werty1248 commited on
Commit
d3c4def
·
verified ·
1 Parent(s): c0895cc

Add files using upload-large-folder tool

Browse files
.gitattributes CHANGED
@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
 
 
33
  *.zip filter=lfs diff=lfs merge=lfs -text
34
  *.zst filter=lfs diff=lfs merge=lfs -text
35
  *tfevents* filter=lfs diff=lfs merge=lfs -text
36
+ tokenizer.json filter=lfs diff=lfs merge=lfs -text
README.md ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ base_model: dragonkue/snowflake-arctic-embed-l-v2.0-ko
4
+ private: true
5
+ ---
6
+
7
+ # Ko-Edu Classifier 2k
8
+
9
+ Private candidate checkpoint from the final Snowflake mixed-context training run.
10
+
11
+ - Repository: `werty1248/Ko-Edu-Classifier-2k-ep4p0-step8868`
12
+ - Aggregate epoch: 4.000
13
+ - Optimizer step: 8868
14
+ - Context mixture: 512/1024/2048 optimizer-step ratio 16:4:1
15
+ - Effective batch: 256
16
+ - Score-4 training target: 10%
17
+ - Base model: `dragonkue/snowflake-arctic-embed-l-v2.0-ko` with capped-quarter fine-tuning
18
+
19
+ Public test aggregate: macro F1 0.703485, class 2/3 mean F1 0.589588, one-sigma accuracy 0.764567.
20
+
21
+ Private-OOD test aggregate: macro F1 0.717750, class 2/3 mean F1 0.577006, one-sigma accuracy 0.775228.
22
+
23
+ Raw training/test text and per-example test predictions are intentionally not included in this repository.
candidate_test_result.json ADDED
@@ -0,0 +1,797 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "global_step": 8868,
3
+ "aggregate_epoch": 4.0,
4
+ "model_path": "/ko_edu_snowflake_final_train/checkpoints/step_08868",
5
+ "validation": {
6
+ "global_step": 8868,
7
+ "aggregate_epoch": 4.0,
8
+ "content_macro_f1": 0.7691223696148852,
9
+ "content_f1_class2_3_mean": 0.7002175340967471
10
+ },
11
+ "tests": {
12
+ "public_test": {
13
+ "aggregate_equal_context_weight": {
14
+ "selection_loss": 0.5996472404267125,
15
+ "content_mae": 0.36416186395414174,
16
+ "content_macro_mae": 0.39134996560001856,
17
+ "content_exact_accuracy": 0.7229817708333334,
18
+ "content_spearman": 0.9219387450855417,
19
+ "content_gaussian_nll": -0.525285516083209,
20
+ "content_mean_predicted_variance": 0.2302397398710588,
21
+ "integrity_accuracy": 0.9256184895833334,
22
+ "integrity_macro_f1": 0.5145415702399313,
23
+ "critical_average_precision": 0.1668109006932243,
24
+ "critical_roc_auc": 0.924077158723946,
25
+ "critical_f1_at_0_5": 0.20155038759689922,
26
+ "critical_brier": 0.007695191131648721,
27
+ "final_mae": 0.3742708417087128,
28
+ "final_macro_mae": 0.40425132911669354,
29
+ "final_exact_accuracy": 0.7154134114583334,
30
+ "final_spearman": 0.9196958171483666,
31
+ "integrity_confusion_matrix": [
32
+ [
33
+ 11192,
34
+ 412,
35
+ 45
36
+ ],
37
+ [
38
+ 385,
39
+ 156,
40
+ 17
41
+ ],
42
+ [
43
+ 42,
44
+ 13,
45
+ 26
46
+ ]
47
+ ],
48
+ "content_distribution": {
49
+ "0": 3088,
50
+ "1": 2782,
51
+ "2": 2964,
52
+ "3": 2007,
53
+ "4": 1447
54
+ },
55
+ "final_distribution": {
56
+ "0": 3088,
57
+ "1": 2782,
58
+ "2": 3029,
59
+ "3": 2086,
60
+ "4": 1303
61
+ },
62
+ "content_per_class": {
63
+ "0": {
64
+ "precision": 0.9783031088082902,
65
+ "recall": 0.923006416131989,
66
+ "f1": 0.9498506524131426,
67
+ "support": 3273
68
+ },
69
+ "1": {
70
+ "precision": 0.6294033069734004,
71
+ "recall": 0.8106481481481481,
72
+ "f1": 0.7086199919061109,
73
+ "support": 2160
74
+ },
75
+ "2": {
76
+ "precision": 0.6271929824561403,
77
+ "recall": 0.6441441441441441,
78
+ "f1": 0.6355555555555555,
79
+ "support": 2886
80
+ },
81
+ "3": {
82
+ "precision": 0.5371200797209765,
83
+ "recall": 0.5502807554874937,
84
+ "f1": 0.5436207766011094,
85
+ "support": 1959
86
+ },
87
+ "4": {
88
+ "precision": 0.8120248790601244,
89
+ "recall": 0.5845771144278606,
90
+ "f1": 0.6797801562048018,
91
+ "support": 2010
92
+ }
93
+ },
94
+ "content_macro_f1": 0.7034854265361441,
95
+ "content_f1_class2_3_mean": 0.5895881660783324,
96
+ "content_mean_variance_correct": 0.16486197243390396,
97
+ "content_mean_variance_wrong": 0.4008672621717884,
98
+ "content_wrong_to_correct_variance_ratio": 2.431532610302252,
99
+ "content_variance_wrong_auroc": 0.799274570902363,
100
+ "content_variance_wrong_average_precision": 0.5623532341417358,
101
+ "content_one_sigma_accuracy": 0.7645670572916666,
102
+ "content_fixed_half_point_accuracy": 0.7229817708333334,
103
+ "content_mean_predicted_sigma": 0.4059100152568453,
104
+ "content_mean_absolute_calibration_gap": 0.23573810502832473
105
+ },
106
+ "by_length": {
107
+ "512": {
108
+ "selection_loss": 0.6475959000101105,
109
+ "content_mae": 0.403041226476617,
110
+ "content_macro_mae": 0.43819375073195327,
111
+ "content_exact_accuracy": 0.69873046875,
112
+ "content_spearman": 0.9094162283707627,
113
+ "content_gaussian_nll": -0.33558642342258216,
114
+ "content_mean_predicted_variance": 0.25569876981790673,
115
+ "integrity_accuracy": 0.92724609375,
116
+ "integrity_macro_f1": 0.5196996608761314,
117
+ "critical_average_precision": 0.16239140288737114,
118
+ "critical_roc_auc": 0.9300401409027608,
119
+ "critical_f1_at_0_5": 0.24489795918367346,
120
+ "critical_brier": 0.008444094912288811,
121
+ "final_mae": 0.4134857682716224,
122
+ "final_macro_mae": 0.4519832393323484,
123
+ "final_exact_accuracy": 0.689453125,
124
+ "final_spearman": 0.9069972446950932,
125
+ "integrity_confusion_matrix": [
126
+ [
127
+ 3741,
128
+ 122,
129
+ 20
130
+ ],
131
+ [
132
+ 133,
133
+ 46,
134
+ 7
135
+ ],
136
+ [
137
+ 13,
138
+ 3,
139
+ 11
140
+ ]
141
+ ],
142
+ "content_distribution": {
143
+ "0": 1031,
144
+ "1": 997,
145
+ "2": 1037,
146
+ "3": 666,
147
+ "4": 365
148
+ },
149
+ "final_distribution": {
150
+ "0": 1031,
151
+ "1": 997,
152
+ "2": 1058,
153
+ "3": 695,
154
+ "4": 315
155
+ },
156
+ "content_per_class": {
157
+ "0": {
158
+ "precision": 0.9747817652764307,
159
+ "recall": 0.921173235563703,
160
+ "f1": 0.9472196041470311,
161
+ "support": 1091
162
+ },
163
+ "1": {
164
+ "precision": 0.5927783350050151,
165
+ "recall": 0.8208333333333333,
166
+ "f1": 0.6884100174723354,
167
+ "support": 720
168
+ },
169
+ "2": {
170
+ "precision": 0.5756991321118612,
171
+ "recall": 0.6205821205821206,
172
+ "f1": 0.5972986493246624,
173
+ "support": 962
174
+ },
175
+ "3": {
176
+ "precision": 0.5225225225225225,
177
+ "recall": 0.5329249617151608,
178
+ "f1": 0.5276724791508719,
179
+ "support": 653
180
+ },
181
+ "4": {
182
+ "precision": 0.8794520547945206,
183
+ "recall": 0.4791044776119403,
184
+ "f1": 0.6202898550724638,
185
+ "support": 670
186
+ }
187
+ },
188
+ "content_macro_f1": 0.6761781210334729,
189
+ "content_f1_class2_3_mean": 0.5624855642377671,
190
+ "content_mean_variance_correct": 0.17687892520976323,
191
+ "content_mean_variance_wrong": 0.4385046006675881,
192
+ "content_wrong_to_correct_variance_ratio": 2.4791229376114723,
193
+ "content_variance_wrong_auroc": 0.8046483457862315,
194
+ "content_variance_wrong_average_precision": 0.5972072289789552,
195
+ "content_one_sigma_accuracy": 0.746337890625,
196
+ "content_fixed_half_point_accuracy": 0.69873046875,
197
+ "content_mean_predicted_sigma": 0.4275034727898084,
198
+ "content_mean_absolute_calibration_gap": 0.2608774770819382
199
+ },
200
+ "1024": {
201
+ "selection_loss": 0.5793806554162239,
202
+ "content_mae": 0.35502210318509775,
203
+ "content_macro_mae": 0.3809260249430456,
204
+ "content_exact_accuracy": 0.729248046875,
205
+ "content_spearman": 0.9258773309917829,
206
+ "content_gaussian_nll": -0.5862765162921657,
207
+ "content_mean_predicted_variance": 0.21921119402604106,
208
+ "integrity_accuracy": 0.931396484375,
209
+ "integrity_macro_f1": 0.5242311737226991,
210
+ "critical_average_precision": 0.20618147810728718,
211
+ "critical_roc_auc": 0.9281923850613947,
212
+ "critical_f1_at_0_5": 0.2,
213
+ "critical_brier": 0.007129219811954048,
214
+ "final_mae": 0.36424440595520335,
215
+ "final_macro_mae": 0.39263800988537184,
216
+ "final_exact_accuracy": 0.72119140625,
217
+ "final_spearman": 0.9237399815073253,
218
+ "integrity_confusion_matrix": [
219
+ [
220
+ 3758,
221
+ 112,
222
+ 13
223
+ ],
224
+ [
