B4K2xx commited on
Commit
f4ed4c1
·
verified ·
1 Parent(s): cdb7a09
m28/benchmark_results.json ADDED
@@ -0,0 +1,540 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m28",
3
+ "checkpoint": "results/m28/final.pt",
4
+ "trained_step": 15258,
5
+ "num_fewshot": 0,
6
+ "limit": null,
7
+ "results": {
8
+ "lambada_openai": {
9
+ "name": "lambada_openai",
10
+ "alias": "lambada_openai",
11
+ "sample_len": 5153,
12
+ "perplexity,none": 566.312332169294,
13
+ "perplexity_stderr,none": 30.16584819271405,
14
+ "acc,none": 0.17543178730836406,
15
+ "acc_stderr,none": 0.005298824200667555
16
+ },
17
+ "hellaswag": {
18
+ "name": "hellaswag",
19
+ "alias": "hellaswag",
20
+ "sample_len": 10042,
21
+ "acc,none": 0.27215694084843656,
22
+ "acc_stderr,none": 0.004441606665788188,
23
+ "acc_norm,none": 0.2787293367855009,
24
+ "acc_norm_stderr,none": 0.004474577054517292
25
+ },
26
+ "piqa": {
27
+ "name": "piqa",
28
+ "alias": "piqa",
29
+ "sample_len": 1838,
30
+ "acc,none": 0.6017410228509249,
31
+ "acc_stderr,none": 0.011421759697789035,
32
+ "acc_norm,none": 0.5919477693144722,
33
+ "acc_norm_stderr,none": 0.011466872778651143
34
+ },
35
+ "sciq": {
36
+ "name": "sciq",
37
+ "alias": "sciq",
38
+ "sample_len": 1000,
39
+ "acc,none": 0.698,
40
+ "acc_stderr,none": 0.014526080235459496,
41
+ "acc_norm,none": 0.606,
42
+ "acc_norm_stderr,none": 0.015459721957493382
43
+ },
44
+ "arc_easy": {
45
+ "name": "arc_easy",
46
+ "alias": "arc_easy",
47
+ "sample_len": 2376,
48
+ "acc,none": 0.4537037037037037,
49
+ "acc_stderr,none": 0.010215708295494234,
50
+ "acc_norm,none": 0.4099326599326599,
51
+ "acc_norm_stderr,none": 0.010091953527506203
52
+ },
53
+ "winogrande": {
54
+ "name": "winogrande",
55
+ "alias": "winogrande",
56
+ "sample_len": 1267,
57
+ "acc,none": 0.5098658247829518,
58
+ "acc_stderr,none": 0.014049749833367674
59
+ },
60
+ "blimp_adjunct_island": {
61
+ "name": "blimp_adjunct_island",
62
+ "alias": "blimp_adjunct_island",
63
+ "sample_len": 1000,
64
+ "acc,none": 0.819,
65
+ "acc_stderr,none": 0.012181436179178027
66
+ },
67
+ "blimp_anaphor_gender_agreement": {
68
+ "name": "blimp_anaphor_gender_agreement",
69
+ "alias": "blimp_anaphor_gender_agreement",
70
+ "sample_len": 1000,
71
+ "acc,none": 0.841,
72
+ "acc_stderr,none": 0.011569479368271195
73
+ },
74
+ "blimp_anaphor_number_agreement": {
75
+ "name": "blimp_anaphor_number_agreement",
76
+ "alias": "blimp_anaphor_number_agreement",
77
+ "sample_len": 1000,
78
+ "acc,none": 0.964,
79
+ "acc_stderr,none": 0.00589395781616553
80
+ },
81
+ "blimp_animate_subject_passive": {
82
+ "name": "blimp_animate_subject_passive",
83
+ "alias": "blimp_animate_subject_passive",
84
+ "sample_len": 1000,
85
+ "acc,none": 0.653,
86
+ "acc_stderr,none": 0.015060472031706637
87
+ },
88
+ "blimp_animate_subject_trans": {
89
+ "name": "blimp_animate_subject_trans",
90
+ "alias": "blimp_animate_subject_trans",
