191
- "accuracy": 0.7273,
192
- "n": 44
193
  },
194
  "ro": {
195
- "support": 14,
196
- "precision": 0.9333,
197
  "recall": 1.0,
198
- "f1": 0.9655,
199
- "fpr": 0.0333,
200
- "accuracy": 0.8864,
201
- "n": 44
202
  },
203
  "sk": {
204
- "support": 14,
205
  "precision": 1.0,
206
- "recall": 0.9286,
207
- "f1": 0.963,
208
  "fpr": 0.0,
209
- "accuracy": 0.7955,
210
- "n": 44
211
  },
212
  "sl": {
213
- "support": 14,
214
- "precision": 1.0,
215
  "recall": 1.0,
216
- "f1": 1.0,
217
- "fpr": 0.0,
218
- "accuracy": 0.9091,
219
- "n": 44
220
  },
221
  "sv": {
222
- "support": 14,
223
  "precision": 1.0,
224
  "recall": 1.0,
225
  "f1": 1.0,
226
  "fpr": 0.0,
227
- "accuracy": 0.8864,
228
- "n": 44
229
  },
230
  "tr": {
231
- "support": 14,
232
- "precision": 0.9333,
233
  "recall": 1.0,
234
- "f1": 0.9655,
235
- "fpr": 0.0333,
236
- "accuracy": 0.8864,
237
- "n": 44
238
  }
239
  },
240
  "per_register": {
241
  "encoded_payload": {
242
- "support": 50,
243
  "precision": 1.0,
244
- "recall": 0.98,
245
- "f1": 0.9899,
246
  "fpr": 0.0,
247
- "accuracy": 0.78,
248
- "n": 50
249
  },
250
  "hypothetical_framing": {
251
- "support": 52,
252
  "precision": 1.0,
253
- "recall": 0.9423,
254
- "f1": 0.9703,
255
  "fpr": 0.0,
256
- "accuracy": 0.8077,
257
- "n": 52
258
  },
259
  "ignore_instructions": {
260
- "support": 51,
261
  "precision": 1.0,
262
- "recall": 0.9804,
263
- "f1": 0.9901,
264
  "fpr": 0.0,
265
- "accuracy": 0.098,
266
- "n": 51
267
  },
268
  "instruction_in_document": {
269
- "support": 52,
270
  "precision": 1.0,
271
  "recall": 1.0,
272
  "f1": 1.0,
273
  "fpr": 0.0,
274
- "accuracy": 0.75,
275
- "n": 52
276
  },
277
  "instruction_in_tool_output": {
278
- "support": 48,
279
  "precision": 1.0,
280
- "recall": 1.0,
281
- "f1": 1.0,
282
  "fpr": 0.0,
283
- "accuracy": 0.8958,
284
- "n": 48
285
  },
286
  "meta_question": {
287
  "support": 0,
288
  "precision": 0.0,
289
  "recall": 0.0,
290
  "f1": 0.0,
291
- "fpr": 0.0288,
292
- "accuracy": 0.9712,
293
- "n": 104
294
  },
295
  "mundane_informational": {
296
  "support": 0,
@@ -299,115 +299,133 @@
299
  "f1": 0.0,
300
  "fpr": 0.0,
301
  "accuracy": 1.0,
302
- "n": 52
303
  },
304
  "mundane_operational": {
305
  "support": 0,
306
  "precision": 0.0,
307
  "recall": 0.0,
308
  "f1": 0.0,
309
- "fpr": 0.0,
310
- "accuracy": 1.0,
311
- "n": 52
312
  },
313
  "mundane_transactional": {
314
  "support": 0,
315
  "precision": 0.0,
316
  "recall": 0.0,
317
  "f1": 0.0,
318
- "fpr": 0.0,
319
- "accuracy": 1.0,
320
- "n": 52
321
  },
322
  "ordinary_instruction": {
323
  "support": 0,
324
  "precision": 0.0,
325
  "recall": 0.0,
326
  "f1": 0.0,
327
- "fpr": 0.0,
328
- "accuracy": 1.0,
329
- "n": 104
330
  },
331
  "ordinary_question": {
332
  "support": 0,
333
  "precision": 0.0,
334
  "recall": 0.0,
335
  "f1": 0.0,
336
- "fpr": 0.0,
337
- "accuracy": 1.0,
338
- "n": 104
339
  },
340
  "persona_override": {
341
- "support": 52,
342
  "precision": 1.0,
343
- "recall": 0.9615,
344
- "f1": 0.9804,
345
  "fpr": 0.0,
346
- "accuracy": 0.3077,
347
- "n": 52
348
  },
349
  "quoted_attack": {
350
  "support": 0,
351
  "precision": 0.0,
352
  "recall": 0.0,
353
  "f1": 0.0,
354
- "fpr": 0.0192,
355
- "accuracy": 0.9808,
356
- "n": 104
357
  },
358
  "roleplay_benign": {
359
  "support": 0,
360
  "precision": 0.0,
361
  "recall": 0.0,
362
  "f1": 0.0,
363
- "fpr": 0.0192,
364
- "accuracy": 0.9808,
365
- "n": 104
366
  },
367
  "security_discussion": {
368
  "support": 0,
369
  "precision": 0.0,
370
  "recall": 0.0,
371
  "f1": 0.0,
372
- "fpr": 0.0288,
373
- "accuracy": 0.9712,
374
- "n": 104
375
  },
376
  "system_prompt_extraction": {
377
- "support": 52,
378
  "precision": 1.0,
379
- "recall": 0.9423,
380
- "f1": 0.9703,
381
  "fpr": 0.0,
382
- "accuracy": 0.3269,
383
- "n": 52
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
384
  }
385
  },
386
  "per_label": {
387
  "direct_injection": {
388
- "support": 153,
389
- "precision": 0.7374,
390
- "recall": 0.9542,
391
- "f1": 0.8319,
392
- "fpr": 0.0528,
393
  "accuracy": 0.0,
394
  "n": 0
395
  },
396
  "indirect_injection": {
397
- "support": 100,
398
- "precision": 0.825,
399
- "recall": 0.99,
400
- "f1": 0.9,
401
- "fpr": 0.0203,
402
  "accuracy": 0.0,
403
  "n": 0
404
  },
405
  "jailbreak": {
406
- "support": 154,
407
- "precision": 0.6042,
408
- "recall": 0.9416,
409
- "f1": 0.736,
410
- "fpr": 0.0966,
411
  "accuracy": 0.0,
412
  "n": 0
413
  }
@@ -415,599 +433,601 @@
415
  "worst_languages": [
416
  [
417
  "mt",
418
- 0.7272727272727273
419
  ],
420
  [
421
- "fr",
422
- 0.9230769230769231
423
  ],
424
  [
425
- "pt",
426
- 0.9333333333333333
427
  ],
428
  [
429
- "cs",
430
- 0.9600000000000001
431
  ],
432
  [
433
- "az",
434
- 0.962962962962963
435
  ]
436
  ],
437
  "calibration": {
438
- "threshold": 0.19,
439
  "objective": "recall_at_fpr",
440