225
+ 133,
226
+ 48,
227
+ 5
228
+ ],
229
+ [
230
+ 14,
231
+ 4,
232
+ 9
233
+ ]
234
+ ],
235
+ "content_distribution": {
236
+ "0": 1027,
237
+ "1": 920,
238
+ "2": 989,
239
+ "3": 669,
240
+ "4": 491
241
+ },
242
+ "final_distribution": {
243
+ "0": 1027,
244
+ "1": 920,
245
+ "2": 1014,
246
+ "3": 688,
247
+ "4": 447
248
+ },
249
+ "content_per_class": {
250
+ "0": {
251
+ "precision": 0.9795520934761441,
252
+ "recall": 0.922089825847846,
253
+ "f1": 0.9499527856468366,
254
+ "support": 1091
255
+ },
256
+ "1": {
257
+ "precision": 0.6326086956521739,
258
+ "recall": 0.8083333333333333,
259
+ "f1": 0.7097560975609756,
260
+ "support": 720
261
+ },
262
+ "2": {
263
+ "precision": 0.6339737108190091,
264
+ "recall": 0.6517671517671517,
265
+ "f1": 0.6427473090722706,
266
+ "support": 962
267
+ },
268
+ "3": {
269
+ "precision": 0.5500747384155455,
270
+ "recall": 0.5635528330781011,
271
+ "f1": 0.556732223903177,
272
+ "support": 653
273
+ },
274
+ "4": {
275
+ "precision": 0.8228105906313645,
276
+ "recall": 0.6029850746268657,
277
+ "f1": 0.6959517657192076,
278
+ "support": 670
279
+ }
280
+ },
281
+ "content_macro_f1": 0.7110280363804935,
282
+ "content_f1_class2_3_mean": 0.5997397664877238,
283
+ "content_mean_variance_correct": 0.15720720139904037,
284
+ "content_mean_variance_wrong": 0.3862138324181519,
285
+ "content_wrong_to_correct_variance_ratio": 2.456718451706433,
286
+ "content_variance_wrong_auroc": 0.8013581848364252,
287
+ "content_variance_wrong_average_precision": 0.5641231749105367,
288
+ "content_one_sigma_accuracy": 0.767822265625,
289
+ "content_fixed_half_point_accuracy": 0.729248046875,
290
+ "content_mean_predicted_sigma": 0.39683885099409644,
291
+ "content_mean_absolute_calibration_gap": 0.22794024060909682
292
+ },
293
+ "2048": {
294
+ "selection_loss": 0.563851957099979,
295
+ "content_mae": 0.33442226220071036,
296
+ "content_macro_mae": 0.3549301211250569,
297
+ "content_exact_accuracy": 0.740966796875,
298
+ "content_spearman": 0.9327042422925155,
299
+ "content_gaussian_nll": -0.653993608534879,
300
+ "content_mean_predicted_variance": 0.21580925576922855,
301
+ "integrity_accuracy": 0.918212890625,
302
+ "integrity_macro_f1": 0.49500456357784506,
303
+ "critical_average_precision": 0.16431265610031154,
304
+ "critical_roc_auc": 0.9154173834685017,
305
+ "critical_f1_at_0_5": 0.15,
306
+ "critical_brier": 0.007512258670703306,
307
+ "final_mae": 0.3450823508993126,
308
+ "final_macro_mae": 0.3681327381323604,
309
+ "final_exact_accuracy": 0.735595703125,
310
+ "final_spearman": 0.9304844155402512,
311
+ "integrity_confusion_matrix": [
312
+ [
313
+ 3693,
314
+ 178,
315
+ 12
316
+ ],
317
+ [
318
+ 119,
319
+ 62,
320
+ 5
321
+ ],
322
+ [
323
+ 15,
324
+ 6,
325
+ 6
326
+ ]
327
+ ],
328
+ "content_distribution": {
329
+ "0": 1030,
330
+ "1": 865,
331
+ "2": 938,
332
+ "3": 672,
333
+ "4": 591
334
+ },
335
+ "final_distribution": {
336
+ "0": 1030,
337
+ "1": 865,
338
+ "2": 957,
339
+ "3": 703,
340
+ "4": 541
341
+ },
342
+ "content_per_class": {
343
+ "0": {
344
+ "precision": 0.9805825242718447,
345
+ "recall": 0.925756186984418,
346
+ "f1": 0.9523809523809523,
347
+ "support": 1091
348
+ },
349
+ "1": {
350
+ "precision": 0.6682080924855491,
351
+ "recall": 0.8027777777777778,
352
+ "f1": 0.7293375394321766,
353
+ "support": 720
354
+ },
355
+ "2": {
356
+ "precision": 0.6769722814498934,
357
+ "recall": 0.66008316008316,
358
+ "f1": 0.6684210526315789,
359
+ "support": 962
360
+ },
361
+ "3": {
362
+ "precision": 0.5386904761904762,
363
+ "recall": 0.554364471669219,
364
+ "f1": 0.5464150943396227,
365
+ "support": 653
366
+ },
367
+ "4": {
368
+ "precision": 0.7614213197969543,
369
+ "recall": 0.6716417910447762,
370
+ "f1": 0.7137192704203014,
371
+ "support": 670
372
+ }
373
+ },
374
+ "content_macro_f1": 0.7220547818409264,
375
+ "content_f1_class2_3_mean": 0.6074180734856007,
376
+ "content_mean_variance_correct": 0.16106371287430865,
377
+ "content_mean_variance_wrong": 0.372409371401728,
378
+ "content_wrong_to_correct_variance_ratio": 2.3121866791457233,
379
+ "content_variance_wrong_auroc": 0.7882823235671796,
380
+ "content_variance_wrong_average_precision": 0.5170919064835311,
381
+ "content_one_sigma_accuracy": 0.779541015625,
382
+ "content_fixed_half_point_accuracy": 0.740966796875,
383
+ "content_mean_predicted_sigma": 0.39338772198663097,
384
+ "content_mean_absolute_calibration_gap": 0.2183965973939393
385
+ }
386
+ }
387
+ },
388
+ "private_ood_test": {
389
+ "aggregate_equal_context_weight": {
390
+ "selection_loss": 0.5928699060725598,
391
+ "content_mae": 0.3732402183663052,
392
+ "content_macro_mae": 0.37847784761947306,
393
+ "content_exact_accuracy": 0.7152506510416666,
394
+ "content_spearman": 0.8885748177452322,
395
+ "content_gaussian_nll": -0.16548063811710678,
396
+ "content_mean_predicted_variance": 0.2825092818944354,
397
+ "integrity_accuracy": 0.9122721354166666,
398
+ "integrity_macro_f1": 0.5191581195438396,
399
+ "critical_average_precision": 0.14243176618765288,
400
+ "critical_roc_auc": 0.8837826503394979,
401
+ "critical_f1_at_0_5": 0.14414414414414414,
402
+ "critical_brier": 0.014604646909798027,
403
+ "final_mae": 0.3995507869677533,
404
+ "final_macro_mae": 0.39592509330157394,
405
+ "final_exact_accuracy": 0.7035319010416666,
406
+ "final_spearman": 0.8782610375325385,
407
+ "integrity_confusion_matrix": [
408
+ [
409
+ 10816,
410
+ 269,
411
+ 21
412
+ ],
413
+ [
414
+ 611,
415
+ 374,
416
+ 11
417
+ ],
418
+ [
419
+ 76,
420
+ 90,
421
+ 20
422
+ ]
423
+ ],
424
+ "content_distribution": {
425
+ "0": 708,
426
+ "1": 2758,
427
+ "2": 3859,
428
+ "3": 1847,
429
+ "4": 3116
430
+ },
431
+ "final_distribution": {
432
+ "0": 708,
433
+ "1": 2758,
434
+ "2": 3894,
435
+ "3": 1990,
436
+ "4": 2938
437
+ },
438
+ "content_per_class": {
439
+ "0": {
440
+ "precision": 0.8728813559322034,
441
+ "recall": 0.8174603174603174,
442
+ "f1": 0.8442622950819673,
443
+ "support": 756
444
+ },
445
+ "1": {
446
+ "precision": 0.7962291515591008,
447
+ "recall": 0.6547406082289803,
448
+ "f1": 0.7185863874345549,
449
+ "support": 3354
450
+ },
451
+ "2": {
452
+ "precision": 0.6351386369525784,
453
+ "recall": 0.7006861063464837,
454
+ "f1": 0.666304200081555,
455
+ "support": 3498
456
+ },
457
+ "3": {
458
+ "precision": 0.4672441797509475,
459
+ "recall": 0.5100472813238771,
460
+ "f1": 0.4877083922011868,
461
+ "support": 1692
462
+ },
463
+ "4": {
464
+ "precision": 0.8539794608472401,
465
+ "recall": 0.8905622489959839,
466
+ "f1": 0.8718872870249017,
467
+ "support": 2988
468
+ }
469
+ },
470
+ "content_macro_f1": 0.7177497123648331,
471
+ "content_f1_class2_3_mean": 0.5770062961413709,
472
+ "content_mean_variance_correct": 0.22792565066763823,
473
+ "content_mean_variance_wrong": 0.4196157508433695,
474
+ "content_wrong_to_correct_variance_ratio": 1.8410203047100402,
475
+ "content_variance_wrong_auroc": 0.7425900435249433,
476
+ "content_variance_wrong_average_precision": 0.468111215636207,
477
+ "content_one_sigma_accuracy": 0.7752278645833334,
478
+ "content_fixed_half_point_accuracy": 0.7152506510416666,
479
+ "content_mean_predicted_sigma": 0.468670747090409,
480
+ "content_mean_absolute_calibration_gap": 0.2627904350268082
481
+ },
482
+ "by_length": {
483
+ "512": {
484
+ "selection_loss": 0.6120346124297937,
485
+ "content_mae": 0.3830975716779529,
486
+ "content_macro_mae": 0.39077505134591706,
487
+ "content_exact_accuracy": 0.706298828125,
488
+ "content_spearman": 0.8809364124482335,
489
+ "content_gaussian_nll": -0.17671267232550236,
490
+ "content_mean_predicted_variance": 0.30312733544612275,
491
+ "integrity_accuracy": 0.90771484375,
492
+ "integrity_macro_f1": 0.5140433335549824,
493
+ "critical_average_precision": 0.11496175566449311,
494
+ "critical_roc_auc": 0.8559302381371248,
495
+ "critical_f1_at_0_5": 0.15,
496
+ "critical_brier": 0.015920309394525538,
497
+ "final_mae": 0.41359218036217615,
498
+ "final_macro_mae": 0.4119320265168561,
499
+ "final_exact_accuracy": 0.689208984375,
500
+ "final_spearman": 0.8698473259231785,
501