91
+ "sample_len": 1000,
92
+ "acc,none": 0.915,
93
+ "acc_stderr,none": 0.008823426366942316
94
+ },
95
+ "blimp_causative": {
96
+ "name": "blimp_causative",
97
+ "alias": "blimp_causative",
98
+ "sample_len": 1000,
99
+ "acc,none": 0.668,
100
+ "acc_stderr,none": 0.014899597242811567
101
+ },
102
+ "blimp_complex_NP_island": {
103
+ "name": "blimp_complex_NP_island",
104
+ "alias": "blimp_complex_NP_island",
105
+ "sample_len": 1000,
106
+ "acc,none": 0.458,
107
+ "acc_stderr,none": 0.015763390640483554
108
+ },
109
+ "blimp_coordinate_structure_constraint_complex_left_branch": {
110
+ "name": "blimp_coordinate_structure_constraint_complex_left_branch",
111
+ "alias": "blimp_coordinate_structure_constraint_complex_left_branch",
112
+ "sample_len": 1000,
113
+ "acc,none": 0.344,
114
+ "acc_stderr,none": 0.015029633724408919
115
+ },
116
+ "blimp_coordinate_structure_constraint_object_extraction": {
117
+ "name": "blimp_coordinate_structure_constraint_object_extraction",
118
+ "alias": "blimp_coordinate_structure_constraint_object_extraction",
119
+ "sample_len": 1000,
120
+ "acc,none": 0.795,
121
+ "acc_stderr,none": 0.012772554096113201
122
+ },
123
+ "blimp_determiner_noun_agreement_1": {
124
+ "name": "blimp_determiner_noun_agreement_1",
125
+ "alias": "blimp_determiner_noun_agreement_1",
126
+ "sample_len": 1000,
127
+ "acc,none": 0.963,
128
+ "acc_stderr,none": 0.0059721576223896395
129
+ },
130
+ "blimp_determiner_noun_agreement_2": {
131
+ "name": "blimp_determiner_noun_agreement_2",
132
+ "alias": "blimp_determiner_noun_agreement_2",
133
+ "sample_len": 1000,
134
+ "acc,none": 0.941,
135
+ "acc_stderr,none": 0.007454835650406693
136
+ },
137
+ "blimp_determiner_noun_agreement_irregular_1": {
138
+ "name": "blimp_determiner_noun_agreement_irregular_1",
139
+ "alias": "blimp_determiner_noun_agreement_irregular_1",
140
+ "sample_len": 1000,
141
+ "acc,none": 0.85,
142
+ "acc_stderr,none": 0.011297239823409418
143
+ },
144
+ "blimp_determiner_noun_agreement_irregular_2": {
145
+ "name": "blimp_determiner_noun_agreement_irregular_2",
146
+ "alias": "blimp_determiner_noun_agreement_irregular_2",
147
+ "sample_len": 1000,
148
+ "acc,none": 0.919,
149
+ "acc_stderr,none": 0.008632121032139986
150
+ },
151
+ "blimp_determiner_noun_agreement_with_adj_2": {
152
+ "name": "blimp_determiner_noun_agreement_with_adj_2",
153
+ "alias": "blimp_determiner_noun_agreement_with_adj_2",
154
+ "sample_len": 1000,
155
+ "acc,none": 0.905,
156
+ "acc_stderr,none": 0.009276910103103353
157
+ },
158
+ "blimp_determiner_noun_agreement_with_adj_irregular_1": {
159
+ "name": "blimp_determiner_noun_agreement_with_adj_irregular_1",
160
+ "alias": "blimp_determiner_noun_agreement_with_adj_irregular_1",
161
+ "sample_len": 1000,
162
+ "acc,none": 0.805,
163
+ "acc_stderr,none": 0.012535235623319256
164
+ },
165