- "value_on_val": 0.9608,
441
  "default_threshold": 0.5,
442
- "default_value_on_val": 0.9384,
443
- "improvement": 0.0224,
444
  "chosen_on": "validation split, never test",
445
  "curve": [
446
  {
447
  "threshold": 0.02,
448
- "macro_f1": 0.435,
449
- "any_label_recall": 0.9888,
450
- "any_label_fpr": 0.6077
451
  },
452
  {
453
  "threshold": 0.03,
454
- "macro_f1": 0.5467,
455
- "any_label_recall": 0.9748,
456
- "any_label_fpr": 0.0436
457
  },
458
  {
459
  "threshold": 0.04,
460
- "macro_f1": 0.558,
461
- "any_label_recall": 0.9664,
462
- "any_label_fpr": 0.0321
463
  },
464
  {
465
  "threshold": 0.05,
466
- "macro_f1": 0.5721,
467
- "any_label_recall": 0.9664,
468
- "any_label_fpr": 0.0231
469
  },
470
  {
471
  "threshold": 0.06,
472
- "macro_f1": 0.591,
473
- "any_label_recall": 0.9664,
474
- "any_label_fpr": 0.0179
475
  },
476
  {
477
  "threshold": 0.07,
478
- "macro_f1": 0.6103,
479
- "any_label_recall": 0.9664,
480
- "any_label_fpr": 0.0154
481
  },
482
  {
483
  "threshold": 0.08,
484
- "macro_f1": 0.6343,
485
- "any_label_recall": 0.9664,
486
- "any_label_fpr": 0.0141
487
  },
488
  {
489
  "threshold": 0.09,
490
- "macro_f1": 0.6544,
491
- "any_label_recall": 0.9664,
492
- "any_label_fpr": 0.0141
493
  },
494
  {
495
  "threshold": 0.1,
496
- "macro_f1": 0.6686,
497
- "any_label_recall": 0.9664,
498
- "any_label_fpr": 0.0141
499
  },
500
  {
501
  "threshold": 0.11,
502
- "macro_f1": 0.6878,
503
- "any_label_recall": 0.9664,
504
- "any_label_fpr": 0.0141
505
  },
506
  {
507
  "threshold": 0.12,
508
- "macro_f1": 0.7066,
509
- "any_label_recall": 0.9664,
510
- "any_label_fpr": 0.0141
511
  },
512
  {
513
  "threshold": 0.13,
514
- "macro_f1": 0.7318,
515
- "any_label_recall": 0.9664,
516
- "any_label_fpr": 0.0128
517
  },
518
  {
519
  "threshold": 0.14,
520
- "macro_f1": 0.7457,
521
- "any_label_recall": 0.9636,
522
- "any_label_fpr": 0.0128
523
  },
524
  {
525
  "threshold": 0.15,
526
- "macro_f1": 0.7677,
527
- "any_label_recall": 0.9636,
528
- "any_label_fpr": 0.0115
529
  },
530
  {
531
  "threshold": 0.16,
532
- "macro_f1": 0.7895,
533
- "any_label_recall": 0.9608,
534
- "any_label_fpr": 0.0115
535
  },
536
  {
537
  "threshold": 0.17,
538
- "macro_f1": 0.808,
539
- "any_label_recall": 0.9608,
540
- "any_label_fpr": 0.0103
541
  },
542
  {
543
  "threshold": 0.18,
544
- "macro_f1": 0.8166,
545
- "any_label_recall": 0.9608,
546
- "any_label_fpr": 0.0103
547
  },
548
  {
549
  "threshold": 0.19,
550
- "macro_f1": 0.8209,
551
- "any_label_recall": 0.9608,
552
- "any_label_fpr": 0.009
553
  },
554
  {
555
  "threshold": 0.2,
556
- "macro_f1": 0.8322,
557
- "any_label_recall": 0.9608,
558
- "any_label_fpr": 0.009
559
  },
560
  {
561
  "threshold": 0.21,
562
- "macro_f1": 0.8355,
563
- "any_label_recall": 0.958,
564
- "any_label_fpr": 0.009
565
  },
566
  {
567
  "threshold": 0.22,
568
- "macro_f1": 0.8431,
569
- "any_label_recall": 0.958,
570
- "any_label_fpr": 0.009
571
  },
572
  {
573
  "threshold": 0.23,
574
- "macro_f1": 0.8467,
575
- "any_label_recall": 0.9552,
576
- "any_label_fpr": 0.009
577
  },
578
  {
579
  "threshold": 0.24,
580
- "macro_f1": 0.8538,
581
- "any_label_recall": 0.9524,
582
- "any_label_fpr": 0.0077
583
  },
584
  {
585
  "threshold": 0.25,
586
- "macro_f1": 0.8593,
587
- "any_label_recall": 0.9524,
588
- "any_label_fpr": 0.0077
589
  },
590
  {
591
  "threshold": 0.26,
592
- "macro_f1": 0.8618,
593
- "any_label_recall": 0.9524,
594
- "any_label_fpr": 0.0064
595
  },
596
  {
597
  "threshold": 0.27,
598
- "macro_f1": 0.8664,
599
- "any_label_recall": 0.9524,
600
- "any_label_fpr": 0.0064
601
  },
602
  {
603
  "threshold": 0.28,
604
- "macro_f1": 0.8679,
605
- "any_label_recall": 0.9524,
606
- "any_label_fpr": 0.0064
607
  },
608
  {
609
  "threshold": 0.29,
610
- "macro_f1": 0.8689,
611
- "any_label_recall": 0.9468,
612
- "any_label_fpr": 0.0064
613
  },
614
  {
615
  "threshold": 0.3,
616
- "macro_f1": 0.8726,
617
- "any_label_recall": 0.9412,
618
- "any_label_fpr": 0.0064
619
  },
620
  {
621
  "threshold": 0.31,
622
- "macro_f1": 0.8766,
623
- "any_label_recall": 0.9412,
624
- "any_label_fpr": 0.0064
625
  },
626
  {
627
  "threshold": 0.32,
628
- "macro_f1": 0.8806,
629
- "any_label_recall": 0.9412,
630
- "any_label_fpr": 0.0064
631
  },
632
  {
633
  "threshold": 0.33,
634
- "macro_f1": 0.8865,
635
- "any_label_recall": 0.9412,
636
- "any_label_fpr": 0.0064
637
  },
638
  {
639
  "threshold": 0.34,
640
- "macro_f1": 0.8909,
641
- "any_label_recall": 0.9412,
642
- "any_label_fpr": 0.0064
643
  },
644
  {
645
  "threshold": 0.35,
646
- "macro_f1": 0.8919,
647
- "any_label_recall": 0.9412,
648
- "any_label_fpr": 0.0064
649
  },
650
  {
651
  "threshold": 0.36,
652
- "macro_f1": 0.8936,
653
- "any_label_recall": 0.9412,
654
- "any_label_fpr": 0.0064
655
  },
656
  {
657
  "threshold": 0.37,
658
- "macro_f1": 0.8961,
659
- "any_label_recall": 0.9412,
