PedramR commited on
Commit
ecf4c4e
·
verified ·
1 Parent(s): 51a91bc

decode sweep

Browse files
edit-qwen3.5-2b/decode_sweep/sweep.json ADDED
@@ -0,0 +1,863 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "validation": {
3
+ "p_none_auc": 0.7889476778365667,
4
+ "mean_p_none_with_edit": 0.5160003902893218,
5
+ "mean_p_none_without_edit": 0.6349248286382652,
6
+ "by_threshold": {
7
+ "0.0": {
8
+ "windows": 144,
9
+ "edit_precision": 0.0,
10
+ "edit_recall": 0.0,
11
+ "window_exact_match": 0.5555555555555556,
12
+ "rejected_lines": 1,
13
+ "predicted_edits": 2,
14
+ "gold_edits": 96
15
+ },
16
+ "0.5": {
17
+ "windows": 144,
18
+ "edit_precision": 0.0,
19
+ "edit_recall": 0.0,
20
+ "window_exact_match": 0.4930555555555556,
21
+ "rejected_lines": 4,
22
+ "predicted_edits": 37,
23
+ "gold_edits": 96
24
+ },
25
+ "0.6": {
26
+ "windows": 144,
27
+ "edit_precision": 0.0,
28
+ "edit_recall": 0.0,
29
+ "window_exact_match": 0.4097222222222222,
30
+ "rejected_lines": 13,
31
+ "predicted_edits": 64,
32
+ "gold_edits": 96
33
+ },
34
+ "0.7": {
35
+ "windows": 144,
36
+ "edit_precision": 0.0,
37
+ "edit_recall": 0.0,
38
+ "window_exact_match": 0.2361111111111111,
39
+ "rejected_lines": 24,
40
+ "predicted_edits": 93,
41
+ "gold_edits": 96
42
+ },
43
+ "0.8": {
44
+ "windows": 144,
45
+ "edit_precision": 0.0,
46
+ "edit_recall": 0.0,
47
+ "window_exact_match": 0.14583333333333334,
48
+ "rejected_lines": 29,
49
+ "predicted_edits": 109,
50
+ "gold_edits": 96
51
+ },
52
+ "0.9": {
53
+ "windows": 144,
54
+ "edit_precision": 0.0,
55
+ "edit_recall": 0.0,
56
+ "window_exact_match": 0.13194444444444445,
57
+ "rejected_lines": 33,
58
+ "predicted_edits": 111,
59
+ "gold_edits": 96
60
+ },
61
+ "0.95": {
62
+ "windows": 144,
63
+ "edit_precision": 0.0,
64
+ "edit_recall": 0.0,
65
+ "window_exact_match": 0.13194444444444445,
66
+ "rejected_lines": 33,
67
+ "predicted_edits": 111,
68
+ "gold_edits": 96
69
+ },
70
+ "0.98": {
71
+ "windows": 144,
72
+ "edit_precision": 0.0,
73
+ "edit_recall": 0.0,
74
+ "window_exact_match": 0.13194444444444445,
75
+ "rejected_lines": 33,
76
+ "predicted_edits": 111,
77
+ "gold_edits": 96
78
+ },
79
+ "0.99": {
80
+ "windows": 144,
81
+ "edit_precision": 0.0,
82
+ "edit_recall": 0.0,
83
+ "window_exact_match": 0.13194444444444445,
84
+ "rejected_lines": 33,
85
+ "predicted_edits": 111,
86
+ "gold_edits": 96
87
+ }
88
+ }
89
+ },
90
+ "test": {
91
+ "transcripts": 300,
92
+ "windows": 1963,
93
+ "by_threshold": {
94
+ "0.0": {
95
+ "main": {
96
+ "n_examples": 300,
97
+ "wer": 0.1966254624199224,
98
+ "wer_zwnj_normalized": 0.18451682757376162,
99
+ "exact_match": 0.0033333333333333335,
100
+ "hallucination_rate": 0.0,
101
+ "rows_changed": 0.6833333333333333,
102
+ "fix_rate": 0.052719092331768386,
103
+ "preservation_rate": 0.9997344604014073,
104
+ "targeted_score": 0.5262267763665879,
105
+ "fix_rate_lenient": 0.05560675883256529,
106
+ "preservation_rate_lenient": 0.9996691371095818,
107
+ "targeted_score_lenient": 0.5276379479710736,
108
+ "fixed_lenient": 543,
109
+ "broken_lenient": 15,
110
+ "net_fixed_lenient": 528,
111
+ "broken_per_100_correct": 0.03308628904182107,
112
+ "fix_rate_recoverable": 0.05654668753260302,
113
+ "unrecoverable_targets": 180,
114
+ "entity_fix_rate": 0.0,
115
+ "entity_keep_rate": 1.0,
116
+ "entity_spans_whisper_wrong": 2,
117
+ "entity_spans_whisper_right": 14,
118
+ "model_edits": 5
119
+ },
120
+ "entity": {
121
+ "n_examples": 145,
122
+ "wer": 0.1965140984551118,
123
+ "wer_zwnj_normalized": 0.1845943318088516,
124
+ "exact_match": 0.006896551724137931,
125
+ "hallucination_rate": 0.0,
126
+ "rows_changed": 0.7310344827586207,
127
+ "fix_rate": 0.05367675973259929,
128
+ "preservation_rate": 0.99969139884495,
129
+ "targeted_score": 0.5266840792887746,
130
+ "fix_rate_lenient": 0.05730659025787966,
131
+ "preservation_rate_lenient": 0.9997360665112391,
132
+ "targeted_score_lenient": 0.5285213283845595,
133
+ "fixed_lenient": 280,
134
+ "broken_lenient": 6,
135
+ "net_fixed_lenient": 274,
136
+ "broken_per_100_correct": 0.026393348876083227,
137
+ "fix_rate_recoverable": 0.05810079133694294,
138
+ "unrecoverable_targets": 84,
139
+ "entity_fix_rate": 0.0,
140
+ "entity_keep_rate": 1.0,
141
+ "entity_spans_whisper_wrong": 2,
142
+ "entity_spans_whisper_right": 14,
143
+ "model_edits": 3