+ "integrity_confusion_matrix": [
502
+ [
503
+ 3585,
504
+ 104,
505
+ 13
506
+ ],
507
+ [
508
+ 201,
509
+ 126,
510
+ 5
511
+ ],
512
+ [
513
+ 27,
514
+ 28,
515
+ 7
516
+ ]
517
+ ],
518
+ "content_distribution": {
519
+ "0": 230,
520
+ "1": 1043,
521
+ "2": 1237,
522
+ "3": 591,
523
+ "4": 995
524
+ },
525
+ "final_distribution": {
526
+ "0": 230,
527
+ "1": 1043,
528
+ "2": 1249,
529
+ "3": 666,
530
+ "4": 908
531
+ },
532
+ "content_per_class": {
533
+ "0": {
534
+ "precision": 0.8782608695652174,
535
+ "recall": 0.8015873015873016,
536
+ "f1": 0.8381742738589212,
537
+ "support": 252
538
+ },
539
+ "1": {
540
+ "precision": 0.7392138063279002,
541
+ "recall": 0.6896243291592129,
542
+ "f1": 0.7135585377140213,
543
+ "support": 1118
544
+ },
545
+ "2": {
546
+ "precision": 0.6313662085691188,
547
+ "recall": 0.6698113207547169,
548
+ "f1": 0.6500208073241781,
549
+ "support": 1166
550
+ },
551
+ "3": {
552
+ "precision": 0.4754653130287648,
553
+ "recall": 0.49822695035460995,
554
+ "f1": 0.48658008658008656,
555
+ "support": 564
556
+ },
557
+ "4": {
558
+ "precision": 0.8623115577889447,
559
+ "recall": 0.8614457831325302,
560
+ "f1": 0.861878453038674,
561
+ "support": 996
562
+ }
563
+ },
564
+ "content_macro_f1": 0.7100424317031763,
565
+ "content_f1_class2_3_mean": 0.5683004469521323,
566
+ "content_mean_variance_correct": 0.24633350799266054,
567
+ "content_mean_variance_wrong": 0.43970634028641054,
568
+ "content_wrong_to_correct_variance_ratio": 1.7850041753130537,
569
+ "content_variance_wrong_auroc": 0.736118857137603,
570
+ "content_variance_wrong_average_precision": 0.46183723218564027,
571
+ "content_one_sigma_accuracy": 0.783447265625,
572
+ "content_fixed_half_point_accuracy": 0.706298828125,
573
+ "content_mean_predicted_sigma": 0.4868545213487623,
574
+ "content_mean_absolute_calibration_gap": 0.2706309738992609
575
+ },
576
+ "1024": {
577
+ "selection_loss": 0.5862670943393096,
578
+ "content_mae": 0.3789426484455323,
579
+ "content_macro_mae": 0.38048229162949315,
580
+ "content_exact_accuracy": 0.7060546875,
581
+ "content_spearman": 0.8906623521795864,
582
+ "content_gaussian_nll": -0.12287645864569759,
583
+ "content_mean_predicted_variance": 0.2687230434777398,
584
+ "integrity_accuracy": 0.914794921875,
585
+ "integrity_macro_f1": 0.5244504767937112,
586
+ "critical_average_precision": 0.17686078916073436,
587
+ "critical_roc_auc": 0.8947194811841285,
588
+ "critical_f1_at_0_5": 0.1891891891891892,
589
+ "critical_brier": 0.013953239596104013,
590
+ "final_mae": 0.4034991220801931,
591
+ "final_macro_mae": 0.39627136522476464,
592
+ "final_exact_accuracy": 0.6962890625,
593
+ "final_spearman": 0.8806180293419044,
594
+ "integrity_confusion_matrix": [
595
+ [
596
+ 3629,
597
+ 69,
598
+ 4
599
+ ],
600
+ [
601
+ 219,
602
+ 110,
603
+ 3
604
+ ],
605
+ [
606
+ 25,
607
+ 29,
608
+ 8
609
+ ]
610
+ ],
611
+ "content_distribution": {
612
+ "0": 242,
613
+ "1": 869,
614
+ "2": 1264,
615
+ "3": 665,
616
+ "4": 1056
617
+ },
618
+ "final_distribution": {
619
+ "0": 242,
620
+ "1": 869,
621
+ "2": 1272,
622
+ "3": 700,
623
+ "4": 1013
624
+ },
625
+ "content_per_class": {
626
+ "0": {
627
+ "precision": 0.8636363636363636,
628
+ "recall": 0.8293650793650794,
629
+ "f1": 0.8461538461538461,
630
+ "support": 252
631
+ },
632
+ "1": {
633
+ "precision": 0.8193325661680092,
634
+ "recall": 0.6368515205724508,
635
+ "f1": 0.7166582788122798,
636
+ "support": 1118
637
+ },
638
+ "2": {
639
+ "precision": 0.6186708860759493,
640
+ "recall": 0.6706689536878216,
641
+ "f1": 0.6436213991769547,
642
+ "support": 1166
643
+ },
644
+ "3": {
645
+ "precision": 0.43909774436090226,
646
+ "recall": 0.5177304964539007,
647
+ "f1": 0.4751830756712775,
648
+ "support": 564
649
+ },
650
+ "4": {
651
+ "precision": 0.8494318181818182,
652
+ "recall": 0.9006024096385542,
653
+ "f1": 0.8742690058479532,
654
+ "support": 996
655
+ }
656
+ },
657
+ "content_macro_f1": 0.7111771211324622,
658
+ "content_f1_class2_3_mean": 0.5594022374241161,
659
+ "content_mean_variance_correct": 0.21197317897944154,
660
+ "content_mean_variance_wrong": 0.40503584092714057,
661
+ "content_wrong_to_correct_variance_ratio": 1.9107881613948123,
662
+ "content_variance_wrong_auroc": 0.7573059258442353,
663
+ "content_variance_wrong_average_precision": 0.4909850888338813,
664
+ "content_one_sigma_accuracy": 0.76025390625,
665
+ "content_fixed_half_point_accuracy": 0.7060546875,
666
+ "content_mean_predicted_sigma": 0.4568070278293196,
667
+ "content_mean_absolute_calibration_gap": 0.2568249294043655
668
+ },
669
+ "2048": {
670
+ "selection_loss": 0.5701281481916703,
671
+ "content_mae": 0.35768043497543045,
672
+ "content_macro_mae": 0.364176199883009,
673
+ "content_exact_accuracy": 0.7333984375,
674
+ "content_spearman": 0.8949834476665999,
675
+ "content_gaussian_nll": -0.1968527833801204,
676
+ "content_mean_predicted_variance": 0.2756774667594437,
677
+ "integrity_accuracy": 0.914306640625,
678
+ "integrity_macro_f1": 0.5180030709490746,
679
+ "critical_average_precision": 0.1761922067653546,
680
+ "critical_roc_auc": 0.9030278919506773,
681
+ "critical_f1_at_0_5": 0.08823529411764706,
682
+ "critical_brier": 0.013940391738764532,
683
+ "final_mae": 0.3815610584608907,
684
+ "final_macro_mae": 0.37957188816310106,
685
+ "final_exact_accuracy": 0.72509765625,
686
+ "final_spearman": 0.885492549972223,
687
+ "integrity_confusion_matrix": [
688
+ [
689
+ 3602,
690
+ 96,
691
+ 4
692
+ ],
693
+ [
694
+ 191,
695
+ 138,
696
+ 3
697
+ ],
698
+ [
699
+ 24,
700
+ 33,
701
+ 5
702
+ ]
703
+ ],
704
+ "content_distribution": {
705
+ "0": 236,
706
+ "1": 846,
707
+ "2": 1358,
708
+ "3": 591,
709
+ "4": 1065
710
+ },
711
+ "final_distribution": {
712
+ "0": 236,
713
+ "1": 846,
714
+ "2": 1373,
715
+ "3": 624,
716
+ "4": 1017
717
+ },
718
+ "content_per_class": {
719
+ "0": {
720
+ "precision": 0.8771186440677966,
721
+ "recall": 0.8214285714285714,
722
+ "f1": 0.8483606557377049,
723
+ "support": 252
724
+ },
725
+ "1": {
726
+ "precision": 0.8427895981087471,
727
+ "recall": 0.6377459749552773,
728
+ "f1": 0.7260692464358453,
729
+ "support": 1118
730
+ },
731
+ "2": {
732
+ "precision": 0.6539027982326951,
733
+ "recall": 0.7615780445969125,
734
+ "f1": 0.7036450079239303,
735
+ "support": 1166
736
+ },
737
+ "3": {
738
+ "precision": 0.4906937394247039,
739
+ "recall": 0.5141843971631206,
740
+ "f1": 0.5021645021645021,
741
+ "support": 564
742
+ },
743
+ "4": {
744
+ "precision": 0.8507042253521127,
745
+ "recall": 0.9096385542168675,
746
+ "f1": 0.8791848617176128,
747
+ "support": 996
748
+ }
749
+ },
750
+ "content_macro_f1": 0.731884854795919,
751
+ "content_f1_class2_3_mean": 0.6029047550442161,
752
+ "content_mean_variance_correct": 0.22555568291829578,
753
+ "content_mean_variance_wrong": 0.4135582713920519,
754
+ "content_wrong_to_correct_variance_ratio": 1.8335085422869053,
755
+ "content_variance_wrong_auroc": 0.7346610197392488,
756
+ "content_variance_wrong_average_precision": 0.4611742547263893,
757
+ "content_one_sigma_accuracy": 0.781982421875,
758
+ "content_fixed_half_point_accuracy": 0.7333984375,
759
+ "content_mean_predicted_sigma": 0.4623506920931452,
760
+ "content_mean_absolute_calibration_gap": 0.2609154017767981
761
+ }
762
+ }
763
+ }
764
+ },
765
+ "prediction_manifest": {
766
+ "private_ood_test_ctx1024.parquet": {
767
+ "rows": 4096,
768
+ "bytes": 317328,
769
+ "sha256": "f2c54290a754bdf608e39e706c9948868cd341d1b3a6cce07f6ae7c018c07858"
770
+ },
771
+ "private_ood_test_ctx2048.parquet": {
772
+ "rows": 4096,
773
+ "bytes": 314219,
774
+ "sha256": "7e409a5182ebf42043873173d5617fea04e406cd6730417bab6e02adea919bf2"
775
+ },
776
+ "private_ood_test_ctx512.parquet": {
777
+ "rows": 4096,
778
+ "bytes": 318598,
779
+ "sha256": "9f63b8d7299fc5f1e395e39b0185edc493899782f330802bf1bf499f0947dd98"
780
+ },
781
+ "public_test_ctx1024.parquet": {
782
+ "rows": 4096,
783
+ "bytes": 313111,
784
+ "sha256": "e930d8592489f8d78670345f98d7573396a9cbe953e6ef1c873c09e8a0dd5d75"
785
+ },
786
+ "public_test_ctx2048.parquet": {
787
+ "rows": 4096,
788
+ "bytes": 310365,