660
- "any_label_fpr": 0.0064
661
  },
662
  {
663
  "threshold": 0.38,
664
- "macro_f1": 0.8978,
665
- "any_label_recall": 0.9412,
666
- "any_label_fpr": 0.0064
667
  },
668
  {
669
  "threshold": 0.39,
670
- "macro_f1": 0.8975,
671
- "any_label_recall": 0.9412,
672
- "any_label_fpr": 0.0051
673
  },
674
  {
675
  "threshold": 0.4,
676
- "macro_f1": 0.8975,
677
- "any_label_recall": 0.9412,
678
- "any_label_fpr": 0.0051
679
  },
680
  {
681
  "threshold": 0.41,
682
- "macro_f1": 0.8993,
683
- "any_label_recall": 0.9412,
684
- "any_label_fpr": 0.0051
685
  },
686
  {
687
  "threshold": 0.42,
688
- "macro_f1": 0.9011,
689
- "any_label_recall": 0.9412,
690
- "any_label_fpr": 0.0051
691
  },
692
  {
693
  "threshold": 0.43,
694
- "macro_f1": 0.9011,
695
- "any_label_recall": 0.9412,
696
- "any_label_fpr": 0.0051
697
  },
698
  {
699
  "threshold": 0.44,
700
- "macro_f1": 0.9029,
701
- "any_label_recall": 0.9412,
702
- "any_label_fpr": 0.0051
703
  },
704
  {
705
  "threshold": 0.45,
706
- "macro_f1": 0.9048,
707
- "any_label_recall": 0.9412,
708
- "any_label_fpr": 0.0051
709
  },
710
  {
711
  "threshold": 0.46,
712
- "macro_f1": 0.9048,
713
- "any_label_recall": 0.9412,
714
- "any_label_fpr": 0.0051
715
  },
716
  {
717
  "threshold": 0.47,
718
- "macro_f1": 0.9054,
719
- "any_label_recall": 0.9384,
720
- "any_label_fpr": 0.0051
721
  },
722
  {
723
  "threshold": 0.48,
724
- "macro_f1": 0.9064,
725
- "any_label_recall": 0.9384,
726
- "any_label_fpr": 0.0051
727
  },
728
  {
729
  "threshold": 0.49,
730
- "macro_f1": 0.9064,
731
- "any_label_recall": 0.9384,
732
- "any_label_fpr": 0.0051
733
  },
734
  {
735
  "threshold": 0.5,
736
- "macro_f1": 0.9082,
737
- "any_label_recall": 0.9384,
738
- "any_label_fpr": 0.0051
739
  },
740
  {
741
  "threshold": 0.51,
742
- "macro_f1": 0.9092,
743
- "any_label_recall": 0.9384,
744
- "any_label_fpr": 0.0051
745
  },
746
  {
747
  "threshold": 0.52,
748
- "macro_f1": 0.9092,
749
- "any_label_recall": 0.9384,
750
- "any_label_fpr": 0.0051
751
  },
752
  {
753
  "threshold": 0.53,
754
- "macro_f1": 0.9092,
755
- "any_label_recall": 0.9384,
756
- "any_label_fpr": 0.0051
757
  },
758
  {
759
  "threshold": 0.54,
760
- "macro_f1": 0.9101,
761
- "any_label_recall": 0.9384,
762
- "any_label_fpr": 0.0051
763
  },
764
  {
765
  "threshold": 0.55,
766
- "macro_f1": 0.9101,
767
- "any_label_recall": 0.9384,
768
- "any_label_fpr": 0.0051
769
  },
770
  {
771
  "threshold": 0.56,
772
- "macro_f1": 0.9101,
773
- "any_label_recall": 0.9384,
774
- "any_label_fpr": 0.0051
775
  },
776
  {
777
  "threshold": 0.57,
778
- "macro_f1": 0.9111,
779
- "any_label_recall": 0.9384,
780
- "any_label_fpr": 0.0051
781
  },
782
  {
783
  "threshold": 0.58,
784
- "macro_f1": 0.9111,
785
- "any_label_recall": 0.9384,
786
- "any_label_fpr": 0.0051
787
  },
788
  {
789
  "threshold": 0.59,
790
- "macro_f1": 0.9121,
791
- "any_label_recall": 0.9384,
792
- "any_label_fpr": 0.0051
793
  },
794
  {
795
  "threshold": 0.6,
796
- "macro_f1": 0.9131,
797
- "any_label_recall": 0.9384,
798
- "any_label_fpr": 0.0051
799
  },
800
  {
801
  "threshold": 0.61,
802
- "macro_f1": 0.9141,
803
- "any_label_recall": 0.9384,
804
- "any_label_fpr": 0.0051
805
  },
806
  {
807
  "threshold": 0.62,
808
- "macro_f1": 0.9139,
809
- "any_label_recall": 0.9356,
810
- "any_label_fpr": 0.0051
811
  },
812
  {
813
  "threshold": 0.63,
814
- "macro_f1": 0.9149,
815
- "any_label_recall": 0.9356,
816
- "any_label_fpr": 0.0051
817
  },
818
  {
819
  "threshold": 0.64,
820
- "macro_f1": 0.9149,
821
- "any_label_recall": 0.9356,
822
- "any_label_fpr": 0.0051
823
  },
824
  {
825
  "threshold": 0.65,
826
- "macro_f1": 0.9146,
827
- "any_label_recall": 0.9328,
828
- "any_label_fpr": 0.0051
829
  },
830
  {
831
  "threshold": 0.66,
832
- "macro_f1": 0.9146,
833
- "any_label_recall": 0.9328,
834
- "any_label_fpr": 0.0051
835
  },
836
  {
837
  "threshold": 0.67,
838
- "macro_f1": 0.9156,
839
- "any_label_recall": 0.9328,
840
- "any_label_fpr": 0.0051
841
  },
842
  {
843
  "threshold": 0.68,
844
- "macro_f1": 0.9154,
845
- "any_label_recall": 0.93,
846
- "any_label_fpr": 0.0051
847
  },
848
  {
849
  "threshold": 0.69,
850
- "macro_f1": 0.9165,
851
- "any_label_recall": 0.93,
852
- "any_label_fpr": 0.0051
853
  },
854
  {
855
  "threshold": 0.7,
856
- "macro_f1": 0.9174,
857
- "any_label_recall": 0.93,
858
- "any_label_fpr": 0.0051
859
  },
860
  {
861
  "threshold": 0.71,
862
- "macro_f1": 0.9172,
863
- "any_label_recall": 0.93,
864
- "any_label_fpr": 0.0051
865
  },
866
  {
867
  "threshold": 0.72,
868
- "macro_f1": 0.917,
869
- "any_label_recall": 0.9272,
870
- "any_label_fpr": 0.0051
871
  },
872
  {
873
  "threshold": 0.73,
874
- "macro_f1": 0.917,
875
- "any_label_recall": 0.9272,
876
- "any_label_fpr": 0.0051
877
  },
878
  {
879
  "threshold": 0.74,
880
- "macro_f1": 0.9185,
881
- "any_label_recall": 0.9272,