144
+ },
145
+ "typo": {
146
+ "n_examples": 45,
147
+ "wer": 0.1501865671641791,
148
+ "wer_zwnj_normalized": 0.1369402985074627,
149
+ "exact_match": 0.0,
150
+ "hallucination_rate": 0.0,
151
+ "rows_changed": 0.4666666666666667,
152
+ "fix_rate": 0.07178217821782178,
153
+ "preservation_rate": 1.0,
154
+ "targeted_score": 0.5358910891089109,
155
+ "fix_rate_lenient": 0.077733860342556,
156
+ "preservation_rate_lenient": 1.0,
157
+ "targeted_score_lenient": 0.538866930171278,
158
+ "fixed_lenient": 59,
159
+ "broken_lenient": 0,
160
+ "net_fixed_lenient": 59,
161
+ "broken_per_100_correct": 0.0,
162
+ "fix_rate_recoverable": 0.07835325365205843,
163
+ "unrecoverable_targets": 6,
164
+ "entity_fix_rate": 0.0,
165
+ "entity_keep_rate": 1.0,
166
+ "entity_spans_whisper_wrong": 1,
167
+ "entity_spans_whisper_right": 4,
168
+ "model_edits": 0
169
+ }
170
+ },
171
+ "0.5": {
172
+ "main": {
173
+ "n_examples": 300,
174
+ "wer": 0.19727510601822612,
175
+ "wer_zwnj_normalized": 0.1852025624830822,
176
+ "exact_match": 0.0033333333333333335,
177
+ "hallucination_rate": 0.0,
178
+ "rows_changed": 0.7633333333333333,
179
+ "fix_rate": 0.05506651017214397,
180
+ "preservation_rate": 0.9984288907083269,
181
+ "targeted_score": 0.5267477004402354,
182
+ "fix_rate_lenient": 0.0579621095750128,
183
+ "preservation_rate_lenient": 0.9983677430739368,
184
+ "targeted_score_lenient": 0.5281649263244748,
185
+ "fixed_lenient": 566,
186
+ "broken_lenient": 74,
187
+ "net_fixed_lenient": 492,
188
+ "broken_per_100_correct": 0.16322569260631728,
189
+ "fix_rate_recoverable": 0.05894627021387585,
190
+ "unrecoverable_targets": 180,
191
+ "entity_fix_rate": 0.0,
192
+ "entity_keep_rate": 1.0,
193
+ "entity_spans_whisper_wrong": 2,
194
+ "entity_spans_whisper_right": 14,
195
+ "model_edits": 162
196
+ },
197
+ "entity": {
198
+ "n_examples": 145,
199
+ "wer": 0.19748640570420253,
200
+ "wer_zwnj_normalized": 0.18556663905794232,
201
+ "exact_match": 0.006896551724137931,
202
+ "hallucination_rate": 0.0,
203
+ "rows_changed": 0.7931034482758621,
204
+ "fix_rate": 0.05524970507274872,
205
+ "preservation_rate": 0.9982365648282855,
206
+ "targeted_score": 0.5267431349505171,
207
+ "fix_rate_lenient": 0.05873925501432665,
208
+ "preservation_rate_lenient": 0.9982844323230546,
209
+ "targeted_score_lenient": 0.5285118436686906,
210
+ "fixed_lenient": 287,
211
+ "broken_lenient": 39,
212
+ "net_fixed_lenient": 248,
213
+ "broken_per_100_correct": 0.17155676769454098,
214
+ "fix_rate_recoverable": 0.059558517284464804,
215
+ "unrecoverable_targets": 84,
216
+ "entity_fix_rate": 0.0,
217
+ "entity_keep_rate": 1.0,
218
+ "entity_spans_whisper_wrong": 2,
219
+ "entity_spans_whisper_right": 14,
220
+ "model_edits": 84
221
+ },
222
+ "typo": {
223
+ "n_examples": 45,
224
+ "wer": 0.15205223880597016,
225
+ "wer_zwnj_normalized": 0.13880597014925372,
226
+ "exact_match": 0.0,
227
+ "hallucination_rate": 0.0,
228
+ "rows_changed": 0.5777777777777777,
229
+ "fix_rate": 0.07301980198019802,
230
+ "preservation_rate": 0.9975834797891037,
231
+ "targeted_score": 0.5353016408846508,
232
+ "fix_rate_lenient": 0.07905138339920949,
233
+ "preservation_rate_lenient": 0.9975893052816129,
234
+ "targeted_score_lenient": 0.5383203443404112,
235
+ "fixed_lenient": 60,
236
+ "broken_lenient": 11,
237
+ "net_fixed_lenient": 49,
238
+ "broken_per_100_correct": 0.24106947183870261,
239
+ "fix_rate_recoverable": 0.0796812749003984,
240
+ "unrecoverable_targets": 6,
241
+ "entity_fix_rate": 0.0,
242
+ "entity_keep_rate": 1.0,
243
+ "entity_spans_whisper_wrong": 1,
244
+ "entity_spans_whisper_right": 4,
245
+ "model_edits": 13
246
+ }
247
+ },
248
+ "0.6": {
249
+ "main": {
250
+ "n_examples": 300,
251
+ "wer": 0.1992059911576288,
252
+ "wer_zwnj_normalized": 0.18722367590002706,
253
+ "exact_match": 0.0033333333333333335,
254
+ "hallucination_rate": 0.0,
255
+ "rows_changed": 0.8466666666666667,
256
+ "fix_rate": 0.05888106416275431,
257
+ "preservation_rate": 0.9952202872253325,
258
+ "targeted_score": 0.5270506756940434,
259
+ "fix_rate_lenient": 0.06175115207373272,
260
+ "preservation_rate_lenient": 0.9951032292218105,
261
+ "targeted_score_lenient": 0.5284271906477715,
262
+ "fixed_lenient": 603,
263
+ "broken_lenient": 222,
264
+ "net_fixed_lenient": 381,
265
+ "broken_per_100_correct": 0.4896770778189518,