789
+ "sha256": "83a8a7d417be256de469edbc65c7c0104dbd5879505952df888c6cbcdb0b202b"
790
+ },
791
+ "public_test_ctx512.parquet": {
792
+ "rows": 4096,
793
+ "bytes": 314796,
794
+ "sha256": "0ab3099b4b1caa5b4b33ce4c0bf89892317984a21fbed0e649d495a39b4cb935"
795
+ }
796
+ }
797
+ }
config.json ADDED
@@ -0,0 +1,44 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_cross_attention": false,
3
+ "architectures": [
4
+ "XLMRobertaForSequenceClassification"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "bos_token_id": 0,
8
+ "classifier_dropout": null,
9
+ "dtype": "bfloat16",
10
+ "eos_token_id": 2,
11
+ "hidden_act": "gelu",
12
+ "hidden_dropout_prob": 0.1,
13
+ "hidden_size": 1024,
14
+ "id2label": {
15
+ "0": "LABEL_0",
16
+ "1": "LABEL_1",
17
+ "2": "LABEL_2",
18
+ "3": "LABEL_3",
19
+ "4": "LABEL_4"
20
+ },
21
+ "initializer_range": 0.02,
22
+ "intermediate_size": 4096,
23
+ "is_decoder": false,
24
+ "label2id": {
25
+ "LABEL_0": 0,
26
+ "LABEL_1": 1,
27
+ "LABEL_2": 2,
28
+ "LABEL_3": 3,
29
+ "LABEL_4": 4
30
+ },
31
+ "layer_norm_eps": 1e-05,
32
+ "max_position_embeddings": 8194,
33
+ "model_type": "xlm-roberta",
34
+ "num_attention_heads": 16,
35
+ "num_hidden_layers": 24,
36
+ "output_past": true,
37
+ "pad_token_id": 1,
38
+ "position_embedding_type": "absolute",
39
+ "tie_word_embeddings": true,
40
+ "transformers_version": "5.16.1",
41
+ "type_vocab_size": 1,
42
+ "use_cache": true,
43
+ "vocab_size": 250002
44
+ }
final_training_config.json ADDED
@@ -0,0 +1,121 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "dragonkue/snowflake-arctic-embed-l-v2.0-ko",
3
+ "variation": "capped_quarter",
4
+ "seed": 20260921,
5
+ "data_seed": 20260921,
6
+ "base_train_rows": 540000,
7
+ "validation_rows": 60000,
8
+ "upsampled_train_population": {
9
+ "base_rows": 540000,
10
+ "extra_score4_rows": 27498,
11
+ "rows": 567498,
12
+ "score4_rows": 56750,
13
+ "score4_fraction": 0.10000035242414951,
14
+ "indices_sha256": "cd146dfd9c4772ef38d09a4bf593d1f213b2bb368824c917a0727d2da1a17bc4"
15
+ },
16
+ "effective_batch_size": 256,
17
+ "train_micro_batches": {
18
+ "512": 32,
19
+ "1024": 16,
20
+ "2048": 8
21
+ },
22
+ "gradient_accumulation": {
23
+ "512": 8,
24
+ "1024": 16,
25
+ "2048": 32
26
+ },
27
+ "context_step_ratio": {
28
+ "512": 16,
29
+ "1024": 4,
30
+ "2048": 1
31
+ },
32
+ "step_counts": {
33
+ "512": 6757,
34
+ "1024": 1689,
35
+ "2048": 422
36
+ },
37
+ "sample_counts": {
38
+ "512": 1729792,
39
+ "1024": 432384,
40
+ "2048": 108032
41
+ },
42
+ "steps_per_epoch": 2217,
43
+ "max_epochs": 4,
44
+ "max_steps": 8868,
45
+ "eval_steps": [
46
+ 1109,
47
+ 2217,
48
+ 3326,
49
+ 4434,
50
+ 5543,
51
+ 6651,
52
+ 7760,
53
+ 8868
54
+ ],
55
+ "warmup_steps": 300,
56
+ "head_learning_rate": 0.0003,
57
+ "encoder_learning_rate": 3e-05,
58
+ "scheduler": "linear warmup then cosine decay to zero",
59
+ "truncation_marker": null,
60
+ "split_manifest": {
61
+ "created_at": "2026-10-03T18:28:15.232161+00:00",
62
+ "seed": 20260921,
63
+ "method": "stratified by public/private provenance, content score, integrity label; cluster-safe",
64
+ "base_rows": 600000,
65
+ "train": {
66
+ "rows": 540000,
67
+ "public_rows": 449998,
68
+ "private_rows": 90002,
69
+ "content_counts": {
70
+ "0": 60618,
71
+ "1": 251387,
72
+ "2": 128263,
73
+ "3": 70480,
74
+ "4": 29252
75
+ },
76
+ "integrity_counts": {
77
+ "0": 528496,
78
+ "1": 9624,
79
+ "2": 1880
80
+ },
81
+ "indices_sha256": "d0384fb70fbea90859eb71fd10f0a34b654a6c5557089429979533631847d645",
82
+ "ids_sha256": "3241fac0b92bf334ee406d1c8fb868bad11be577433c73b35043e1d92f03b728",
83
+ "clusters_sha256": "2c0728e27091ce4e2699cd0cf25fba53792209b75cb040dc6d4d568e70a9f01d"
84
+ },
85
+ "validation": {
86
+ "rows": 60000,
87
+ "public_rows": 50002,
88
+ "private_rows": 9998,
89
+ "content_counts": {
90
+ "0": 6735,
91
+ "1": 27932,
92
+ "2": 14252,
93
+ "3": 7830,
94
+ "4": 3251
95
+ },
96
+ "integrity_counts": {
97
+ "0": 58720,
98
+ "1": 1070,
99
+ "2": 210
100
+ },
101
+ "indices_sha256": "80d7239d50b44a23d2ccab4e6a5af50d789be7e4977d8f6172783cc8bde72e9d",
102
+ "ids_sha256": "05a6505df655a1edee4bbfcf66e055d1b1c63f3a8a7cb61677cc430788e7fd4f",
103
+ "clusters_sha256": "36222da42a8344e1381e320a00190f58af507d39afdc9a88096a07df0bbbe14b"
104
+ },
105
+ "overlap": {
106
+ "indices": 0,
107
+ "clusters": 0
108
+ }
109
+ },
110
+ "schedule_sha256": "06270c088964c90c980ab8856de5b8f3b1c5bb4c478f53100e944bf1078432f3",
111
+ "population_indices_sha256": "ad8477c94140db4b9f66b7cd0320da8e760c8577f8c67bde41849a56d3b59f7e",
112
+ "checkpoint_selection": {
113
+ "objectives": [
114
+ "validation content macro F1",
115
+ "validation mean content F1 classes 2 and 3"
116
+ ],
117
+ "dominated_definition": "both objectives strictly lower than an earlier checkpoint",
118
+ "minimum_epochs": 2,
119
+ "patience": "two consecutive half-epoch dominated checkpoints"
120
+ }
121
+ }
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e823b590afc3f81236344952455bee164beee069dd73bfb8bfe607189dc36bed
3
+ size 1135568290
test_metrics.json ADDED
@@ -0,0 +1,799 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "global_step": 8868,
3
+ "aggregate_epoch": 4.0,
4
+ "model_path": "/ko_edu_snowflake_final_train/checkpoints/step_08868",
5
+ "validation": {
6
+ "global_step": 8868,
7
+ "aggregate_epoch": 4.0,
8
+ "content_macro_f1": 0.7691223696148852,
9
+ "content_f1_class2_3_mean": 0.7002175340967471
10
+ },
11
+ "tests": {
12
+ "public_test": {
13
+ "aggregate_equal_context_weight": {
14
+ "selection_loss": 0.5996472404267125,
15
+ "content_mae": 0.36416186395414174,
16
+ "content_macro_mae": 0.39134996560001856,
17
+ "content_exact_accuracy": 0.7229817708333334,
18
+ "content_spearman": 0.9219387450855417,
19
+ "content_gaussian_nll": -0.525285516083209,
20
+ "content_mean_predicted_variance": 0.2302397398710588,
21
+ "integrity_accuracy": 0.9256184895833334,
22
+ "integrity_macro_f1": 0.5145415702399313,
23
+ "critical_average_precision": 0.1668109006932243,
24
+ "critical_roc_auc": 0.924077158723946,
25
+ "critical_f1_at_0_5": 0.20155038759689922,
26
+ "critical_brier": 0.007695191131648721,
27
+ "final_mae": 0.3742708417087128,
28
+ "final_macro_mae": 0.40425132911669354,
29
+ "final_exact_accuracy": 0.7154134114583334,
30
+ "final_spearman": 0.9196958171483666,
31
+ "integrity_confusion_matrix": [
32
+ [
33
+ 11192,
34
+ 412,
35
+ 45
36
+ ],
37
+ [
38
+ 385,
39
+ 156,
40
+ 17
41
+ ],
42
+ [
43
+ 42,
44
+ 13,
45
+ 26
46
+ ]
47
+ ],
48
+ "content_distribution": {
49
+ "0": 3088,
50
+ "1": 2782,
51
+ "2": 2964,
52
+ "3": 2007,
53
+ "4": 1447
54
+ },
55
+ "final_distribution": {
56
+ "0": 3088,
57
+ "1": 2782,
58
+ "2": 3029,
59
+ "3": 2086,
60
+ "4": 1303
61
+ },
62
+ "content_per_class": {
63
+ "0": {
64
+ "precision": 0.9783031088082902,
65
+ "recall": 0.923006416131989,
66
+ "f1": 0.9498506524131426,
67
+ "support": 3273
68
+ },
69
+ "1": {
70
+ "precision": 0.6294033069734004,
71
+ "recall": 0.8106481481481481,
72
+ "f1": 0.7086199919061109,
73
+ "support": 2160
74
+ },
75
+ "2": {
76
+ "precision": 0.6271929824561403,
77
+ "recall": 0.6441441441441441,
78
+ "f1": 0.6355555555555555,
79
+ "support": 2886
80
+ },
81
+ "3": {
82
+ "precision": 0.5371200797209765,
83
+ "recall": 0.5502807554874937,
84
+ "f1": 0.5436207766011094,
85
+ "support": 1959
86
+ },
87
+ "4": {
88
+ "precision": 0.8120248790601244,
89
+ "recall": 0.5845771144278606,
90
+ "f1": 0.6797801562048018,
91
+ "support": 2010
92
+ }
93
+ },
94
+ "content_macro_f1": 0.7034854265361441,
95
+ "content_f1_class2_3_mean": 0.5895881660783324,
96
+ "content_mean_variance_correct": 0.16486197243390396,
97
+ "content_mean_variance_wrong": 0.4008672621717884,
98
+ "content_wrong_to_correct_variance_ratio": 2.431532610302252,
99
+ "content_variance_wrong_auroc": 0.799274570902363,
100
+ "content_variance_wrong_average_precision": 0.5623532341417358,
101
+ "content_one_sigma_accuracy": 0.7645670572916666,
102
+ "content_fixed_half_point_accuracy": 0.7229817708333334,
103
+ "content_mean_predicted_sigma": 0.4059100152568453,
104
+ "content_mean_absolute_calibration_gap": 0.23573810502832473