882
- "any_label_fpr": 0.0038
883
  },
884
  {
885
  "threshold": 0.75,
886
- "macro_f1": 0.9215,
887
- "any_label_recall": 0.9244,
888
- "any_label_fpr": 0.0038
889
  },
890
  {
891
  "threshold": 0.76,
892
- "macro_f1": 0.9225,
893
- "any_label_recall": 0.9244,
894
- "any_label_fpr": 0.0038
895
  },
896
  {
897
  "threshold": 0.77,
898
- "macro_f1": 0.9244,
899
- "any_label_recall": 0.9244,
900
- "any_label_fpr": 0.0038
901
  },
902
  {
903
  "threshold": 0.78,
904
- "macro_f1": 0.9254,
905
- "any_label_recall": 0.9216,
906
- "any_label_fpr": 0.0038
907
  },
908
  {
909
  "threshold": 0.79,
910
- "macro_f1": 0.9254,
911
- "any_label_recall": 0.9216,
912
- "any_label_fpr": 0.0038
913
  },
914
  {
915
  "threshold": 0.8,
916
- "macro_f1": 0.9246,
917
- "any_label_recall": 0.9216,
918
- "any_label_fpr": 0.0038
919
  },
920
  {
921
  "threshold": 0.81,
922
- "macro_f1": 0.9234,
923
- "any_label_recall": 0.9216,
924
- "any_label_fpr": 0.0038
925
  },
926
  {
927
  "threshold": 0.82,
928
- "macro_f1": 0.9234,
929
- "any_label_recall": 0.9216,
930
- "any_label_fpr": 0.0038
931
  },
932
  {
933
  "threshold": 0.83,
934
- "macro_f1": 0.9255,
935
- "any_label_recall": 0.9216,
936
- "any_label_fpr": 0.0038
937
  },
938
  {
939
  "threshold": 0.84,
940
- "macro_f1": 0.9275,
941
- "any_label_recall": 0.9216,
942
- "any_label_fpr": 0.0038
943
  },
944
  {
945
  "threshold": 0.85,
946
- "macro_f1": 0.9275,
947
- "any_label_recall": 0.9216,
948
- "any_label_fpr": 0.0038
949
  },
950
  {
951
  "threshold": 0.86,
952
- "macro_f1": 0.9263,
953
- "any_label_recall": 0.9188,
954
- "any_label_fpr": 0.0038
955
  },
956
  {
957
  "threshold": 0.87,
958
- "macro_f1": 0.9251,
959
- "any_label_recall": 0.916,
960
- "any_label_fpr": 0.0038
961
  },
962
  {
963
  "threshold": 0.88,
964
- "macro_f1": 0.9261,
965
- "any_label_recall": 0.916,
966
- "any_label_fpr": 0.0038
967
  },
968
  {
969
  "threshold": 0.89,
970
- "macro_f1": 0.9261,
971
- "any_label_recall": 0.916,
972
- "any_label_fpr": 0.0038
973
  },
974
  {
975
  "threshold": 0.9,
976
- "macro_f1": 0.9261,
977
- "any_label_recall": 0.916,
978
- "any_label_fpr": 0.0038
979
  },
980
  {
981
  "threshold": 0.91,
982
- "macro_f1": 0.9261,
983
- "any_label_recall": 0.916,
984
- "any_label_fpr": 0.0038
985
  },
986
  {
987
  "threshold": 0.92,
988
- "macro_f1": 0.9298,
989
- "any_label_recall": 0.916,
990
- "any_label_fpr": 0.0038
991
  },
992
  {
993
  "threshold": 0.93,
994
- "macro_f1": 0.9302,
995
- "any_label_recall": 0.9132,
996
- "any_label_fpr": 0.0026
997
  },
998
  {
999
  "threshold": 0.94,
1000
- "macro_f1": 0.9288,
1001
- "any_label_recall": 0.9048,
1002
- "any_label_fpr": 0.0026
1003
  },
1004
  {
1005
  "threshold": 0.95,
1006
- "macro_f1": 0.9206,
1007
- "any_label_recall": 0.8908,
1008
- "any_label_fpr": 0.0026
1009
  }
1010
  ],
1011
- "notes": []
 
 
1012
  }
1013
  }
 
1
  {
2
  "task": "injection",
3
+ "threshold": 0.02,
4
  "per_language": {
5
  "az": {
6
+ "support": 42,
7
  "precision": 1.0,
8
+ "recall": 1.0,
9
+ "f1": 1.0,
10
  "fpr": 0.0,
11
+ "accuracy": 0.994,
12
+ "n": 167
13
  },
14
  "bg": {
15
+ "support": 41,
16
  "precision": 1.0,
17
  "recall": 1.0,
18
  "f1": 1.0,
19
  "fpr": 0.0,
20
+ "accuracy": 0.9879,
21
+ "n": 165
22
  },
23
  "cs": {
24
+ "support": 42,
25
+ "precision": 0.9545,
26
+ "recall": 1.0,
27
+ "f1": 0.9767,
28
+ "fpr": 0.016,
29
+ "accuracy": 0.988,
30
+ "n": 167
31
  },
32
  "da": {
33
+ "support": 41,
34
+ "precision": 0.9762,
35
  "recall": 1.0,
36
+ "f1": 0.988,
37
+ "fpr": 0.008,
38
+ "accuracy": 0.988,
39
+ "n": 166
40
  },
41
  "de": {
42
+ "support": 42,
43
+ "precision": 1.0,
44
  "recall": 1.0,
45
+ "f1": 1.0,
46
+ "fpr": 0.0,
47
+ "accuracy": 1.0,
48
+ "n": 167
49
  },
50
  "el": {
51
+ "support": 42,
52
  "precision": 1.0,
53
  "recall": 1.0,
54
  "f1": 1.0,
55
  "fpr": 0.0,
56
+ "accuracy": 0.988,
57
+ "n": 166
58
  },
59
  "en": {
60
+ "support": 41,
61
+ "precision": 0.9762,
62
  "recall": 1.0,
63
+ "f1": 0.988,
64
+ "fpr": 0.008,
65
+ "accuracy": 0.988,
66
+ "n": 166
67
  },
68
  "es": {
69
+ "support": 40,
70
  "precision": 1.0,
71
  "recall": 1.0,
72
  "f1": 1.0,
73
  "fpr": 0.0,
74
+ "accuracy": 0.9939,
75
+ "n": 165
76
  },
77
  "et": {
78
+ "support": 41,
79
+ "precision": 1.0,
80
  "recall": 1.0,
81
+ "f1": 1.0,
82
+ "fpr": 0.0,
83
+ "accuracy": 0.994,
84
+ "n": 166
85
  },
86
  "fi": {
87
+ "support": 42,
88
+ "precision": 1.0,
89
  "recall": 1.0,
90
+ "f1": 1.0,
91
+ "fpr": 0.0,
92
+ "accuracy": 1.0,
93
+ "n": 167
94
  },
95
  "fr": {
96
+ "support": 41,
 
 
 
 
 
 
 
 
 
97
  "precision": 1.0,
98
+ "recall": 0.9756,
99
+ "f1": 0.9877,
100
  "fpr": 0.0,
101
+ "accuracy": 0.9759,
102
+ "n": 166
103
+ },
104
+ "ga": {
105
+ "support": 42,
106
+ "precision": 0.9762,
107
+ "recall": 0.9762,
108
+ "f1": 0.9762,
109
+ "fpr": 0.008,
110
+ "accuracy": 0.9701,
111
+ "n": 167
112
  },
113