266
+ "fix_rate_recoverable": 0.06280646844027125,
267
+ "unrecoverable_targets": 180,
268
+ "entity_fix_rate": 0.0,
269
+ "entity_keep_rate": 1.0,
270
+ "entity_spans_whisper_wrong": 2,
271
+ "entity_spans_whisper_right": 14,
272
+ "model_edits": 447
273
+ },
274
+ "entity": {
275
+ "n_examples": 145,
276
+ "wer": 0.19907090640642444,
277
+ "wer_zwnj_normalized": 0.18718715113976017,
278
+ "exact_match": 0.006896551724137931,
279
+ "hallucination_rate": 0.0,
280
+ "rows_changed": 0.8827586206896552,
281
+ "fix_rate": 0.05859221392056626,
282
+ "preservation_rate": 0.9954591544328352,
283
+ "targeted_score": 0.5270256841767007,
284
+ "fix_rate_lenient": 0.06180925092099877,
285
+ "preservation_rate_lenient": 0.995337175031892,
286
+ "targeted_score_lenient": 0.5285732129764454,
287
+ "fixed_lenient": 302,
288
+ "broken_lenient": 106,
289
+ "net_fixed_lenient": 196,
290
+ "broken_per_100_correct": 0.4662824968108037,
291
+ "fix_rate_recoverable": 0.06268221574344024,
292
+ "unrecoverable_targets": 84,
293
+ "entity_fix_rate": 0.0,
294
+ "entity_keep_rate": 1.0,
295
+ "entity_spans_whisper_wrong": 2,
296
+ "entity_spans_whisper_right": 14,
297
+ "model_edits": 225
298
+ },
299
+ "typo": {
300
+ "n_examples": 45,
301
+ "wer": 0.1541044776119403,
302
+ "wer_zwnj_normalized": 0.14085820895522388,
303
+ "exact_match": 0.0,
304
+ "hallucination_rate": 0.0,
305
+ "rows_changed": 0.6666666666666666,
306
+ "fix_rate": 0.07549504950495049,
307
+ "preservation_rate": 0.9951669595782073,
308
+ "targeted_score": 0.5353310045415789,
309
+ "fix_rate_lenient": 0.08168642951251646,
310
+ "preservation_rate_lenient": 0.995178610563226,
311
+ "targeted_score_lenient": 0.5384325200378712,
312
+ "fixed_lenient": 62,
313
+ "broken_lenient": 22,
314
+ "net_fixed_lenient": 40,
315
+ "broken_per_100_correct": 0.48213894367740523,
316
+ "fix_rate_recoverable": 0.08233731739707835,
317
+ "unrecoverable_targets": 6,
318
+ "entity_fix_rate": 0.0,
319
+ "entity_keep_rate": 1.0,
320
+ "entity_spans_whisper_wrong": 1,
321
+ "entity_spans_whisper_right": 4,
322
+ "model_edits": 31
323
+ }
324
+ },
325
+ "0.7": {
326
+ "main": {
327
+ "n_examples": 300,
328
+ "wer": 0.20517910313092122,
329
+ "wer_zwnj_normalized": 0.19326897049535324,
330
+ "exact_match": 0.0033333333333333335,
331
+ "hallucination_rate": 0.0,
332
+ "rows_changed": 0.9266666666666666,
333
+ "fix_rate": 0.061130672926447575,
334
+ "preservation_rate": 0.9876524086654422,
335
+ "targeted_score": 0.5243915407959449,
336
+ "fix_rate_lenient": 0.06328725038402458,
337
+ "preservation_rate_lenient": 0.9875154402682195,
338
+ "targeted_score_lenient": 0.525401345326122,
339
+ "fixed_lenient": 618,
340
+ "broken_lenient": 566,
341
+ "net_fixed_lenient": 52,
342
+ "broken_per_100_correct": 1.2484559731780482,
343
+ "fix_rate_recoverable": 0.06437141366718832,
344
+ "unrecoverable_targets": 180,
345
+ "entity_fix_rate": 0.0,
346
+ "entity_keep_rate": 1.0,
347
+ "entity_spans_whisper_wrong": 2,
348
+ "entity_spans_whisper_right": 14,
349
+ "model_edits": 884
350
+ },
351
+ "entity": {
352
+ "n_examples": 145,
353
+ "wer": 0.20461665886420108,
354
+ "wer_zwnj_normalized": 0.19280492635672872,
355
+ "exact_match": 0.006896551724137931,
356
+ "hallucination_rate": 0.0,
357
+ "rows_changed": 0.9517241379310345,
358
+ "fix_rate": 0.06154148643334644,
359
+ "preservation_rate": 0.9882731561080986,
360
+ "targeted_score": 0.5249073212707225,
361
+ "fix_rate_lenient": 0.06385591485878019,
362
+ "preservation_rate_lenient": 0.9882989486649364,
363
+ "targeted_score_lenient": 0.5260774317618583,
364
+ "fixed_lenient": 312,
365
+ "broken_lenient": 266,
366
+ "net_fixed_lenient": 46,
367
+ "broken_per_100_correct": 1.1701051335063564,
368
+ "fix_rate_recoverable": 0.06476468138275718,
369
+ "unrecoverable_targets": 84,
370
+ "entity_fix_rate": 0.0,
371
+ "entity_keep_rate": 1.0,
372
+ "entity_spans_whisper_wrong": 2,
373
+ "entity_spans_whisper_right": 14,
374
+ "model_edits": 438
375
+ },
376
+ "typo": {
377
+ "n_examples": 45,
378
+ "wer": 0.16026119402985076,
379
+ "wer_zwnj_normalized": 0.14682835820895523,
380
+ "exact_match": 0.0,
381
+ "hallucination_rate": 0.0,
382
+ "rows_changed": 0.8444444444444444,
383
+ "fix_rate": 0.07797029702970297,
384
+ "preservation_rate": 0.9876977152899824,
385