105
+ },
106
+ "by_length": {
107
+ "512": {
108
+ "selection_loss": 0.6475959000101105,
109
+ "content_mae": 0.403041226476617,
110
+ "content_macro_mae": 0.43819375073195327,
111
+ "content_exact_accuracy": 0.69873046875,
112
+ "content_spearman": 0.9094162283707627,
113
+ "content_gaussian_nll": -0.33558642342258216,
114
+ "content_mean_predicted_variance": 0.25569876981790673,
115
+ "integrity_accuracy": 0.92724609375,
116
+ "integrity_macro_f1": 0.5196996608761314,
117
+ "critical_average_precision": 0.16239140288737114,
118
+ "critical_roc_auc": 0.9300401409027608,
119
+ "critical_f1_at_0_5": 0.24489795918367346,
120
+ "critical_brier": 0.008444094912288811,
121
+ "final_mae": 0.4134857682716224,
122
+ "final_macro_mae": 0.4519832393323484,
123
+ "final_exact_accuracy": 0.689453125,
124
+ "final_spearman": 0.9069972446950932,
125
+ "integrity_confusion_matrix": [
126
+ [
127
+ 3741,
128
+ 122,
129
+ 20
130
+ ],
131
+ [
132
+ 133,
133
+ 46,
134
+ 7
135
+ ],
136
+ [
137
+ 13,
138
+ 3,
139
+ 11
140
+ ]
141
+ ],
142
+ "content_distribution": {
143
+ "0": 1031,
144
+ "1": 997,
145
+ "2": 1037,
146
+ "3": 666,
147
+ "4": 365
148
+ },
149
+ "final_distribution": {
150
+ "0": 1031,
151
+ "1": 997,
152
+ "2": 1058,
153
+ "3": 695,
154
+ "4": 315
155
+ },
156
+ "content_per_class": {
157
+ "0": {
158
+ "precision": 0.9747817652764307,
159
+ "recall": 0.921173235563703,
160
+ "f1": 0.9472196041470311,
161
+ "support": 1091
162
+ },
163
+ "1": {
164
+ "precision": 0.5927783350050151,
165
+ "recall": 0.8208333333333333,
166
+ "f1": 0.6884100174723354,
167
+ "support": 720
168
+ },
169
+ "2": {
170
+ "precision": 0.5756991321118612,
171
+ "recall": 0.6205821205821206,
172
+ "f1": 0.5972986493246624,
173
+ "support": 962
174
+ },
175
+ "3": {
176
+ "precision": 0.5225225225225225,
177
+ "recall": 0.5329249617151608,
178
+ "f1": 0.5276724791508719,
179
+ "support": 653
180
+ },
181
+ "4": {
182
+ "precision": 0.8794520547945206,
183
+ "recall": 0.4791044776119403,
184
+ "f1": 0.6202898550724638,
185
+ "support": 670
186
+ }
187
+ },
188
+ "content_macro_f1": 0.6761781210334729,
189
+ "content_f1_class2_3_mean": 0.5624855642377671,
190
+ "content_mean_variance_correct": 0.17687892520976323,
191
+ "content_mean_variance_wrong": 0.4385046006675881,
192
+ "content_wrong_to_correct_variance_ratio": 2.4791229376114723,
193
+ "content_variance_wrong_auroc": 0.8046483457862315,
194
+ "content_variance_wrong_average_precision": 0.5972072289789552,
195
+ "content_one_sigma_accuracy": 0.746337890625,
196
+ "content_fixed_half_point_accuracy": 0.69873046875,
197
+ "content_mean_predicted_sigma": 0.4275034727898084,
198
+ "content_mean_absolute_calibration_gap": 0.2608774770819382
199
+ },
200
+ "1024": {
201
+ "selection_loss": 0.5793806554162239,
202
+ "content_mae": 0.35502210318509775,
203
+ "content_macro_mae": 0.3809260249430456,
204
+ "content_exact_accuracy": 0.729248046875,
205
+ "content_spearman": 0.9258773309917829,
206
+ "content_gaussian_nll": -0.5862765162921657,
207
+ "content_mean_predicted_variance": 0.21921119402604106,
208
+ "integrity_accuracy": 0.931396484375,
209
+ "integrity_macro_f1": 0.5242311737226991,
210
+ "critical_average_precision": 0.20618147810728718,
211
+ "critical_roc_auc": 0.9281923850613947,
212
+ "critical_f1_at_0_5": 0.2,
213
+ "critical_brier": 0.007129219811954048,
214
+ "final_mae": 0.36424440595520335,
215
+ "final_macro_mae": 0.39263800988537184,
216
+ "final_exact_accuracy": 0.72119140625,
217
+ "final_spearman": 0.9237399815073253,
218
+ "integrity_confusion_matrix": [
219
+ [
220
+ 3758,
221
+ 112,
222
+ 13
223
+ ],
224
+ [
225
+ 133,
226
+ 48,
227
+ 5
228
+ ],
229
+ [
230
+ 14,
231
+ 4,
232
+ 9
233
+ ]
234
+ ],
235
+ "content_distribution": {
236
+ "0": 1027,
237
+ "1": 920,
238
+ "2": 989,
239
+ "3": 669,
240
+ "4": 491
241
+ },
242
+ "final_distribution": {
243
+ "0": 1027,
244
+ "1": 920,
245
+ "2": 1014,
246
+ "3": 688,
247
+ "4": 447
248
+ },
249
+ "content_per_class": {
250
+ "0": {
251
+ "precision": 0.9795520934761441,
252
+ "recall": 0.922089825847846,
253
+ "f1": 0.9499527856468366,
254
+ "support": 1091
255
+ },
256
+ "1": {
257
+ "precision": 0.6326086956521739,
258
+ "recall": 0.8083333333333333,
259
+ "f1": 0.7097560975609756,
260
+ "support": 720
261
+ },
262
+ "2": {
263
+ "precision": 0.6339737108190091,
264
+ "recall": 0.6517671517671517,
265
+ "f1": 0.6427473090722706,
266
+ "support": 962
267
+ },
268
+ "3": {
269
+ "precision": 0.5500747384155455,
270
+ "recall": 0.5635528330781011,
271
+ "f1": 0.556732223903177,
272
+ "support": 653
273
+ },
274
+ "4": {
275
+ "precision": 0.8228105906313645,
276
+ "recall": 0.6029850746268657,
277
+ "f1": 0.6959517657192076,
278
+ "support": 670
279
+ }
280
+ },
281
+ "content_macro_f1": 0.7110280363804935,
282
+ "content_f1_class2_3_mean": 0.5997397664877238,
283
+ "content_mean_variance_correct": 0.15720720139904037,
284
+ "content_mean_variance_wrong": 0.3862138324181519,
285
+ "content_wrong_to_correct_variance_ratio": 2.456718451706433,
286
+ "content_variance_wrong_auroc": 0.8013581848364252,
287
+ "content_variance_wrong_average_precision": 0.5641231749105367,
288
+ "content_one_sigma_accuracy": 0.767822265625,
289
+ "content_fixed_half_point_accuracy": 0.729248046875,
290
+ "content_mean_predicted_sigma": 0.39683885099409644,
291
+ "content_mean_absolute_calibration_gap": 0.22794024060909682
292
+ },
293
+ "2048": {
294
+ "selection_loss": 0.563851957099979,
295
+ "content_mae": 0.33442226220071036,
296
+ "content_macro_mae": 0.3549301211250569,
297
+ "content_exact_accuracy": 0.740966796875,
298
+ "content_spearman": 0.9327042422925155,
299
+ "content_gaussian_nll": -0.653993608534879,
300
+ "content_mean_predicted_variance": 0.21580925576922855,
301
+ "integrity_accuracy": 0.918212890625,
302
+ "integrity_macro_f1": 0.49500456357784506,
303
+ "critical_average_precision": 0.16431265610031154,
304
+ "critical_roc_auc": 0.9154173834685017,
305
+ "critical_f1_at_0_5": 0.15,
306
+ "critical_brier": 0.007512258670703306,
307
+ "final_mae": 0.3450823508993126,
308
+ "final_macro_mae": 0.3681327381323604,
309
+ "final_exact_accuracy": 0.735595703125,
310
+ "final_spearman": 0.9304844155402512,
311
+ "integrity_confusion_matrix": [
312
+ [
313
+ 3693,
314
+ 178,
315
+ 12
316
+ ],
317
+ [
318
+ 119,
319
+ 62,
320
+ 5
321
+ ],
322
+ [
323
+ 15,
324
+ 6,
325
+ 6
326
+ ]
327
+ ],
328
+ "content_distribution": {
329
+ "0": 1030,
330
+ "1": 865,
331
+ "2": 938,
332
+ "3": 672,
333
+ "4": 591
334
+ },
335
+ "final_distribution": {
336
+ "0": 1030,
337
+ "1": 865,
338
+ "2": 957,
339
+ "3": 703,
340
+ "4": 541
341
+ },
342
+ "content_per_class": {
343
+ "0": {
344
+ "precision": 0.9805825242718447,
345
+ "recall": 0.925756186984418,
346
+ "f1": 0.9523809523809523,
347
+ "support": 1091
348
+ },
349
+ "1": {
350
+ "precision": 0.6682080924855491,
351
+ "recall": 0.8027777777777778,
352
+ "f1": 0.7293375394321766,
353
+ "support": 720
354
+ },
355
+ "2": {
356
+ "precision": 0.6769722814498934,
357
+ "recall": 0.66008316008316,
358
+ "f1": 0.6684210526315789,
359
+ "support": 962
360
+ },
361
+ "3": {
362
+ "precision": 0.5386904761904762,
363
+ "recall": 0.554364471669219,
364
+ "f1": 0.5464150943396227,
365
+ "support": 653
366
+ },
367
+ "4": {
368
+ "precision": 0.7614213197969543,
369
+ "recall": 0.6716417910447762,
370
+ "f1": 0.7137192704203014,
371
+ "support": 670
372
+ }
373
+ },
374
+ "content_macro_f1": 0.7220547818409264,
375
+ "content_f1_class2_3_mean": 0.6074180734856007,
376
+ "content_mean_variance_correct": 0.16106371287430865,
377
+ "content_mean_variance_wrong": 0.372409371401728,
378
+ "content_wrong_to_correct_variance_ratio": 2.3121866791457233,
379
+ "content_variance_wrong_auroc": 0.7882823235671796,
380
+ "content_variance_wrong_average_precision": 0.5170919064835311,
381
+ "content_one_sigma_accuracy": 0.779541015625,
382
+ "content_fixed_half_point_accuracy": 0.740966796875,
383
+ "content_mean_predicted_sigma": 0.39338772198663097,
384
+ "content_mean_absolute_calibration_gap": 0.2183965973939393