  "hr": {
114
+ "support": 42,
115
+ "precision": 0.9545,
116
  "recall": 1.0,
117
+ "f1": 0.9767,
118
+ "fpr": 0.016,
119
+ "accuracy": 0.988,
120
+ "n": 167
121
  },
122
  "hu": {
123
+ "support": 42,
124
+ "precision": 0.9545,
125
  "recall": 1.0,
126
+ "f1": 0.9767,
127
+ "fpr": 0.016,
128
+ "accuracy": 0.976,
129
+ "n": 167
130
  },
131
  "it": {
132
+ "support": 42,
133
  "precision": 1.0,
134
  "recall": 1.0,
135
  "f1": 1.0,
136
  "fpr": 0.0,
137
+ "accuracy": 1.0,
138
+ "n": 167
139
  },
140
  "lt": {
141
+ "support": 42,
142
  "precision": 1.0,
143
  "recall": 1.0,
144
  "f1": 1.0,
145
  "fpr": 0.0,
146
+ "accuracy": 0.994,
147
+ "n": 167
148
  },
149
  "lv": {
150
+ "support": 41,
151
+ "precision": 0.9762,
152
  "recall": 1.0,
153
+ "f1": 0.988,
154
+ "fpr": 0.008,
155
+ "accuracy": 0.9819,
156
+ "n": 166
157
  },
158
  "mt": {
159
+ "support": 42,
160
+ "precision": 0.7321,
161
+ "recall": 0.9762,
162
+ "f1": 0.8367,
163
+ "fpr": 0.12,
164
+ "accuracy": 0.8802,
165
+ "n": 167
166
  },
167
  "nl": {
168
+ "support": 41,
169
+ "precision": 0.9762,
170
  "recall": 1.0,
171
+ "f1": 0.988,
172
+ "fpr": 0.008,
173
+ "accuracy": 0.994,
174
+ "n": 166
175
  },
176
  "pl": {
177
+ "support": 41,
178
  "precision": 1.0,
179
  "recall": 1.0,
180
  "f1": 1.0,
181
  "fpr": 0.0,
182
+ "accuracy": 0.994,
183
+ "n": 166
184
  },
185
  "pt": {
186
+ "support": 42,
187
+ "precision": 0.9767,
188
  "recall": 1.0,
189
+ "f1": 0.9882,
190
+ "fpr": 0.008,
191
+ "accuracy": 0.988,
192
+ "n": 167
193
  },
194
  "ro": {
195
+ "support": 42,
196
+ "precision": 0.9767,
197
  "recall": 1.0,
198
+ "f1": 0.9882,
199
+ "fpr": 0.0081,
200
+ "accuracy": 0.988,
201
+ "n": 166
202
  },
203
  "sk": {
204
+ "support": 41,
205
  "precision": 1.0,
206
+ "recall": 1.0,
207
+ "f1": 1.0,
208
  "fpr": 0.0,
209
+ "accuracy": 0.994,
210
+ "n": 166
211
  },
212
  "sl": {
213
+ "support": 42,
214
+ "precision": 0.9767,
215
  "recall": 1.0,
216
+ "f1": 0.9882,
217
+ "fpr": 0.008,
218
+ "accuracy": 0.988,
219
+ "n": 167
220
  },
221
  "sv": {
222
+ "support": 41,
223
  "precision": 1.0,
224
  "recall": 1.0,
225
  "f1": 1.0,
226
  "fpr": 0.0,
227
+ "accuracy": 0.994,
228
+ "n": 166
229
  },
230
  "tr": {
231
+ "support": 42,
232
+ "precision": 0.9545,
233
  "recall": 1.0,
234
+ "f1": 0.9767,
235
+ "fpr": 0.016,
236
+ "accuracy": 0.982,
237
+ "n": 167
238
  }
239
  },
240
  "per_register": {
241
  "encoded_payload": {
242
+ "support": 155,
243
  "precision": 1.0,
244
+ "recall": 0.9935,
245
+ "f1": 0.9968,
246
  "fpr": 0.0,
247
+ "accuracy": 0.9742,
248
+ "n": 155
249
  },
250
  "hypothetical_framing": {
251
+ "support": 156,
252
  "precision": 1.0,
253
+ "recall": 1.0,
254
+ "f1": 1.0,
255
  "fpr": 0.0,
256
+ "accuracy": 0.9872,
257
+ "n": 156
258
  },
259
  "ignore_instructions": {
260
+ "support": 155,
261
  "precision": 1.0,
262
+ "recall": 1.0,
263
+ "f1": 1.0,
264
  "fpr": 0.0,
265
+ "accuracy": 0.9484,
266
+ "n": 155
267
  },
268
  "instruction_in_document": {
269
+ "support": 155,
270
  "precision": 1.0,
271
  "recall": 1.0,
272
  "f1": 1.0,
273
  "fpr": 0.0,
274
+ "accuracy": 0.9613,
275
+ "n": 155
276
  },
277
  "instruction_in_tool_output": {
278
+ "support": 148,
279
  "precision": 1.0,
280
+ "recall": 0.9865,
281
+ "f1": 0.9932,
282
  "fpr": 0.0,
283
+ "accuracy": 0.9865,
284
+ "n": 148
285
  },
286
  "meta_question": {
287
  "support": 0,
288
  "precision": 0.0,
289
  "recall": 0.0,
290
  "f1": 0.0,
291
+ "fpr": 0.0148,
292
+ "accuracy": 0.9852,
293
+ "n": 338
294
  },
295
  "mundane_informational": {
296
  "support": 0,
 
299
  "f1": 0.0,
300
  "fpr": 0.0,
301
  "accuracy": 1.0,
302
+ "n": 182
303
  },
304
  "mundane_operational": {
305
  "support": 0,
306
  "precision": 0.0,
307
  "recall": 0.0,
308
  "f1": 0.0,
309
+ "fpr": 0.0055,
310
+ "accuracy": 0.9945,
311
+ "n": 182
312
  },
313
  "mundane_transactional": {
314
  "support": 0,
315
  "precision": 0.0,
316
  "recall": 0.0,
317
  "f1": 0.0,
318
+ "fpr": 0.0055,
319
+ "accuracy": 0.9945,
320
+ "n": 182
321
  },
322
  "ordinary_instruction": {
323
  "support": 0,
324
  "precision": 0.0,
325
  "recall": 0.0,
326
  "f1": 0.0,
327
+ "fpr": 0.0148,
328
+ "accuracy": 0.9852,
329
+ "n": 338
330
  },
331
  "ordinary_question": {
332
  "support": 0,
333
  "precision": 0.0,
334
  "recall": 0.0,
335
  "f1": 0.0,
336
+ "fpr": 0.003,
337
+ "accuracy": 0.997,
338
+ "n": 338
339
  },
340
  "persona_override": {
341
+ "support": 155,
342
  "precision": 1.0,
343
+ "recall": 1.0,
344
+ "f1": 1.0,
345
  "fpr": 0.0,
346
+ "accuracy": 0.9226,
347
+ "n": 155
348
  },
349
  "quoted_attack": {
350
  "support": 0,
351
  "precision": 0.0,
352
  "recall": 0.0,
353
  "f1": 0.0,
354
+ "fpr": 0.0178,
355
+ "accuracy": 0.9822,
356
+ "n": 338
357
  },
358
  "roleplay_benign": {
359
  "support": 0,
360
  "precision": 0.0,
361
  "recall": 0.0,
362
  "f1": 0.0,
363
+ "fpr": 0.0148,
364