+ "targeted_score": 0.5328340061598427,
386
+ "fix_rate_lenient": 0.08432147562582346,
387
+ "preservation_rate_lenient": 0.987727372342757,
388
+ "targeted_score_lenient": 0.5360244239842902,
389
+ "fixed_lenient": 64,
390
+ "broken_lenient": 56,
391
+ "net_fixed_lenient": 8,
392
+ "broken_per_100_correct": 1.2272627657243043,
393
+ "fix_rate_recoverable": 0.0849933598937583,
394
+ "unrecoverable_targets": 6,
395
+ "entity_fix_rate": 0.0,
396
+ "entity_keep_rate": 1.0,
397
+ "entity_spans_whisper_wrong": 1,
398
+ "entity_spans_whisper_right": 4,
399
+ "model_edits": 71
400
+ }
401
+ },
402
+ "0.8": {
403
+ "main": {
404
+ "n_examples": 300,
405
+ "wer": 0.20768744924659388,
406
+ "wer_zwnj_normalized": 0.19583145357755122,
407
+ "exact_match": 0.0033333333333333335,
408
+ "hallucination_rate": 0.0,
409
+ "rows_changed": 0.9566666666666667,
410
+ "fix_rate": 0.06122848200312989,
411
+ "preservation_rate": 0.9845987032816269,
412
+ "targeted_score": 0.5229135926423784,
413
+ "fix_rate_lenient": 0.06338965693804403,
414
+ "preservation_rate_lenient": 0.984471501676372,
415
+ "targeted_score_lenient": 0.523930579307208,
416
+ "fixed_lenient": 619,
417
+ "broken_lenient": 704,
418
+ "net_fixed_lenient": -85,
419
+ "broken_per_100_correct": 1.5528498323628022,
420
+ "fix_rate_recoverable": 0.06447574334898279,
421
+ "unrecoverable_targets": 180,
422
+ "entity_fix_rate": 0.0,
423
+ "entity_keep_rate": 1.0,
424
+ "entity_spans_whisper_wrong": 2,
425
+ "entity_spans_whisper_right": 14,
426
+ "model_edits": 1036
427
+ },
428
+ "entity": {
429
+ "n_examples": 145,
430
+ "wer": 0.2073175123338975,
431
+ "wer_zwnj_normalized": 0.19557780258561705,
432
+ "exact_match": 0.006896551724137931,
433
+ "hallucination_rate": 0.0,
434
+ "rows_changed": 0.9793103448275862,
435
+ "fix_rate": 0.0609516319307904,
436
+ "preservation_rate": 0.9850989727990125,
437
+ "targeted_score": 0.5230253023649014,
438
+ "fix_rate_lenient": 0.06324191567744576,
439
+ "preservation_rate_lenient": 0.985087757885013,
440
+ "targeted_score_lenient": 0.5241648367812294,
441
+ "fixed_lenient": 309,
442
+ "broken_lenient": 339,
443
+ "net_fixed_lenient": -30,
444
+ "broken_per_100_correct": 1.4912242114987024,
445
+ "fix_rate_recoverable": 0.0641399416909621,
446
+ "unrecoverable_targets": 84,
447
+ "entity_fix_rate": 0.0,
448
+ "entity_keep_rate": 1.0,
449
+ "entity_spans_whisper_wrong": 2,
450
+ "entity_spans_whisper_right": 14,
451
+ "model_edits": 516
452
+ },
453
+ "typo": {
454
+ "n_examples": 45,
455
+ "wer": 0.16399253731343283,
456
+ "wer_zwnj_normalized": 0.1505597014925373,
457
+ "exact_match": 0.0,
458
+ "hallucination_rate": 0.0,
459
+ "rows_changed": 0.8888888888888888,
460
+ "fix_rate": 0.07797029702970297,
461
+ "preservation_rate": 0.9835237258347979,
462
+ "targeted_score": 0.5307470114322504,
463
+ "fix_rate_lenient": 0.08432147562582346,
464
+ "preservation_rate_lenient": 0.9835634451019066,
465
+ "targeted_score_lenient": 0.5339424603638651,
466
+ "fixed_lenient": 64,
467
+ "broken_lenient": 75,
468
+ "net_fixed_lenient": -11,
469
+ "broken_per_100_correct": 1.643655489809336,
470
+ "fix_rate_recoverable": 0.0849933598937583,
471
+ "unrecoverable_targets": 6,
472
+ "entity_fix_rate": 0.0,
473
+ "entity_keep_rate": 1.0,
474
+ "entity_spans_whisper_wrong": 1,
475
+ "entity_spans_whisper_right": 4,
476
+ "model_edits": 91
477
+ }
478
+ },
479
+ "0.9": {
480
+ "main": {
481
+ "n_examples": 300,
482
+ "wer": 0.2077415862131192,
483
+ "wer_zwnj_normalized": 0.1958855905440765,
484
+ "exact_match": 0.0,
485
+ "hallucination_rate": 0.0,
486
+ "rows_changed": 0.9766666666666667,
487
+ "fix_rate": 0.061326291079812204,
488
+ "preservation_rate": 0.9845323183819787,
489
+ "targeted_score": 0.5229293047308955,
490
+ "fix_rate_lenient": 0.06349206349206349,
491
+ "preservation_rate_lenient": 0.9844053290982884,
492
+ "targeted_score_lenient": 0.5239486962951759,
493
+ "fixed_lenient": 620,
494
+ "broken_lenient": 707,
495
+ "net_fixed_lenient": -87,
496
+ "broken_per_100_correct": 1.5594670901711665,
497
+ "fix_rate_recoverable": 0.06458007303077726,
498
+ "unrecoverable_targets": 180,
499
+ "entity_fix_rate": 0.0,
500
+ "entity_keep_rate": 1.0,
501
+ "entity_spans_whisper_wrong": 2,
502
+ "entity_spans_whisper_right": 14,
503
+ "model_edits": 1042
504
+ },
505
+ "entity": {
506