385
+ }
386
+ }
387
+ },
388
+ "private_ood_test": {
389
+ "aggregate_equal_context_weight": {
390
+ "selection_loss": 0.5928699060725598,
391
+ "content_mae": 0.3732402183663052,
392
+ "content_macro_mae": 0.37847784761947306,
393
+ "content_exact_accuracy": 0.7152506510416666,
394
+ "content_spearman": 0.8885748177452322,
395
+ "content_gaussian_nll": -0.16548063811710678,
396
+ "content_mean_predicted_variance": 0.2825092818944354,
397
+ "integrity_accuracy": 0.9122721354166666,
398
+ "integrity_macro_f1": 0.5191581195438396,
399
+ "critical_average_precision": 0.14243176618765288,
400
+ "critical_roc_auc": 0.8837826503394979,
401
+ "critical_f1_at_0_5": 0.14414414414414414,
402
+ "critical_brier": 0.014604646909798027,
403
+ "final_mae": 0.3995507869677533,
404
+ "final_macro_mae": 0.39592509330157394,
405
+ "final_exact_accuracy": 0.7035319010416666,
406
+ "final_spearman": 0.8782610375325385,
407
+ "integrity_confusion_matrix": [
408
+ [
409
+ 10816,
410
+ 269,
411
+ 21
412
+ ],
413
+ [
414
+ 611,
415
+ 374,
416
+ 11
417
+ ],
418
+ [
419
+ 76,
420
+ 90,
421
+ 20
422
+ ]
423
+ ],
424
+ "content_distribution": {
425
+ "0": 708,
426
+ "1": 2758,
427
+ "2": 3859,
428
+ "3": 1847,
429
+ "4": 3116
430
+ },
431
+ "final_distribution": {
432
+ "0": 708,
433
+ "1": 2758,
434
+ "2": 3894,
435
+ "3": 1990,
436
+ "4": 2938
437
+ },
438
+ "content_per_class": {
439
+ "0": {
440
+ "precision": 0.8728813559322034,
441
+ "recall": 0.8174603174603174,
442
+ "f1": 0.8442622950819673,
443
+ "support": 756
444
+ },
445
+ "1": {
446
+ "precision": 0.7962291515591008,
447
+ "recall": 0.6547406082289803,
448
+ "f1": 0.7185863874345549,
449
+ "support": 3354
450
+ },
451
+ "2": {
452
+ "precision": 0.6351386369525784,
453
+ "recall": 0.7006861063464837,
454
+ "f1": 0.666304200081555,
455
+ "support": 3498
456
+ },
457
+ "3": {
458
+ "precision": 0.4672441797509475,
459
+ "recall": 0.5100472813238771,
460
+ "f1": 0.4877083922011868,
461
+ "support": 1692
462
+ },
463
+ "4": {
464
+ "precision": 0.8539794608472401,
465
+ "recall": 0.8905622489959839,
466
+ "f1": 0.8718872870249017,
467
+ "support": 2988
468
+ }
469
+ },
470
+ "content_macro_f1": 0.7177497123648331,
471
+ "content_f1_class2_3_mean": 0.5770062961413709,
472
+ "content_mean_variance_correct": 0.22792565066763823,
473
+ "content_mean_variance_wrong": 0.4196157508433695,
474
+ "content_wrong_to_correct_variance_ratio": 1.8410203047100402,
475
+ "content_variance_wrong_auroc": 0.7425900435249433,
476
+ "content_variance_wrong_average_precision": 0.468111215636207,
477
+ "content_one_sigma_accuracy": 0.7752278645833334,
478
+ "content_fixed_half_point_accuracy": 0.7152506510416666,
479
+ "content_mean_predicted_sigma": 0.468670747090409,
480
+ "content_mean_absolute_calibration_gap": 0.2627904350268082
481
+ },
482
+ "by_length": {
483
+ "512": {
484
+ "selection_loss": 0.6120346124297937,
485
+ "content_mae": 0.3830975716779529,
486
+ "content_macro_mae": 0.39077505134591706,
487
+ "content_exact_accuracy": 0.706298828125,
488
+ "content_spearman": 0.8809364124482335,
489
+ "content_gaussian_nll": -0.17671267232550236,
490
+ "content_mean_predicted_variance": 0.30312733544612275,
491
+ "integrity_accuracy": 0.90771484375,
492
+ "integrity_macro_f1": 0.5140433335549824,
493
+ "critical_average_precision": 0.11496175566449311,
494
+ "critical_roc_auc": 0.8559302381371248,
495
+ "critical_f1_at_0_5": 0.15,
496
+ "critical_brier": 0.015920309394525538,
497
+ "final_mae": 0.41359218036217615,
498
+ "final_macro_mae": 0.4119320265168561,
499
+ "final_exact_accuracy": 0.689208984375,
500
+ "final_spearman": 0.8698473259231785,
501
+ "integrity_confusion_matrix": [
502
+ [
503
+ 3585,
504
+ 104,
505
+ 13
506
+ ],
507
+ [
508
+ 201,
509
+ 126,
510
+ 5
511
+ ],
512
+ [
513
+ 27,
514
+ 28,
515
+ 7
516
+ ]
517
+ ],
518
+ "content_distribution": {
519
+ "0": 230,
520
+ "1": 1043,
521
+ "2": 1237,
522
+ "3": 591,
523
+ "4": 995
524
+ },
525
+ "final_distribution": {
526
+ "0": 230,
527
+ "1": 1043,
528
+ "2": 1249,
529
+ "3": 666,
530
+ "4": 908
531
+ },
532
+ "content_per_class": {
533
+ "0": {
534
+ "precision": 0.8782608695652174,
535
+ "recall": 0.8015873015873016,
536
+ "f1": 0.8381742738589212,
537
+ "support": 252
538
+ },
539
+ "1": {
540
+ "precision": 0.7392138063279002,
541
+ "recall": 0.6896243291592129,
542
+ "f1": 0.7135585377140213,
543
+ "support": 1118
544
+ },
545
+ "2": {
546
+ "precision": 0.6313662085691188,
547
+ "recall": 0.6698113207547169,
548
+ "f1": 0.6500208073241781,
549
+ "support": 1166
550
+ },
551
+ "3": {
552
+ "precision": 0.4754653130287648,
553
+ "recall": 0.49822695035460995,
554
+ "f1": 0.48658008658008656,
555
+ "support": 564
556
+ },
557
+ "4": {
558
+ "precision": 0.8623115577889447,
559
+ "recall": 0.8614457831325302,
560
+ "f1": 0.861878453038674,
561
+ "support": 996
562
+ }
563
+ },
564
+ "content_macro_f1": 0.7100424317031763,
565
+ "content_f1_class2_3_mean": 0.5683004469521323,
566
+ "content_mean_variance_correct": 0.24633350799266054,
567
+ "content_mean_variance_wrong": 0.43970634028641054,
568
+ "content_wrong_to_correct_variance_ratio": 1.7850041753130537,
569
+ "content_variance_wrong_auroc": 0.736118857137603,
570
+ "content_variance_wrong_average_precision": 0.46183723218564027,
571
+ "content_one_sigma_accuracy": 0.783447265625,
572
+ "content_fixed_half_point_accuracy": 0.706298828125,
573
+ "content_mean_predicted_sigma": 0.4868545213487623,
574
+ "content_mean_absolute_calibration_gap": 0.2706309738992609
575
+ },
576
+ "1024": {
577
+ "selection_loss": 0.5862670943393096,
578
+ "content_mae": 0.3789426484455323,
579
+ "content_macro_mae": 0.38048229162949315,
580
+ "content_exact_accuracy": 0.7060546875,
581
+ "content_spearman": 0.8906623521795864,
582
+ "content_gaussian_nll": -0.12287645864569759,
583
+ "content_mean_predicted_variance": 0.2687230434777398,
584
+ "integrity_accuracy": 0.914794921875,
585
+ "integrity_macro_f1": 0.5244504767937112,
586
+ "critical_average_precision": 0.17686078916073436,
587
+ "critical_roc_auc": 0.8947194811841285,
588
+ "critical_f1_at_0_5": 0.1891891891891892,
589
+ "critical_brier": 0.013953239596104013,
590
+ "final_mae": 0.4034991220801931,
591
+ "final_macro_mae": 0.39627136522476464,
592
+ "final_exact_accuracy": 0.6962890625,
593
+ "final_spearman": 0.8806180293419044,
594
+ "integrity_confusion_matrix": [
595
+ [
596
+ 3629,
597
+ 69,
598
+ 4
599
+ ],
600
+ [
601
+ 219,
602
+ 110,
603
+ 3
604
+ ],
605
+ [
606
+ 25,
607
+ 29,
608
+ 8
609
+ ]
610
+ ],
611
+ "content_distribution": {
612
+ "0": 242,
613
+ "1": 869,
614
+ "2": 1264,
615
+ "3": 665,
616
+ "4": 1056
617
+ },
618
+ "final_distribution": {
619
+ "0": 242,
620
+ "1": 869,
621
+ "2": 1272,
622
+ "3": 700,
623
+ "4": 1013
624
+ },
625
+ "content_per_class": {
626
+ "0": {
627
+ "precision": 0.8636363636363636,
628
+ "recall": 0.8293650793650794,
629
+ "f1": 0.8461538461538461,
630
+ "support": 252
631
+ },
632
+ "1": {
633
+ "precision": 0.8193325661680092,
634
+ "recall": 0.6368515205724508,
635
+ "f1": 0.7166582788122798,
636
+ "support": 1118
637
+ },
638
+ "2": {
639
+ "precision": 0.6186708860759493,
640
+ "recall": 0.6706689536878216,
641
+ "f1": 0.6436213991769547,
642
+ "support": 1166
643
+ },
644
+ "3": {
645
+ "precision": 0.43909774436090226,
646
+ "recall": 0.5177304964539007,
647
+ "f1": 0.4751830756712775,
648
+ "support": 564
649
+ },
650
+ "4": {
651
+ "precision": 0.8494318181818182,
652
+ "recall": 0.9006024096385542,
653
+ "f1": 0.8742690058479532,
654
+ "support": 996
655
+ }
656
+ },
657
+ "content_macro_f1": 0.7111771211324622,
658
+ "content_f1_class2_3_mean": 0.5594022374241161,
659
+ "content_mean_variance_correct": 0.21197317897944154,
660
+ "content_mean_variance_wrong": 0.40503584092714057,
661
+ "content_wrong_to_correct_variance_ratio": 1.9107881613948123,
662
+ "content_variance_wrong_auroc": 0.7573059258442353,
663
+ "content_variance_wrong_average_precision": 0.4909850888338813,
664
+ "content_one_sigma_accuracy": 0.76025390625,
665
+ "content_fixed_half_point_accuracy": 0.7060546875,
666