+ "accuracy": 0.9852,
365
+ "n": 338
366
  },
367
  "security_discussion": {
368
  "support": 0,
369
  "precision": 0.0,
370
  "recall": 0.0,
371
  "f1": 0.0,
372
+ "fpr": 0.0148,
373
+ "accuracy": 0.9852,
374
+ "n": 338
375
  },
376
  "system_prompt_extraction": {
377
+ "support": 156,
378
  "precision": 1.0,
379
+ "recall": 1.0,
380
+ "f1": 1.0,
381
  "fpr": 0.0,
382
+ "accuracy": 1.0,
383
+ "n": 156
384
+ },
385
+ "technical_identifiers": {
386
+ "support": 0,
387
+ "precision": 0.0,
388
+ "recall": 0.0,
389
+ "f1": 0.0,
390
+ "fpr": 0.003,
391
+ "accuracy": 0.997,
392
+ "n": 336
393
+ },
394
+ "technical_payload": {
395
+ "support": 0,
396
+ "precision": 0.0,
397
+ "recall": 0.0,
398
+ "f1": 0.0,
399
+ "fpr": 0.003,
400
+ "accuracy": 0.997,
401
+ "n": 337
402
  }
403
  },
404
  "per_label": {
405
  "direct_injection": {
406
+ "support": 466,
407
+ "precision": 0.9546,
408
+ "recall": 0.9936,
409
+ "f1": 0.9737,
410
+ "fpr": 0.0057,
411
  "accuracy": 0.0,
412
  "n": 0
413
  },
414
  "indirect_injection": {
415
+ "support": 303,
416
+ "precision": 0.9495,
417
+ "recall": 0.9934,
418
+ "f1": 0.971,
419
+ "fpr": 0.004,
420
  "accuracy": 0.0,
421
  "n": 0
422
  },
423
  "jailbreak": {
424
+ "support": 466,
425
+ "precision": 0.9331,
426
+ "recall": 0.9871,
427
+ "f1": 0.9593,
428
+ "fpr": 0.0085,
429
  "accuracy": 0.0,
430
  "n": 0
431
  }
 
433
  "worst_languages": [
434
  [
435
  "mt",
436
+ 0.836734693877551
437
  ],
438
  [
439
+ "ga",
440
+ 0.9761904761904762
441
  ],
442
  [
443
+ "cs",
444
+ 0.9767441860465117
445
  ],
446
  [
447
+ "hr",
448
+ 0.9767441860465117
449
  ],
450
  [
451
+ "hu",
452
+ 0.9767441860465117
453
  ]
454
  ],
455
  "calibration": {
456
+ "threshold": 0.02,
457
  "objective": "recall_at_fpr",
458
+ "value_on_val": 0.9926,
459
  "default_threshold": 0.5,
460
+ "default_value_on_val": 0.9852,
461
+ "improvement": 0.0074,
462
  "chosen_on": "validation split, never test",
463
  "curve": [
464
  {
465
  "threshold": 0.02,
466
+ "macro_f1": 0.959,
467
+ "any_label_recall": 0.9926,
468
+ "any_label_fpr": 0.0089
469
  },
470
  {
471
  "threshold": 0.03,
472
+ "macro_f1": 0.9659,
473
+ "any_label_recall": 0.9907,
474
+ "any_label_fpr": 0.0071
475
  },
476
  {
477
  "threshold": 0.04,
478
+ "macro_f1": 0.9677,
479
+ "any_label_recall": 0.9898,
480
+ "any_label_fpr": 0.0062
481
  },
482
  {
483
  "threshold": 0.05,
484
+ "macro_f1": 0.9687,
485
+ "any_label_recall": 0.9898,
486
+ "any_label_fpr": 0.0059
487
  },
488
  {
489
  "threshold": 0.06,
490
+ "macro_f1": 0.9687,
491
+ "any_label_recall": 0.9898,
492
+ "any_label_fpr": 0.0059
493
  },
494
  {
495
  "threshold": 0.07,
496
+ "macro_f1": 0.9692,
497
+ "any_label_recall": 0.9898,
498
+ "any_label_fpr": 0.0055
499
  },
500
  {
501
  "threshold": 0.08,
502
+ "macro_f1": 0.9688,
503
+ "any_label_recall": 0.9898,
504
+ "any_label_fpr": 0.0052
505
  },
506
  {
507
  "threshold": 0.09,
508
+ "macro_f1": 0.9692,
509
+ "any_label_recall": 0.9889,
510
+ "any_label_fpr": 0.0052
511
  },
512
  {
513
  "threshold": 0.1,
514
+ "macro_f1": 0.9684,
515
+ "any_label_recall": 0.988,
516
+ "any_label_fpr": 0.0049
517
  },
518
  {
519
  "threshold": 0.11,
520
+ "macro_f1": 0.968,
521
+ "any_label_recall": 0.987,
522
+ "any_label_fpr": 0.0049
523
  },
524
  {
525
  "threshold": 0.12,
526
+ "macro_f1": 0.9686,
527
+ "any_label_recall": 0.987,
528
+ "any_label_fpr": 0.0049
529
  },
530
  {
531
  "threshold": 0.13,
532
+ "macro_f1": 0.9692,
533
+ "any_label_recall": 0.987,
534
+ "any_label_fpr": 0.0043
535
  },
536
  {
537
  "threshold": 0.14,
538
+ "macro_f1": 0.9696,
539
+ "any_label_recall": 0.987,
540
+ "any_label_fpr": 0.0043
541
  },
542
  {
543
  "threshold": 0.15,
544
+ "macro_f1": 0.9699,
545
+ "any_label_recall": 0.987,
546
+ "any_label_fpr": 0.004
547
  },
548
  {
549
  "threshold": 0.16,
550
+ "macro_f1": 0.9703,
551
+ "any_label_recall": 0.987,
552
+ "any_label_fpr": 0.004
553
  },
554
  {
555
  "threshold": 0.17,
556
+ "macro_f1": 0.9703,
557
+ "any_label_recall": 0.987,
558
+ "any_label_fpr": 0.004
559
  },
560
  {
561
  "threshold": 0.18,
562
+ "macro_f1": 0.9703,
563
+ "any_label_recall": 0.987,
564
+ "any_label_fpr": 0.004
565
  },
566
  {
567
  "threshold": 0.19,
568
+ "macro_f1": 0.9713,
569
+ "any_label_recall": 0.987,
570
+ "any_label_fpr": 0.004
571
  },
572
  {
573
  "threshold": 0.2,
574
+ "macro_f1": 0.9713,
575
+ "any_label_recall": 0.987,
576
+ "any_label_fpr": 0.004
577
  },
578
  {
579
  "threshold": 0.21,
580
+ "macro_f1": 0.9717,
581
+ "any_label_recall": 0.987,
582
+ "any_label_fpr": 0.0037
583
  },
584
  {
585
  "threshold": 0.22,
586
+ "macro_f1": 0.9717,
587
+ "any_label_recall": 0.987,
588
+ "any_label_fpr": 0.0037
589
  },
590
  {
591
  "threshold": 0.23,
592
+ "macro_f1": 0.9717,
593
+ "any_label_recall": 0.987,
594
+ "any_label_fpr": 0.0037
595
  },
596
  {
597
  "threshold": 0.24,
598