+ "n_examples": 145,
507
+ "wer": 0.20735352371349347,
508
+ "wer_zwnj_normalized": 0.19561381396521302,
509
+ "exact_match": 0.0,
510
+ "hallucination_rate": 0.0,
511
+ "rows_changed": 0.9862068965517241,
512
+ "fix_rate": 0.0609516319307904,
513
+ "preservation_rate": 0.9850548869197197,
514
+ "targeted_score": 0.523003259425255,
515
+ "fix_rate_lenient": 0.06324191567744576,
516
+ "preservation_rate_lenient": 0.9850437689702195,
517
+ "targeted_score_lenient": 0.5241428423238327,
518
+ "fixed_lenient": 309,
519
+ "broken_lenient": 340,
520
+ "net_fixed_lenient": -31,
521
+ "broken_per_100_correct": 1.4956231029780496,
522
+ "fix_rate_recoverable": 0.0641399416909621,
523
+ "unrecoverable_targets": 84,
524
+ "entity_fix_rate": 0.0,
525
+ "entity_keep_rate": 1.0,
526
+ "entity_spans_whisper_wrong": 2,
527
+ "entity_spans_whisper_right": 14,
528
+ "model_edits": 517
529
+ },
530
+ "typo": {
531
+ "n_examples": 45,
532
+ "wer": 0.16417910447761194,
533
+ "wer_zwnj_normalized": 0.15074626865671642,
534
+ "exact_match": 0.0,
535
+ "hallucination_rate": 0.0,
536
+ "rows_changed": 0.9333333333333333,
537
+ "fix_rate": 0.07797029702970297,
538
+ "preservation_rate": 0.9835237258347979,
539
+ "targeted_score": 0.5307470114322504,
540
+ "fix_rate_lenient": 0.08432147562582346,
541
+ "preservation_rate_lenient": 0.9835634451019066,
542
+ "targeted_score_lenient": 0.5339424603638651,
543
+ "fixed_lenient": 64,
544
+ "broken_lenient": 75,
545
+ "net_fixed_lenient": -11,
546
+ "broken_per_100_correct": 1.643655489809336,
547
+ "fix_rate_recoverable": 0.0849933598937583,
548
+ "unrecoverable_targets": 6,
549
+ "entity_fix_rate": 0.0,
550
+ "entity_keep_rate": 1.0,
551
+ "entity_spans_whisper_wrong": 1,
552
+ "entity_spans_whisper_right": 4,
553
+ "model_edits": 93
554
+ }
555
+ },
556
+ "0.95": {
557
+ "main": {
558
+ "n_examples": 300,
559
+ "wer": 0.2077415862131192,
560
+ "wer_zwnj_normalized": 0.1958855905440765,
561
+ "exact_match": 0.0,
562
+ "hallucination_rate": 0.0,
563
+ "rows_changed": 0.9766666666666667,
564
+ "fix_rate": 0.061326291079812204,
565
+ "preservation_rate": 0.9845323183819787,
566
+ "targeted_score": 0.5229293047308955,
567
+ "fix_rate_lenient": 0.06349206349206349,
568
+ "preservation_rate_lenient": 0.9844053290982884,
569
+ "targeted_score_lenient": 0.5239486962951759,
570
+ "fixed_lenient": 620,
571
+ "broken_lenient": 707,
572
+ "net_fixed_lenient": -87,
573
+ "broken_per_100_correct": 1.5594670901711665,
574
+ "fix_rate_recoverable": 0.06458007303077726,
575
+ "unrecoverable_targets": 180,
576
+ "entity_fix_rate": 0.0,
577
+ "entity_keep_rate": 1.0,
578
+ "entity_spans_whisper_wrong": 2,
579
+ "entity_spans_whisper_right": 14,
580
+ "model_edits": 1042
581
+ },
582
+ "entity": {
583
+ "n_examples": 145,
584
+ "wer": 0.20735352371349347,
585
+ "wer_zwnj_normalized": 0.19561381396521302,
586
+ "exact_match": 0.0,
587
+ "hallucination_rate": 0.0,
588
+ "rows_changed": 0.9862068965517241,
589
+ "fix_rate": 0.0609516319307904,
590
+ "preservation_rate": 0.9850548869197197,
591
+ "targeted_score": 0.523003259425255,
592
+ "fix_rate_lenient": 0.06324191567744576,
593
+ "preservation_rate_lenient": 0.9850437689702195,
594
+ "targeted_score_lenient": 0.5241428423238327,
595
+ "fixed_lenient": 309,
596
+ "broken_lenient": 340,
597
+ "net_fixed_lenient": -31,
598
+ "broken_per_100_correct": 1.4956231029780496,
599
+ "fix_rate_recoverable": 0.0641399416909621,
600
+ "unrecoverable_targets": 84,
601
+ "entity_fix_rate": 0.0,
602
+ "entity_keep_rate": 1.0,
603
+ "entity_spans_whisper_wrong": 2,
604
+ "entity_spans_whisper_right": 14,
605
+ "model_edits": 517
606
+ },
607
+ "typo": {
608
+ "n_examples": 45,
609
+ "wer": 0.16417910447761194,
610
+ "wer_zwnj_normalized": 0.15074626865671642,
611
+ "exact_match": 0.0,
612
+ "hallucination_rate": 0.0,
613
+ "rows_changed": 0.9333333333333333,
614
+ "fix_rate": 0.07797029702970297,
615
+ "preservation_rate": 0.9835237258347979,
616
+ "targeted_score": 0.5307470114322504,
617
+ "fix_rate_lenient": 0.08432147562582346,
618
+ "preservation_rate_lenient": 0.9835634451019066,
619
+ "targeted_score_lenient": 0.5339424603638651,
620
+ "fixed_lenient": 64,
621
+ "broken_lenient": 75,
622
+ "net_fixed_lenient": -11,
623
+ "broken_per_100_correct": 1.643655489809336,
624
+ "fix_rate_recoverable": 0.0849933598937583,