+ "blimp_determiner_noun_agreement_with_adj_irregular_2": {
166
+ "name": "blimp_determiner_noun_agreement_with_adj_irregular_2",
167
+ "alias": "blimp_determiner_noun_agreement_with_adj_irregular_2",
168
+ "sample_len": 1000,
169
+ "acc,none": 0.885,
170
+ "acc_stderr,none": 0.010093407594904551
171
+ },
172
+ "blimp_determiner_noun_agreement_with_adjective_1": {
173
+ "name": "blimp_determiner_noun_agreement_with_adjective_1",
174
+ "alias": "blimp_determiner_noun_agreement_with_adjective_1",
175
+ "sample_len": 1000,
176
+ "acc,none": 0.93,
177
+ "acc_stderr,none": 0.008072494358323525
178
+ },
179
+ "blimp_distractor_agreement_relational_noun": {
180
+ "name": "blimp_distractor_agreement_relational_noun",
181
+ "alias": "blimp_distractor_agreement_relational_noun",
182
+ "sample_len": 1000,
183
+ "acc,none": 0.809,
184
+ "acc_stderr,none": 0.012436787112179482
185
+ },
186
+ "blimp_distractor_agreement_relative_clause": {
187
+ "name": "blimp_distractor_agreement_relative_clause",
188
+ "alias": "blimp_distractor_agreement_relative_clause",
189
+ "sample_len": 1000,
190
+ "acc,none": 0.686,
191
+ "acc_stderr,none": 0.014683991951088084
192
+ },
193
+ "blimp_drop_argument": {
194
+ "name": "blimp_drop_argument",
195
+ "alias": "blimp_drop_argument",
196
+ "sample_len": 1000,
197
+ "acc,none": 0.738,
198
+ "acc_stderr,none": 0.013912208651021217
199
+ },
200
+ "blimp_ellipsis_n_bar_1": {
201
+ "name": "blimp_ellipsis_n_bar_1",
202
+ "alias": "blimp_ellipsis_n_bar_1",
203
+ "sample_len": 1000,
204
+ "acc,none": 0.813,
205
+ "acc_stderr,none": 0.012336254828074168
206
+ },
207
+ "blimp_ellipsis_n_bar_2": {
208
+ "name": "blimp_ellipsis_n_bar_2",
209
+ "alias": "blimp_ellipsis_n_bar_2",
210
+ "sample_len": 1000,
211
+ "acc,none": 0.826,
212
+ "acc_stderr,none": 0.011994493230973447
213
+ },
214
+ "blimp_existential_there_object_raising": {
215
+ "name": "blimp_existential_there_object_raising",
216
+ "alias": "blimp_existential_there_object_raising",
217
+ "sample_len": 1000,
218
+ "acc,none": 0.763,
219
+ "acc_stderr,none": 0.013454070462577935
220
+ },
221
+ "blimp_existential_there_quantifiers_1": {
222
+ "name": "blimp_existential_there_quantifiers_1",
223
+ "alias": "blimp_existential_there_quantifiers_1",
224
+ "sample_len": 1000,
225
+ "acc,none": 0.957,
226
+ "acc_stderr,none": 0.006418114379799739
227
+ },
228
+ "blimp_existential_there_quantifiers_2": {
229
+ "name": "blimp_existential_there_quantifiers_2",
230
+ "alias": "blimp_existential_there_quantifiers_2",
231
+ "sample_len": 1000,
232
+ "acc,none": 0.229,
233
+ "acc_stderr,none": 0.013294199326613684
234
+ },
235
+ "blimp_existential_there_subject_raising": {
236
+ "name": "blimp_existential_there_subject_raising",
237
+ "alias": "blimp_existential_there_subject_raising",
238
+ "sample_len": 1000,
239
+ "acc,none": 0.81,
240
+ "acc_stderr,none": 0.012411851354816254
241
+ },
242