+ "macro_f1": 0.9717,
599
+ "any_label_recall": 0.987,
600
+ "any_label_fpr": 0.0037
601
  },
602
  {
603
  "threshold": 0.25,
604
+ "macro_f1": 0.9717,
605
+ "any_label_recall": 0.987,
606
+ "any_label_fpr": 0.0037
607
  },
608
  {
609
  "threshold": 0.26,
610
+ "macro_f1": 0.972,
611
+ "any_label_recall": 0.987,
612
+ "any_label_fpr": 0.0037
613
  },
614
  {
615
  "threshold": 0.27,
616
+ "macro_f1": 0.972,
617
+ "any_label_recall": 0.987,
618
+ "any_label_fpr": 0.0037
619
  },
620
  {
621
  "threshold": 0.28,
622
+ "macro_f1": 0.9716,
623
+ "any_label_recall": 0.9861,
624
+ "any_label_fpr": 0.0037
625
  },
626
  {
627
  "threshold": 0.29,
628
+ "macro_f1": 0.9716,
629
+ "any_label_recall": 0.9861,
630
+ "any_label_fpr": 0.0037
631
  },
632
  {
633
  "threshold": 0.3,
634
+ "macro_f1": 0.9722,
635
+ "any_label_recall": 0.9861,
636
+ "any_label_fpr": 0.0034
637
  },
638
  {
639
  "threshold": 0.31,
640
+ "macro_f1": 0.9725,
641
+ "any_label_recall": 0.9861,
642
+ "any_label_fpr": 0.0034
643
  },
644
  {
645
  "threshold": 0.32,
646
+ "macro_f1": 0.9725,
647
+ "any_label_recall": 0.9861,
648
+ "any_label_fpr": 0.0034
649
  },
650
  {
651
  "threshold": 0.33,
652
+ "macro_f1": 0.9725,
653
+ "any_label_recall": 0.9861,
654
+ "any_label_fpr": 0.0034
655
  },
656
  {
657
  "threshold": 0.34,
658
+ "macro_f1": 0.9731,
659
+ "any_label_recall": 0.9861,
660
+ "any_label_fpr": 0.0031
661
  },
662
  {
663
  "threshold": 0.35,
664
+ "macro_f1": 0.9734,
665
+ "any_label_recall": 0.9861,
666
+ "any_label_fpr": 0.0031
667
  },
668
  {
669
  "threshold": 0.36,
670
+ "macro_f1": 0.9728,
671
+ "any_label_recall": 0.9852,
672
+ "any_label_fpr": 0.0031
673
  },
674
  {
675
  "threshold": 0.37,
676
+ "macro_f1": 0.9728,
677
+ "any_label_recall": 0.9852,
678
+ "any_label_fpr": 0.0031
679
  },
680
  {
681
  "threshold": 0.38,
682
+ "macro_f1": 0.9728,
683
+ "any_label_recall": 0.9852,
684
+ "any_label_fpr": 0.0031
685
  },
686
  {
687
  "threshold": 0.39,
688
+ "macro_f1": 0.9728,
689
+ "any_label_recall": 0.9852,
690
+ "any_label_fpr": 0.0031
691
  },
692
  {
693
  "threshold": 0.4,
694
+ "macro_f1": 0.9728,
695
+ "any_label_recall": 0.9852,
696
+ "any_label_fpr": 0.0031
697
  },
698
  {
699
  "threshold": 0.41,
700
+ "macro_f1": 0.9732,
701
+ "any_label_recall": 0.9852,
702
+ "any_label_fpr": 0.0028
703
  },
704
  {
705
  "threshold": 0.42,
706
+ "macro_f1": 0.9732,
707
+ "any_label_recall": 0.9852,
708
+ "any_label_fpr": 0.0028
709
  },
710
  {
711
  "threshold": 0.43,
712
+ "macro_f1": 0.9732,
713
+ "any_label_recall": 0.9852,
714
+ "any_label_fpr": 0.0028
715
  },
716
  {
717
  "threshold": 0.44,
718
+ "macro_f1": 0.9732,
719
+ "any_label_recall": 0.9852,
720
+ "any_label_fpr": 0.0028
721
  },
722
  {
723
  "threshold": 0.45,
724
+ "macro_f1": 0.9732,
725
+ "any_label_recall": 0.9852,
726
+ "any_label_fpr": 0.0028
727
  },
728
  {
729
  "threshold": 0.46,
730
+ "macro_f1": 0.9732,
731
+ "any_label_recall": 0.9852,
732
+ "any_label_fpr": 0.0028
733
  },
734
  {
735
  "threshold": 0.47,
736
+ "macro_f1": 0.9728,
737
+ "any_label_recall": 0.9852,
738
+ "any_label_fpr": 0.0028
739
  },
740
  {
741
  "threshold": 0.48,
742
+ "macro_f1": 0.9732,
743
+ "any_label_recall": 0.9852,
744
+ "any_label_fpr": 0.0025
745
  },
746
  {
747
  "threshold": 0.49,
748
+ "macro_f1": 0.9732,
749
+ "any_label_recall": 0.9852,
750
+ "any_label_fpr": 0.0025
751
  },
752
  {
753
  "threshold": 0.5,
754
+ "macro_f1": 0.9728,
755
+ "any_label_recall": 0.9852,
756
+ "any_label_fpr": 0.0025
757
  },
758
  {
759
  "threshold": 0.51,
760
+ "macro_f1": 0.9728,
761
+ "any_label_recall": 0.9852,
762
+ "any_label_fpr": 0.0025
763
  },
764
  {
765
  "threshold": 0.52,
766
+ "macro_f1": 0.9724,
767
+ "any_label_recall": 0.9852,
768
+ "any_label_fpr": 0.0025
769
  },
770
  {
771
  "threshold": 0.53,
772
+ "macro_f1": 0.9724,
773
+ "any_label_recall": 0.9852,
774
+ "any_label_fpr": 0.0025
775
  },
776
  {
777
  "threshold": 0.54,
778
+ "macro_f1": 0.9724,
779
+ "any_label_recall": 0.9852,
780
+ "any_label_fpr": 0.0025
781
  },
782
  {
783
  "threshold": 0.55,
784
+ "macro_f1": 0.9724,
785
+ "any_label_recall": 0.9852,
786
+ "any_label_fpr": 0.0025
787
  },
788
  {
789
  "threshold": 0.56,
790
+ "macro_f1": 0.9721,
791
+ "any_label_recall": 0.9852,
792
+ "any_label_fpr": 0.0025
793
  },
794
  {
795
  "threshold": 0.57,
796
+ "macro_f1": 0.9721,
797
+ "any_label_recall": 0.9852,
798
+ "any_label_fpr": 0.0025
799
  },
800
  {
801
  "threshold": 0.58,
802
+ "macro_f1": 0.9724,
803
+ "any_label_recall": 0.9852,
804
+ "any_label_fpr": 0.0022
805
  },
806
  {
807
  "threshold": 0.59,
808
+ "macro_f1": 0.9724,
809
+ "any_label_recall": 0.9852,
810
+ "any_label_fpr": 0.0022
811
  },
812
  {
813
  "threshold": 0.6,
814
+ "macro_f1": 0.9724,
815
+ "any_label_recall": 0.9852,
816
+ "any_label_fpr": 0.0022
817
  },
818
  {
819
  "threshold": 0.61,
820