625
+ "unrecoverable_targets": 6,
626
+ "entity_fix_rate": 0.0,
627
+ "entity_keep_rate": 1.0,
628
+ "entity_spans_whisper_wrong": 1,
629
+ "entity_spans_whisper_right": 4,
630
+ "model_edits": 93
631
+ }
632
+ },
633
+ "0.98": {
634
+ "main": {
635
+ "n_examples": 300,
636
+ "wer": 0.2077415862131192,
637
+ "wer_zwnj_normalized": 0.1958855905440765,
638
+ "exact_match": 0.0,
639
+ "hallucination_rate": 0.0,
640
+ "rows_changed": 0.9766666666666667,
641
+ "fix_rate": 0.061326291079812204,
642
+ "preservation_rate": 0.9845323183819787,
643
+ "targeted_score": 0.5229293047308955,
644
+ "fix_rate_lenient": 0.06349206349206349,
645
+ "preservation_rate_lenient": 0.9844053290982884,
646
+ "targeted_score_lenient": 0.5239486962951759,
647
+ "fixed_lenient": 620,
648
+ "broken_lenient": 707,
649
+ "net_fixed_lenient": -87,
650
+ "broken_per_100_correct": 1.5594670901711665,
651
+ "fix_rate_recoverable": 0.06458007303077726,
652
+ "unrecoverable_targets": 180,
653
+ "entity_fix_rate": 0.0,
654
+ "entity_keep_rate": 1.0,
655
+ "entity_spans_whisper_wrong": 2,
656
+ "entity_spans_whisper_right": 14,
657
+ "model_edits": 1042
658
+ },
659
+ "entity": {
660
+ "n_examples": 145,
661
+ "wer": 0.20735352371349347,
662
+ "wer_zwnj_normalized": 0.19561381396521302,
663
+ "exact_match": 0.0,
664
+ "hallucination_rate": 0.0,
665
+ "rows_changed": 0.9862068965517241,
666
+ "fix_rate": 0.0609516319307904,
667
+ "preservation_rate": 0.9850548869197197,
668
+ "targeted_score": 0.523003259425255,
669
+ "fix_rate_lenient": 0.06324191567744576,
670
+ "preservation_rate_lenient": 0.9850437689702195,
671
+ "targeted_score_lenient": 0.5241428423238327,
672
+ "fixed_lenient": 309,
673
+ "broken_lenient": 340,
674
+ "net_fixed_lenient": -31,
675
+ "broken_per_100_correct": 1.4956231029780496,
676
+ "fix_rate_recoverable": 0.0641399416909621,
677
+ "unrecoverable_targets": 84,
678
+ "entity_fix_rate": 0.0,
679
+ "entity_keep_rate": 1.0,
680
+ "entity_spans_whisper_wrong": 2,
681
+ "entity_spans_whisper_right": 14,
682
+ "model_edits": 517
683
+ },
684
+ "typo": {
685
+ "n_examples": 45,
686
+ "wer": 0.16417910447761194,
687
+ "wer_zwnj_normalized": 0.15074626865671642,
688
+ "exact_match": 0.0,
689
+ "hallucination_rate": 0.0,
690
+ "rows_changed": 0.9333333333333333,
691
+ "fix_rate": 0.07797029702970297,
692
+ "preservation_rate": 0.9835237258347979,
693
+ "targeted_score": 0.5307470114322504,
694
+ "fix_rate_lenient": 0.08432147562582346,
695
+ "preservation_rate_lenient": 0.9835634451019066,
696
+ "targeted_score_lenient": 0.5339424603638651,
697
+ "fixed_lenient": 64,
698
+ "broken_lenient": 75,
699
+ "net_fixed_lenient": -11,
700
+ "broken_per_100_correct": 1.643655489809336,
701
+ "fix_rate_recoverable": 0.0849933598937583,
702
+ "unrecoverable_targets": 6,
703
+ "entity_fix_rate": 0.0,
704
+ "entity_keep_rate": 1.0,
705
+ "entity_spans_whisper_wrong": 1,
706
+ "entity_spans_whisper_right": 4,
707
+ "model_edits": 93
708
+ }
709
+ },
710
+ "0.99": {
711
+ "main": {
712
+ "n_examples": 300,
713
+ "wer": 0.2077415862131192,
714
+ "wer_zwnj_normalized": 0.1958855905440765,
715
+ "exact_match": 0.0,
716
+ "hallucination_rate": 0.0,
717
+ "rows_changed": 0.9766666666666667,
718
+ "fix_rate": 0.061326291079812204,
719
+ "preservation_rate": 0.9845323183819787,
720
+ "targeted_score": 0.5229293047308955,
721
+ "fix_rate_lenient": 0.06349206349206349,
722
+ "preservation_rate_lenient": 0.9844053290982884,
723
+ "targeted_score_lenient": 0.5239486962951759,
724
+ "fixed_lenient": 620,
725
+ "broken_lenient": 707,
726
+ "net_fixed_lenient": -87,
727
+ "broken_per_100_correct": 1.5594670901711665,
728
+ "fix_rate_recoverable": 0.06458007303077726,
729
+ "unrecoverable_targets": 180,
730
+ "entity_fix_rate": 0.0,
731
+ "entity_keep_rate": 1.0,
732
+ "entity_spans_whisper_wrong": 2,
733
+ "entity_spans_whisper_right": 14,
734
+ "model_edits": 1042
735
+ },
736
+ "entity": {
737
+ "n_examples": 145,
738
+ "wer": 0.20735352371349347,
739
+ "wer_zwnj_normalized": 0.19561381396521302,
740
+ "exact_match": 0.0,
741
+ "hallucination_rate": 0.0,
742
+ "rows_changed": 0.9862068965517241,
743
+ "fix_rate": 0.0609516319307904,
744
+ "preservation_rate": 0.9850548869197197,
745
+ "targeted_score": 0.523003259425255,
746
+ "fix_rate_lenient": 0.06324191567744576,
747