+ "blimp_expletive_it_object_raising": {
243
+ "name": "blimp_expletive_it_object_raising",
244
+ "alias": "blimp_expletive_it_object_raising",
245
+ "sample_len": 1000,
246
+ "acc,none": 0.761,
247
+ "acc_stderr,none": 0.013493000446937704
248
+ },
249
+ "blimp_inchoative": {
250
+ "name": "blimp_inchoative",
251
+ "alias": "blimp_inchoative",
252
+ "sample_len": 1000,
253
+ "acc,none": 0.611,
254
+ "acc_stderr,none": 0.015424555647308514
255
+ },
256
+ "blimp_intransitive": {
257
+ "name": "blimp_intransitive",
258
+ "alias": "blimp_intransitive",
259
+ "sample_len": 1000,
260
+ "acc,none": 0.693,
261
+ "acc_stderr,none": 0.01459328489285274
262
+ },
263
+ "blimp_irregular_past_participle_adjectives": {
264
+ "name": "blimp_irregular_past_participle_adjectives",
265
+ "alias": "blimp_irregular_past_participle_adjectives",
266
+ "sample_len": 1000,
267
+ "acc,none": 0.908,
268
+ "acc_stderr,none": 0.009144376393151129
269
+ },
270
+ "blimp_irregular_past_participle_verbs": {
271
+ "name": "blimp_irregular_past_participle_verbs",
272
+ "alias": "blimp_irregular_past_participle_verbs",
273
+ "sample_len": 1000,
274
+ "acc,none": 0.886,
275
+ "acc_stderr,none": 0.010055103435823278
276
+ },
277
+ "blimp_irregular_plural_subject_verb_agreement_1": {
278
+ "name": "blimp_irregular_plural_subject_verb_agreement_1",
279
+ "alias": "blimp_irregular_plural_subject_verb_agreement_1",
280
+ "sample_len": 1000,
281
+ "acc,none": 0.845,
282
+ "acc_stderr,none": 0.011450157470799522
283
+ },
284
+ "blimp_irregular_plural_subject_verb_agreement_2": {
285
+ "name": "blimp_irregular_plural_subject_verb_agreement_2",
286
+ "alias": "blimp_irregular_plural_subject_verb_agreement_2",
287
+ "sample_len": 1000,
288
+ "acc,none": 0.881,
289
+ "acc_stderr,none": 0.01024421514533661
290
+ },
291
+ "blimp_left_branch_island_echo_question": {
292
+ "name": "blimp_left_branch_island_echo_question",
293
+ "alias": "blimp_left_branch_island_echo_question",
294
+ "sample_len": 1000,
295
+ "acc,none": 0.551,
296
+ "acc_stderr,none": 0.015736792768752044
297
+ },
298
+ "blimp_left_branch_island_simple_question": {
299
+ "name": "blimp_left_branch_island_simple_question",
300
+ "alias": "blimp_left_branch_island_simple_question",
301
+ "sample_len": 1000,
302
+ "acc,none": 0.438,
303
+ "acc_stderr,none": 0.01569721001969466
304
+ },
305
+ "blimp_matrix_question_npi_licensor_present": {
306
+ "name": "blimp_matrix_question_npi_licensor_present",
307
+ "alias": "blimp_matrix_question_npi_licensor_present",
308
+ "sample_len": 1000,
309
+ "acc,none": 0.571,
310
+ "acc_stderr,none": 0.015658997547870177
311
+ },
312
+ "blimp_npi_present_1": {
313
+ "name": "blimp_npi_present_1",
314
+ "alias": "blimp_npi_present_1",
315
+ "sample_len": 1000,
316
+ "acc,none": 0.491,
317
+ "acc_stderr,none": 0.0158167369950053
318
+ },
319
+ "blimp_npi_present_2": {
320
+ "name": "blimp_npi_present_2",