+ "content_mean_predicted_sigma": 0.4568070278293196,
667
+ "content_mean_absolute_calibration_gap": 0.2568249294043655
668
+ },
669
+ "2048": {
670
+ "selection_loss": 0.5701281481916703,
671
+ "content_mae": 0.35768043497543045,
672
+ "content_macro_mae": 0.364176199883009,
673
+ "content_exact_accuracy": 0.7333984375,
674
+ "content_spearman": 0.8949834476665999,
675
+ "content_gaussian_nll": -0.1968527833801204,
676
+ "content_mean_predicted_variance": 0.2756774667594437,
677
+ "integrity_accuracy": 0.914306640625,
678
+ "integrity_macro_f1": 0.5180030709490746,
679
+ "critical_average_precision": 0.1761922067653546,
680
+ "critical_roc_auc": 0.9030278919506773,
681
+ "critical_f1_at_0_5": 0.08823529411764706,
682
+ "critical_brier": 0.013940391738764532,
683
+ "final_mae": 0.3815610584608907,
684
+ "final_macro_mae": 0.37957188816310106,
685
+ "final_exact_accuracy": 0.72509765625,
686
+ "final_spearman": 0.885492549972223,
687
+ "integrity_confusion_matrix": [
688
+ [
689
+ 3602,
690
+ 96,
691
+ 4
692
+ ],
693
+ [
694
+ 191,
695
+ 138,
696
+ 3
697
+ ],
698
+ [
699
+ 24,
700
+ 33,
701
+ 5
702
+ ]
703
+ ],
704
+ "content_distribution": {
705
+ "0": 236,
706
+ "1": 846,
707
+ "2": 1358,
708
+ "3": 591,
709
+ "4": 1065
710
+ },
711
+ "final_distribution": {
712
+ "0": 236,
713
+ "1": 846,
714
+ "2": 1373,
715
+ "3": 624,
716
+ "4": 1017
717
+ },
718
+ "content_per_class": {
719
+ "0": {
720
+ "precision": 0.8771186440677966,
721
+ "recall": 0.8214285714285714,
722
+ "f1": 0.8483606557377049,
723
+ "support": 252
724
+ },
725
+ "1": {
726
+ "precision": 0.8427895981087471,
727
+ "recall": 0.6377459749552773,
728
+ "f1": 0.7260692464358453,
729
+ "support": 1118
730
+ },
731
+ "2": {
732
+ "precision": 0.6539027982326951,
733
+ "recall": 0.7615780445969125,
734
+ "f1": 0.7036450079239303,
735
+ "support": 1166
736
+ },
737
+ "3": {
738
+ "precision": 0.4906937394247039,
739
+ "recall": 0.5141843971631206,
740
+ "f1": 0.5021645021645021,
741
+ "support": 564
742
+ },
743
+ "4": {
744
+ "precision": 0.8507042253521127,
745
+ "recall": 0.9096385542168675,
746
+ "f1": 0.8791848617176128,
747
+ "support": 996
748
+ }
749
+ },
750
+ "content_macro_f1": 0.731884854795919,
751
+ "content_f1_class2_3_mean": 0.6029047550442161,
752
+ "content_mean_variance_correct": 0.22555568291829578,
753
+ "content_mean_variance_wrong": 0.4135582713920519,
754
+ "content_wrong_to_correct_variance_ratio": 1.8335085422869053,
755
+ "content_variance_wrong_auroc": 0.7346610197392488,
756
+ "content_variance_wrong_average_precision": 0.4611742547263893,
757
+ "content_one_sigma_accuracy": 0.781982421875,
758
+ "content_fixed_half_point_accuracy": 0.7333984375,
759
+ "content_mean_predicted_sigma": 0.4623506920931452,
760
+ "content_mean_absolute_calibration_gap": 0.2609154017767981
761
+ }
762
+ }
763
+ }
764
+ },
765
+ "prediction_manifest": {
766
+ "private_ood_test_ctx1024.parquet": {
767
+ "rows": 4096,
768
+ "bytes": 317328,
769
+ "sha256": "f2c54290a754bdf608e39e706c9948868cd341d1b3a6cce07f6ae7c018c07858"
770
+ },
771
+ "private_ood_test_ctx2048.parquet": {
772
+ "rows": 4096,
773
+ "bytes": 314219,
774
+ "sha256": "7e409a5182ebf42043873173d5617fea04e406cd6730417bab6e02adea919bf2"
775
+ },
776
+ "private_ood_test_ctx512.parquet": {
777
+ "rows": 4096,
778
+ "bytes": 318598,
779
+ "sha256": "9f63b8d7299fc5f1e395e39b0185edc493899782f330802bf1bf499f0947dd98"
780
+ },
781
+ "public_test_ctx1024.parquet": {
782
+ "rows": 4096,
783
+ "bytes": 313111,
784
+ "sha256": "e930d8592489f8d78670345f98d7573396a9cbe953e6ef1c873c09e8a0dd5d75"
785
+ },
786
+ "public_test_ctx2048.parquet": {
787
+ "rows": 4096,
788
+ "bytes": 310365,
789
+ "sha256": "83a8a7d417be256de469edbc65c7c0104dbd5879505952df888c6cbcdb0b202b"
790
+ },
791
+ "public_test_ctx512.parquet": {
792
+ "rows": 4096,
793
+ "bytes": 314796,
794
+ "sha256": "0ab3099b4b1caa5b4b33ce4c0bf89892317984a21fbed0e649d495a39b4cb935"
795
+ }
796
+ },
797
+ "repo_id": "werty1248/Ko-Edu-Classifier-2k-ep4p0-step8868",
798
+ "selected_for_upload": true
799
+ }
tokenizer.json ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a56def25aa40facc030ea8b0b87f3688e4b3c39eb8b45d5702b3a1300fe2a20
3
+ size 17082734
tokenizer_config.json ADDED
@@ -0,0 +1,23 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "add_prefix_space": true,
3
+ "backend": "tokenizers",
4
+ "bos_token": "<s>",
5
+ "clean_up_tokenization_spaces": true,
6
+ "cls_token": "<s>",
7
+ "eos_token": "</s>",
8
+ "is_local": false,
9
+ "local_files_only": false,
10
+ "mask_token": "<mask>",
11
+ "max_length": 512,
12
+ "model_max_length": 8192,
13
+ "pad_to_multiple_of": null,
14
+ "pad_token": "<pad>",
15
+ "pad_token_type_id": 0,
16
+ "padding_side": "right",
17
+ "sep_token": "</s>",
18
+ "stride": 0,
19
+ "tokenizer_class": "XLMRobertaTokenizer",
20
+ "truncation_side": "right",
21
+ "truncation_strategy": "longest_first",
22
+ "unk_token": "<unk>"
23
+ }
validation.json ADDED
@@ -0,0 +1,386 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "kind": "validation",
3
+ "global_step": 8868,
4
+ "aggregate_epoch": 4.0,
5
+ "content_macro_f1": 0.7691223696148852,
6
+ "content_f1_class2_3_mean": 0.7002175340967471,
7
+ "dominated_in_both_objectives": false,
8
+ "bad_streak": 0,
9
+ "metrics": {
10
+ "aggregate_equal_context_weight": {
11
+ "selection_loss": 0.5140893343795243,
12
+ "content_mae": 0.2737932315832202,
13
+ "content_macro_mae": 0.3100500287727975,
14
+ "content_exact_accuracy": 0.7986166666666666,
15
+ "content_spearman": 0.8996877860492928,
16
+ "content_gaussian_nll": -0.7082690140020707,
17
+ "content_mean_predicted_variance": 0.1657321429377615,
18
+ "integrity_accuracy": 0.9669555555555556,
19
+ "integrity_macro_f1": 0.5279255895888997,
20
+ "critical_average_precision": 0.18384277757309278,
21
+ "critical_roc_auc": 0.9641408554278599,
22
+ "critical_f1_at_0_5": 0.22842197035745423,
23
+ "critical_brier": 0.004478420865291646,
24
+ "final_mae": 0.2757862484826717,
25
+ "final_macro_mae": 0.3203746614823401,
26
+ "final_exact_accuracy": 0.7975944444444445,
27
+ "final_spearman": 0.8992452079736484,
28
+ "integrity_confusion_matrix": [
29
+ [
30
+ 172551,
31
+ 3035,
32
+ 574
33
+ ],
34
+ [
35
+ 1757,
36
+ 1296,
37
+ 157
38
+ ],
39
+ [
40
+ 300,
41
+ 125,
42
+ 205
43
+ ]
44
+ ],
45
+ "content_distribution": {
46
+ "0": 17487,
47
+ "1": 85223,
48
+ "2": 44101,
49
+ "3": 22566,
50
+ "4": 10623
51
+ },
52
+ "final_distribution": {
53
+ "0": 17487,
54
+ "1": 85223,
55
+ "2": 44690,
56
+ "3": 22868,
57
+ "4": 9732
58
+ },
59
+ "content_per_class": {
60
+ "0": {
61
+ "precision": 0.8771087093269285,
62
+ "recall": 0.7591190299430834,
63
+ "f1": 0.8138597049771835,
64
+ "support": 20205
65
+ },
66
+ "1": {
67
+ "precision": 0.8654705889255248,
68
+ "recall": 0.8802090791923242,
69
+ "f1": 0.8727776167176471,
70
+ "support": 83796
71
+ },
72
+ "2": {
73
+ "precision": 0.7085100111108592,
74
+ "recall": 0.7307980166526336,
75
+ "f1": 0.719481446515537,
76
+ "support": 42756
77
+ },
78
+ "3": {
79
+ "precision": 0.6948949747407605,
80
+ "recall": 0.6675606641123882,
81
+ "f1": 0.6809536216779573,
82
+ "support": 23490
83
+ },
84
+ "4": {
85
+ "precision": 0.7274781135272522,
86
+ "recall": 0.7923715779760074,
87
+ "f1": 0.7585394581861012,
88
+ "support": 9753
89
+ }
90
+ },
91
+ "content_macro_f1": 0.7691223696148852,
92
+ "content_f1_class2_3_mean": 0.7002175340967471,
93
+ "content_mean_variance_correct": 0.14257208054755013,
94
+ "content_mean_variance_wrong": 0.25757694220547306,
95
+ "content_wrong_to_correct_variance_ratio": 1.806643637493716,
96
+ "content_variance_wrong_auroc": 0.7498307193383456,
97
+ "content_variance_wrong_average_precision": 0.37411770726590243,
98
+ "content_one_sigma_accuracy": 0.7989111111111111,
99
+ "content_fixed_half_point_accuracy": 0.7986166666666666,
100
+ "content_mean_predicted_sigma": 0.3627248205399464,
101
+ "content_mean_absolute_calibration_gap": 0.19831773181999218
102
+ },
103
+ "by_length": {
104
+ "512": {
105
+ "selection_loss": 0.5347708140474868,
106
+ "content_mae": 0.2877122715353697,
107
+ "content_macro_mae": 0.3326414165132969,