+ "preservation_rate_lenient": 0.9850437689702195,
748
+ "targeted_score_lenient": 0.5241428423238327,
749
+ "fixed_lenient": 309,
750
+ "broken_lenient": 340,
751
+ "net_fixed_lenient": -31,
752
+ "broken_per_100_correct": 1.4956231029780496,
753
+ "fix_rate_recoverable": 0.0641399416909621,
754
+ "unrecoverable_targets": 84,
755
+ "entity_fix_rate": 0.0,
756
+ "entity_keep_rate": 1.0,
757
+ "entity_spans_whisper_wrong": 2,
758
+ "entity_spans_whisper_right": 14,
759
+ "model_edits": 517
760
+ },
761
+ "typo": {
762
+ "n_examples": 45,
763
+ "wer": 0.16417910447761194,
764
+ "wer_zwnj_normalized": 0.15074626865671642,
765
+ "exact_match": 0.0,
766
+ "hallucination_rate": 0.0,
767
+ "rows_changed": 0.9333333333333333,
768
+ "fix_rate": 0.07797029702970297,
769
+ "preservation_rate": 0.9835237258347979,
770
+ "targeted_score": 0.5307470114322504,
771
+ "fix_rate_lenient": 0.08432147562582346,
772
+ "preservation_rate_lenient": 0.9835634451019066,
773
+ "targeted_score_lenient": 0.5339424603638651,
774
+ "fixed_lenient": 64,
775
+ "broken_lenient": 75,
776
+ "net_fixed_lenient": -11,
777
+ "broken_per_100_correct": 1.643655489809336,
778
+ "fix_rate_recoverable": 0.0849933598937583,
779
+ "unrecoverable_targets": 6,
780
+ "entity_fix_rate": 0.0,
781
+ "entity_keep_rate": 1.0,
782
+ "entity_spans_whisper_wrong": 1,
783
+ "entity_spans_whisper_right": 4,
784
+ "model_edits": 93
785
+ }
786
+ }
787
+ },
788
+ "rules": {
789
+ "main": {
790
+ "n_examples": 300,
791
+ "wer": 0.19658937110890554,
792
+ "wer_zwnj_normalized": 0.1844626906072363,
793
+ "exact_match": 0.0033333333333333335,
794
+ "hallucination_rate": 0.0,
795
+ "rows_changed": 0.6833333333333333,
796
+ "fix_rate": 0.052719092331768386,
797
+ "preservation_rate": 0.9997787170011728,
798
+ "targeted_score": 0.5262489046664706,
799
+ "fix_rate_lenient": 0.05560675883256529,
800
+ "preservation_rate_lenient": 0.9997132521616375,
801
+ "targeted_score_lenient": 0.5276600054971015,
802
+ "fixed_lenient": 543,
803
+ "broken_lenient": 13,
804
+ "net_fixed_lenient": 530,
805
+ "broken_per_100_correct": 0.028674783836244926,
806
+ "fix_rate_recoverable": 0.05654668753260302,
807
+ "unrecoverable_targets": 180,
808
+ "entity_fix_rate": 0.0,
809
+ "entity_keep_rate": 1.0,
810
+ "entity_spans_whisper_wrong": 2,
811
+ "entity_spans_whisper_right": 14
812
+ },
813
+ "entity": {
814
+ "n_examples": 145,
815
+ "wer": 0.1964420756959199,
816
+ "wer_zwnj_normalized": 0.18448629767006375,
817
+ "exact_match": 0.006896551724137931,
818
+ "hallucination_rate": 0.0,
819
+ "rows_changed": 0.7310344827586207,
820
+ "fix_rate": 0.05367675973259929,
821
+ "preservation_rate": 0.9997795706035357,
822
+ "targeted_score": 0.5267281651680675,
823
+ "fix_rate_lenient": 0.05730659025787966,
824
+ "preservation_rate_lenient": 0.9998240443408261,
825
+ "targeted_score_lenient": 0.5285653172993529,
826
+ "fixed_lenient": 280,
827
+ "broken_lenient": 4,
828
+ "net_fixed_lenient": 276,
829
+ "broken_per_100_correct": 0.017595565917388818,
830
+ "fix_rate_recoverable": 0.05810079133694294,
831
+ "unrecoverable_targets": 84,
832
+ "entity_fix_rate": 0.0,
833
+ "entity_keep_rate": 1.0,
834
+ "entity_spans_whisper_wrong": 2,
835
+ "entity_spans_whisper_right": 14
836
+ },
837
+ "typo": {
838
+ "n_examples": 45,
839
+ "wer": 0.1501865671641791,
840
+ "wer_zwnj_normalized": 0.1369402985074627,
841
+ "exact_match": 0.0,
842
+ "hallucination_rate": 0.0,
843
+ "rows_changed": 0.4666666666666667,
844
+ "fix_rate": 0.07178217821782178,
845
+ "preservation_rate": 1.0,
846
+ "targeted_score": 0.5358910891089109,
847
+ "fix_rate_lenient": 0.077733860342556,
848
+ "preservation_rate_lenient": 1.0,
849
+ "targeted_score_lenient": 0.538866930171278,
850
+ "fixed_lenient": 59,
851
+ "broken_lenient": 0,
852
+ "net_fixed_lenient": 59,
853
+ "broken_per_100_correct": 0.0,
854
+ "fix_rate_recoverable": 0.07835325365205843,
855
+ "unrecoverable_targets": 6,
856
+ "entity_fix_rate": 0.0,
857
+ "entity_keep_rate": 1.0,
858
+ "entity_spans_whisper_wrong": 1,
859
+ "entity_spans_whisper_right": 4
860
+ }
861
+ }
862
+ }
863
+ }
edit-qwen3.5-2b/decode_sweep/sweep.md ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Threshold decoding: edit-qwen3.5-2b
2
+
3
+ Validation (144 windows): p_none AUC 0.789; mean p_none 0.516 on windows needing an edit vs 0.635 on others.