321
+ "alias": "blimp_npi_present_2",
322
+ "sample_len": 1000,
323
+ "acc,none": 0.644,
324
+ "acc_stderr,none": 0.015149042659306678
325
+ },
326
+ "blimp_only_npi_licensor_present": {
327
+ "name": "blimp_only_npi_licensor_present",
328
+ "alias": "blimp_only_npi_licensor_present",
329
+ "sample_len": 1000,
330
+ "acc,none": 0.569,
331
+ "acc_stderr,none": 0.015667944488173442
332
+ },
333
+ "blimp_only_npi_scope": {
334
+ "name": "blimp_only_npi_scope",
335
+ "alias": "blimp_only_npi_scope",
336
+ "sample_len": 1000,
337
+ "acc,none": 0.512,
338
+ "acc_stderr,none": 0.015814743314581686
339
+ },
340
+ "blimp_passive_1": {
341
+ "name": "blimp_passive_1",
342
+ "alias": "blimp_passive_1",
343
+ "sample_len": 1000,
344
+ "acc,none": 0.851,
345
+ "acc_stderr,none": 0.011266140684632246
346
+ },
347
+ "blimp_passive_2": {
348
+ "name": "blimp_passive_2",
349
+ "alias": "blimp_passive_2",
350
+ "sample_len": 1000,
351
+ "acc,none": 0.851,
352
+ "acc_stderr,none": 0.011266140684632246
353
+ },
354
+ "blimp_principle_A_c_command": {
355
+ "name": "blimp_principle_A_c_command",
356
+ "alias": "blimp_principle_A_c_command",
357
+ "sample_len": 1000,
358
+ "acc,none": 0.699,
359
+ "acc_stderr,none": 0.01451239503354321
360
+ },
361
+ "blimp_principle_A_case_1": {
362
+ "name": "blimp_principle_A_case_1",
363
+ "alias": "blimp_principle_A_case_1",
364
+ "sample_len": 1000,
365
+ "acc,none": 1.0,
366
+ "acc_stderr,none": 0.0
367
+ },
368
+ "blimp_principle_A_case_2": {
369
+ "name": "blimp_principle_A_case_2",
370
+ "alias": "blimp_principle_A_case_2",
371
+ "sample_len": 1000,
372
+ "acc,none": 0.896,
373
+ "acc_stderr,none": 0.009658016218524242
374
+ },
375
+ "blimp_principle_A_domain_1": {
376
+ "name": "blimp_principle_A_domain_1",
377
+ "alias": "blimp_principle_A_domain_1",
378
+ "sample_len": 1000,
379
+ "acc,none": 0.848,
380
+ "acc_stderr,none": 0.01135891830347525
381
+ },
382
+ "blimp_principle_A_domain_2": {
383
+ "name": "blimp_principle_A_domain_2",
384
+ "alias": "blimp_principle_A_domain_2",
385
+ "sample_len": 1000,
386
+ "acc,none": 0.718,
387
+ "acc_stderr,none": 0.01423652621529138
388
+ },
389
+ "blimp_principle_A_domain_3": {
390
+ "name": "blimp_principle_A_domain_3",
391
+ "alias": "blimp_principle_A_domain_3",
392
+ "sample_len": 1000,
393
+ "acc,none": 0.604,
394
+ "acc_stderr,none": 0.015473313265859495
395
+ },
396
+ "blimp_principle_A_reconstruction": {
397
+ "name": "blimp_principle_A_reconstruction",
398
+ "alias": "blimp_principle_A_reconstruction",
399
+ "sample_len": 1000,
400
+ "acc,none": 0.321,
401
+ "acc_stderr,none": 0.01477082181793475
402
+ },
403
+ "blimp_regular_plural_subject_verb_agreement_1": {
404
+ "name": "blimp_regular_plural_subject_verb_agreement_1",
405
+ "alias": "blimp_regular_plural_subject_verb_agreement_1",
406
+ "sample_len": 1000,
407
+ "acc,none": 0.86,
408