+ "macro_f1": 0.9724,
821
+ "any_label_recall": 0.9852,
822
+ "any_label_fpr": 0.0022
823
  },
824
  {
825
  "threshold": 0.62,
826
+ "macro_f1": 0.9724,
827
+ "any_label_recall": 0.9852,
828
+ "any_label_fpr": 0.0022
829
  },
830
  {
831
  "threshold": 0.63,
832
+ "macro_f1": 0.9724,
833
+ "any_label_recall": 0.9852,
834
+ "any_label_fpr": 0.0022
835
  },
836
  {
837
  "threshold": 0.64,
838
+ "macro_f1": 0.9724,
839
+ "any_label_recall": 0.9852,
840
+ "any_label_fpr": 0.0022
841
  },
842
  {
843
  "threshold": 0.65,
844
+ "macro_f1": 0.9724,
845
+ "any_label_recall": 0.9852,
846
+ "any_label_fpr": 0.0022
847
  },
848
  {
849
  "threshold": 0.66,
850
+ "macro_f1": 0.9724,
851
+ "any_label_recall": 0.9852,
852
+ "any_label_fpr": 0.0022
853
  },
854
  {
855
  "threshold": 0.67,
856
+ "macro_f1": 0.9724,
857
+ "any_label_recall": 0.9852,
858
+ "any_label_fpr": 0.0022
859
  },
860
  {
861
  "threshold": 0.68,
862
+ "macro_f1": 0.9724,
863
+ "any_label_recall": 0.9852,
864
+ "any_label_fpr": 0.0022
865
  },
866
  {
867
  "threshold": 0.69,
868
+ "macro_f1": 0.9724,
869
+ "any_label_recall": 0.9852,
870
+ "any_label_fpr": 0.0022
871
  },
872
  {
873
  "threshold": 0.7,
874
+ "macro_f1": 0.9724,
875
+ "any_label_recall": 0.9852,
876
+ "any_label_fpr": 0.0022
877
  },
878
  {
879
  "threshold": 0.71,
880
+ "macro_f1": 0.9724,
881
+ "any_label_recall": 0.9852,
882
+ "any_label_fpr": 0.0022
883
  },
884
  {
885
  "threshold": 0.72,
886
+ "macro_f1": 0.9724,
887
+ "any_label_recall": 0.9852,
888
+ "any_label_fpr": 0.0022
889
  },
890
  {
891
  "threshold": 0.73,
892
+ "macro_f1": 0.9724,
893
+ "any_label_recall": 0.9852,
894
+ "any_label_fpr": 0.0022
895
  },
896
  {
897
  "threshold": 0.74,
898
+ "macro_f1": 0.9717,
899
+ "any_label_recall": 0.9852,
900
+ "any_label_fpr": 0.0022
901
  },
902
  {
903
  "threshold": 0.75,
904
+ "macro_f1": 0.9717,
905
+ "any_label_recall": 0.9852,
906
+ "any_label_fpr": 0.0022
907
  },
908
  {
909
  "threshold": 0.76,
910
+ "macro_f1": 0.9713,
911
+ "any_label_recall": 0.9852,
912
+ "any_label_fpr": 0.0022
913
  },
914
  {
915
  "threshold": 0.77,
916
+ "macro_f1": 0.9709,
917
+ "any_label_recall": 0.9852,
918
+ "any_label_fpr": 0.0022
919
  },
920
  {
921
  "threshold": 0.78,
922
+ "macro_f1": 0.9706,
923
+ "any_label_recall": 0.9852,
924
+ "any_label_fpr": 0.0022
925
  },
926
  {
927
  "threshold": 0.79,
928
+ "macro_f1": 0.9702,
929
+ "any_label_recall": 0.9852,
930
+ "any_label_fpr": 0.0022
931
  },
932
  {
933
  "threshold": 0.8,
934
+ "macro_f1": 0.9702,
935
+ "any_label_recall": 0.9852,
936
+ "any_label_fpr": 0.0022
937
  },
938
  {
939
  "threshold": 0.81,
940
+ "macro_f1": 0.9702,
941
+ "any_label_recall": 0.9852,
942
+ "any_label_fpr": 0.0022
943
  },
944
  {
945
  "threshold": 0.82,
946
+ "macro_f1": 0.9702,
947
+ "any_label_recall": 0.9852,
948
+ "any_label_fpr": 0.0022
949
  },
950
  {
951
  "threshold": 0.83,
952
+ "macro_f1": 0.9698,
953
+ "any_label_recall": 0.9852,
954
+ "any_label_fpr": 0.0022
955
  },
956
  {
957
  "threshold": 0.84,
958
+ "macro_f1": 0.9698,
959
+ "any_label_recall": 0.9852,
960
+ "any_label_fpr": 0.0022
961
  },
962
  {
963
  "threshold": 0.85,
964
+ "macro_f1": 0.9698,
965
+ "any_label_recall": 0.9852,
966
+ "any_label_fpr": 0.0022
967
  },
968
  {
969
  "threshold": 0.86,
970
+ "macro_f1": 0.9692,
971
+ "any_label_recall": 0.9843,
972
+ "any_label_fpr": 0.0022
973
  },
974
  {
975
  "threshold": 0.87,
976
+ "macro_f1": 0.9689,
977
+ "any_label_recall": 0.9833,
978
+ "any_label_fpr": 0.0022
979
  },
980
  {
981
  "threshold": 0.88,
982
+ "macro_f1": 0.9685,
983
+ "any_label_recall": 0.9833,
984
+ "any_label_fpr": 0.0022
985
  },
986
  {
987
  "threshold": 0.89,
988
+ "macro_f1": 0.9685,
989
+ "any_label_recall": 0.9833,
990
+ "any_label_fpr": 0.0022
991
  },
992
  {
993
  "threshold": 0.9,
994
+ "macro_f1": 0.9679,
995
+ "any_label_recall": 0.9824,
996
+ "any_label_fpr": 0.0022
997
  },
998
  {
999
  "threshold": 0.91,
1000
+ "macro_f1": 0.9679,
1001
+ "any_label_recall": 0.9824,
1002
+ "any_label_fpr": 0.0022
1003
  },
1004
  {
1005
  "threshold": 0.92,
1006
+ "macro_f1": 0.9679,
1007
+ "any_label_recall": 0.9824,
1008
+ "any_label_fpr": 0.0022
1009
  },
1010
  {
1011
  "threshold": 0.93,
1012
+ "macro_f1": 0.9672,
1013
+ "any_label_recall": 0.9815,
1014
+ "any_label_fpr": 0.0022
1015
  },
1016
  {
1017
  "threshold": 0.94,
1018
+ "macro_f1": 0.9671,
1019
+ "any_label_recall": 0.9796,
1020
+ "any_label_fpr": 0.0022
1021
  },
1022
  {
1023
  "threshold": 0.95,
1024
+ "macro_f1": 0.9671,
1025
+ "any_label_recall": 0.9796,
1026
+ "any_label_fpr": 0.0022
1027
  }
1028
  ],
1029
+ "notes": [
1030
+ "the chosen threshold 0.02 is the lowest value in the sweep, so the scores are compressed toward zero and this is underfitting rather than a calibration success. Check average precision before trusting the F1."
1031
+ ]
1032
  }
1033
  }