4
+
5
+ | threshold | edit precision | edit recall | predicted | gold | rejected lines |
6
+ |---|---|---|---|---|---|
7
+ | 0.0 | 0.000 | 0.000 | 2 | 96 | 1 |
8
+ | 0.5 | 0.000 | 0.000 | 37 | 96 | 4 |
9
+ | 0.6 | 0.000 | 0.000 | 64 | 96 | 13 |
10
+ | 0.7 | 0.000 | 0.000 | 93 | 96 | 24 |
11
+ | 0.8 | 0.000 | 0.000 | 109 | 96 | 29 |
12
+ | 0.9 | 0.000 | 0.000 | 111 | 96 | 33 |
13
+ | 0.95 | 0.000 | 0.000 | 111 | 96 | 33 |
14
+ | 0.98 | 0.000 | 0.000 | 111 | 96 | 33 |
15
+ | 0.99 | 0.000 | 0.000 | 111 | 96 | 33 |
16
+
17
+ Test, first 300 transcripts (threshold 0.0 = greedy):
18
+
19
+ | slice | threshold | WER | fixed | broken | net | broken/100 | entity fix | model edits |
20
+ |---|---|---|---|---|---|---|---|---|
21
+ | main | rules only | 0.1966 | 543 | 13 | 530 | 0.03 | 0.000 | 0 |
22
+ | main | 0.0 | 0.1966 | 543 | 15 | 528 | 0.03 | 0.000 | 5 |
23
+ | main | 0.5 | 0.1973 | 566 | 74 | 492 | 0.16 | 0.000 | 162 |
24
+ | main | 0.6 | 0.1992 | 603 | 222 | 381 | 0.49 | 0.000 | 447 |
25
+ | main | 0.7 | 0.2052 | 618 | 566 | 52 | 1.25 | 0.000 | 884 |
26
+ | main | 0.8 | 0.2077 | 619 | 704 | -85 | 1.55 | 0.000 | 1036 |
27
+ | main | 0.9 | 0.2077 | 620 | 707 | -87 | 1.56 | 0.000 | 1042 |
28
+ | main | 0.95 | 0.2077 | 620 | 707 | -87 | 1.56 | 0.000 | 1042 |
29
+ | main | 0.98 | 0.2077 | 620 | 707 | -87 | 1.56 | 0.000 | 1042 |
30
+ | main | 0.99 | 0.2077 | 620 | 707 | -87 | 1.56 | 0.000 | 1042 |
31
+ | entity | rules only | 0.1964 | 280 | 4 | 276 | 0.02 | 0.000 | 0 |
32
+ | entity | 0.0 | 0.1965 | 280 | 6 | 274 | 0.03 | 0.000 | 3 |
33
+ | entity | 0.5 | 0.1975 | 287 | 39 | 248 | 0.17 | 0.000 | 84 |
34
+ | entity | 0.6 | 0.1991 | 302 | 106 | 196 | 0.47 | 0.000 | 225 |
35
+ | entity | 0.7 | 0.2046 | 312 | 266 | 46 | 1.17 | 0.000 | 438 |
36
+ | entity | 0.8 | 0.2073 | 309 | 339 | -30 | 1.49 | 0.000 | 516 |
37
+ | entity | 0.9 | 0.2074 | 309 | 340 | -31 | 1.50 | 0.000 | 517 |
38
+ | entity | 0.95 | 0.2074 | 309 | 340 | -31 | 1.50 | 0.000 | 517 |
39
+ | entity | 0.98 | 0.2074 | 309 | 340 | -31 | 1.50 | 0.000 | 517 |
40
+ | entity | 0.99 | 0.2074 | 309 | 340 | -31 | 1.50 | 0.000 | 517 |
41
+ | typo | rules only | 0.1502 | 59 | 0 | 59 | 0.00 | 0.000 | 0 |
42
+ | typo | 0.0 | 0.1502 | 59 | 0 | 59 | 0.00 | 0.000 | 0 |
43
+ | typo | 0.5 | 0.1521 | 60 | 11 | 49 | 0.24 | 0.000 | 13 |
44
+ | typo | 0.6 | 0.1541 | 62 | 22 | 40 | 0.48 | 0.000 | 31 |
45
+ | typo | 0.7 | 0.1603 | 64 | 56 | 8 | 1.23 | 0.000 | 71 |
46
+ | typo | 0.8 | 0.1640 | 64 | 75 | -11 | 1.64 | 0.000 | 91 |
47
+ | typo | 0.9 | 0.1642 | 64 | 75 | -11 | 1.64 | 0.000 | 93 |
48
+ | typo | 0.95 | 0.1642 | 64 | 75 | -11 | 1.64 | 0.000 | 93 |
49
+ | typo | 0.98 | 0.1642 | 64 | 75 | -11 | 1.64 | 0.000 | 93 |
50
+ | typo | 0.99 | 0.1642 | 64 | 75 | -11 | 1.64 | 0.000 | 93 |
edit-qwen3.5-2b/decode_sweep/test_windows.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
edit-qwen3.5-2b/decode_sweep/validation_windows.jsonl ADDED
The diff for this file is too large to render. See raw diff