108
+ "content_exact_accuracy": 0.7846833333333333,
109
+ "content_spearman": 0.8920195781597556,
110
+ "content_gaussian_nll": -0.6381107757869618,
111
+ "content_mean_predicted_variance": 0.1784611644987229,
112
+ "integrity_accuracy": 0.9649666666666666,
113
+ "integrity_macro_f1": 0.5231860914821642,
114
+ "critical_average_precision": 0.1914824098632402,
115
+ "critical_roc_auc": 0.9600141766022349,
116
+ "critical_f1_at_0_5": 0.24,
117
+ "critical_brier": 0.004935151331732246,
118
+ "final_mae": 0.2900404591834574,
119
+ "final_macro_mae": 0.345389726812951,
120
+ "final_exact_accuracy": 0.7832666666666667,
121
+ "final_spearman": 0.8915067951762922,
122
+ "integrity_confusion_matrix": [
123
+ [
124
+ 57398,
125
+ 1095,
126
+ 227
127
+ ],
128
+ [
129
+ 583,
130
+ 424,
131
+ 63
132
+ ],
133
+ [
134
+ 96,
135
+ 38,
136
+ 76
137
+ ]
138
+ ],
139
+ "content_distribution": {
140
+ "0": 5814,
141
+ "1": 28407,
142
+ "2": 14862,
143
+ "3": 7542,
144
+ "4": 3375
145
+ },
146
+ "final_distribution": {
147
+ "0": 5814,
148
+ "1": 28407,
149
+ "2": 15133,
150
+ "3": 7661,
151
+ "4": 2985
152
+ },
153
+ "content_per_class": {
154
+ "0": {
155
+ "precision": 0.8753009975920193,
156
+ "recall": 0.7556050482553823,
157
+ "f1": 0.8110606422822536,
158
+ "support": 6735
159
+ },
160
+ "1": {
161
+ "precision": 0.857922343084451,
162
+ "recall": 0.8725118144064156,
163
+ "f1": 0.8651555760663129,
164
+ "support": 27932
165
+ },
166
+ "2": {
167
+ "precision": 0.6829498048714843,
168
+ "recall": 0.712180746561886,
169
+ "f1": 0.6972590506285635,
170
+ "support": 14252
171
+ },
172
+ "3": {
173
+ "precision": 0.6675948024396712,
174
+ "recall": 0.6430395913154534,
175
+ "f1": 0.6550871714806141,
176
+ "support": 7830
177
+ },
178
+ "4": {
179
+ "precision": 0.7217777777777777,
180
+ "recall": 0.74930790525992,
181
+ "f1": 0.735285239963779,
182
+ "support": 3251
183
+ }
184
+ },
185
+ "content_macro_f1": 0.7527695360843045,
186
+ "content_f1_class2_3_mean": 0.6761731110545888,
187
+ "content_mean_variance_correct": 0.1514430107925402,
188
+ "content_mean_variance_wrong": 0.27692402498643776,
189
+ "content_wrong_to_correct_variance_ratio": 1.8285691993128186,
190
+ "content_variance_wrong_auroc": 0.7523747674967851,
191
+ "content_variance_wrong_average_precision": 0.3968017995070952,
192
+ "content_one_sigma_accuracy": 0.7935333333333333,
193
+ "content_fixed_half_point_accuracy": 0.7846833333333333,
194
+ "content_mean_predicted_sigma": 0.37587916885792183,
195
+ "content_mean_absolute_calibration_gap": 0.20801088074152765
196
+ },
197
+ "1024": {
198
+ "selection_loss": 0.5039974774694058,
199
+ "content_mae": 0.26787705369252196,
200
+ "content_macro_mae": 0.3007754810817112,
201
+ "content_exact_accuracy": 0.8042833333333334,
202
+ "content_spearman": 0.9028312134289629,
203
+ "content_gaussian_nll": -0.7359567302086514,
204
+ "content_mean_predicted_variance": 0.1592265743720674,
205
+ "integrity_accuracy": 0.96835,
206
+ "integrity_macro_f1": 0.530696196209282,
207
+ "critical_average_precision": 0.18711201444922157,
208
+ "critical_roc_auc": 0.9656394205114728,
209
+ "critical_f1_at_0_5": 0.22589531680440772,
210
+ "critical_brier": 0.004262811449088174,
211
+ "final_mae": 0.26969235729647084,
212
+ "final_macro_mae": 0.3098881766812208,
213
+ "final_exact_accuracy": 0.8035666666666667,
214
+ "final_spearman": 0.9024137669725919,
215
+ "integrity_confusion_matrix": [
216
+ [
217
+ 57606,
218
+ 938,
219
+ 176
220
+ ],
221
+ [
222
+ 593,
223
+ 430,
224
+ 47
225
+ ],
226
+ [
227
+ 101,
228
+ 44,
229
+ 65
230
+ ]
231
+ ],
232
+ "content_distribution": {
233
+ "0": 5802,
234
+ "1": 28349,
235
+ "2": 14688,
236
+ "3": 7550,
237
+ "4": 3611
238
+ },
239
+ "final_distribution": {
240
+ "0": 5802,
241
+ "1": 28349,
242
+ "2": 14855,
243
+ "3": 7639,
244
+ "4": 3355
245
+ },
246
+ "content_per_class": {
247
+ "0": {
248
+ "precision": 0.8796966563254051,
249
+ "recall": 0.7578322197475872,
250
+ "f1": 0.8142298795565127,
251
+ "support": 6735
252
+ },
253
+ "1": {
254
+ "precision": 0.8693075593495362,
255
+ "recall": 0.8822855506229415,
256
+ "f1": 0.8757484763952311,
257
+ "support": 27932
258
+ },
259
+ "2": {
260
+ "precision": 0.71875,
261
+ "recall": 0.7407381420151558,
262
+ "f1": 0.7295784381478921,
263
+ "support": 14252
264
+ },
265
+ "3": {
266
+ "precision": 0.7054304635761589,
267
+ "recall": 0.6802043422733078,
268
+ "f1": 0.6925877763328999,
269
+ "support": 7830
270
+ },
271
+ "4": {
272
+ "precision": 0.7272223760731099,
273
+ "recall": 0.8077514610888957,
274
+ "f1": 0.7653745263771495,
275
+ "support": 3251
276
+ }
277
+ },
278
+ "content_macro_f1": 0.775503819361937,
279
+ "content_f1_class2_3_mean": 0.711083107240396,
280
+ "content_mean_variance_correct": 0.13785730128542895,
281
+ "content_mean_variance_wrong": 0.24704203986997347,
282
+ "content_wrong_to_correct_variance_ratio": 1.7920127375660806,
283
+ "content_variance_wrong_auroc": 0.7488252815378621,
284
+ "content_variance_wrong_average_precision": 0.365055072543468,
285
+ "content_one_sigma_accuracy": 0.7999333333333334,
286
+ "content_fixed_half_point_accuracy": 0.8042833333333334,
287
+ "content_mean_predicted_sigma": 0.35608551753900974,
288
+ "content_mean_absolute_calibration_gap": 0.19391285502780448
289
+ },
290
+ "2048": {
291
+ "selection_loss": 0.5010323490574853,
292
+ "content_mae": 0.265790369521769,
293
+ "content_macro_mae": 0.2967331887233845,
294
+ "content_exact_accuracy": 0.8068833333333333,
295
+ "content_spearman": 0.9041704167079047,
296
+ "content_gaussian_nll": -0.7507395360105985,
297
+ "content_mean_predicted_variance": 0.15950868994249426,
298
+ "integrity_accuracy": 0.96755,
299
+ "integrity_macro_f1": 0.5300005982213699,
300
+ "critical_average_precision": 0.182803358663597,
301
+ "critical_roc_auc": 0.9674798301993486,
302
+ "critical_f1_at_0_5": 0.21727019498607242,
303
+ "critical_brier": 0.004237299815054516,
304
+ "final_mae": 0.2676259289680868,
305
+ "final_macro_mae": 0.3058460809528486,
306
+ "final_exact_accuracy": 0.80595,
307
+ "final_spearman": 0.903772047903423,
308
+ "integrity_confusion_matrix": [
309
+ [
310
+ 57547,
311
+ 1002,
312
+ 171
313
+ ],
314
+ [
315
+ 581,
316
+ 442,
317
+ 47
318
+ ],
319
+ [
320
+ 103,
321
+ 43,
322
+ 64
323
+ ]
324
+ ],
325
+ "content_distribution": {
326
+ "0": 5871,
327
+ "1": 28467,
328
+ "2": 14551,
329
+ "3": 7474,
330
+ "4": 3637
331
+ },
332
+ "final_distribution": {
333
+ "0": 5871,
334
+ "1": 28467,
335
+ "2": 14702,
336
+ "3": 7568,
337
+ "4": 3392
338
+ },
339
+ "content_per_class": {
340
+ "0": {
341
+ "precision": 0.8763413387838528,
342
+ "recall": 0.7639198218262806,
343
+ "f1": 0.8162779628748215,
344
+ "support": 6735
345
+ },
346
+ "1": {
347
+ "precision": 0.8691818596971932,
348
+ "recall": 0.8858298725476157,
349
+ "f1": 0.8774269047323534,
350
+ "support": 27932
351
+ },
352
+ "2": {
353
+ "precision": 0.7242801182049343,
354
+ "recall": 0.7394751613808588,
355
+ "f1": 0.7317987709613581,
356
+ "support": 14252
357
+ },
358
+ "3": {
359
+ "precision": 0.7118009098207118,
360
+ "recall": 0.6794380587484036,
361
+ "f1": 0.6952430737062206,
362
+ "support": 7830
363
+ },
364
+ "4": {
365
+ "precision": 0.7330217211987902,
366
+ "recall": 0.8200553675792064,
367
+ "f1": 0.7740998838559814,
368
+ "support": 3251
369
+ }
370
+ },
371
+ "content_macro_f1": 0.7789693192261471,
372
+ "content_f1_class2_3_mean": 0.7135209223337894,
373
+ "content_mean_variance_correct": 0.13864480555917522,
374
+ "content_mean_variance_wrong": 0.24668252567647403,
375
+ "content_wrong_to_correct_variance_ratio": 1.77924102299806,
376
+ "content_variance_wrong_auroc": 0.7460495586192982,
377
+ "content_variance_wrong_average_precision": 0.35514109972948527,
378
+ "content_one_sigma_accuracy": 0.8032666666666667,
379
+ "content_fixed_half_point_accuracy": 0.8068833333333333,
380
+ "content_mean_predicted_sigma": 0.35620977522290764,
381
+ "content_mean_absolute_calibration_gap": 0.19302945969064433
382
+ }
383
+ }
384
+ },
385
+ "logged_at": "2026-10-04T00:33:26.173277+00:00"
386
+ }