+ "acc_stderr,none": 0.010978183844357888
409
+ },
410
+ "blimp_regular_plural_subject_verb_agreement_2": {
411
+ "name": "blimp_regular_plural_subject_verb_agreement_2",
412
+ "alias": "blimp_regular_plural_subject_verb_agreement_2",
413
+ "sample_len": 1000,
414
+ "acc,none": 0.894,
415
+ "acc_stderr,none": 0.00973955126578524
416
+ },
417
+ "blimp_sentential_negation_npi_licensor_present": {
418
+ "name": "blimp_sentential_negation_npi_licensor_present",
419
+ "alias": "blimp_sentential_negation_npi_licensor_present",
420
+ "sample_len": 1000,
421
+ "acc,none": 0.998,
422
+ "acc_stderr,none": 0.0014135055705578234
423
+ },
424
+ "blimp_sentential_negation_npi_scope": {
425
+ "name": "blimp_sentential_negation_npi_scope",
426
+ "alias": "blimp_sentential_negation_npi_scope",
427
+ "sample_len": 1000,
428
+ "acc,none": 0.6,
429
+ "acc_stderr,none": 0.015499685165842571
430
+ },
431
+ "blimp_sentential_subject_island": {
432
+ "name": "blimp_sentential_subject_island",
433
+ "alias": "blimp_sentential_subject_island",
434
+ "sample_len": 1000,
435
+ "acc,none": 0.499,
436
+ "acc_stderr,none": 0.015819268290576817
437
+ },
438
+ "blimp_superlative_quantifiers_1": {
439
+ "name": "blimp_superlative_quantifiers_1",
440
+ "alias": "blimp_superlative_quantifiers_1",
441
+ "sample_len": 1000,
442
+ "acc,none": 0.535,
443
+ "acc_stderr,none": 0.015780495050030086
444
+ },
445
+ "blimp_superlative_quantifiers_2": {
446
+ "name": "blimp_superlative_quantifiers_2",
447
+ "alias": "blimp_superlative_quantifiers_2",
448
+ "sample_len": 1000,
449
+ "acc,none": 0.471,
450
+ "acc_stderr,none": 0.015792669451628764
451
+ },
452
+ "blimp_tough_vs_raising_1": {
453
+ "name": "blimp_tough_vs_raising_1",
454
+ "alias": "blimp_tough_vs_raising_1",
455
+ "sample_len": 1000,
456
+ "acc,none": 0.469,
457
+ "acc_stderr,none": 0.015788865959538965
458
+ },
459
+ "blimp_tough_vs_raising_2": {
460
+ "name": "blimp_tough_vs_raising_2",
461
+ "alias": "blimp_tough_vs_raising_2",
462
+ "sample_len": 1000,
463
+ "acc,none": 0.854,
464
+ "acc_stderr,none": 0.011171786285496551
465
+ },
466
+ "blimp_transitive": {
467
+ "name": "blimp_transitive",
468
+ "alias": "blimp_transitive",
469
+ "sample_len": 1000,
470
+ "acc,none": 0.783,
471
+ "acc_stderr,none": 0.013041513757270706
472
+ },
473
+ "blimp_wh_island": {
474
+ "name": "blimp_wh_island",
475
+ "alias": "blimp_wh_island",
476
+ "sample_len": 1000,
477
+ "acc,none": 0.743,
478
+ "acc_stderr,none": 0.013825416526895026
479
+ },
480
+ "blimp_wh_questions_object_gap": {
481
+ "name": "blimp_wh_questions_object_gap",
482
+ "alias": "blimp_wh_questions_object_gap",
483
+ "sample_len": 1000,
484
+ "acc,none": 0.708,
485
+ "acc_stderr,none": 0.014385511563477433
486
+ },
487
+ "blimp_wh_questions_subject_gap": {
488
+ "name": "blimp_wh_questions_subject_gap",
489
+ "alias": "blimp_wh_questions_subject_gap",
490
+ "sample_len": 1000,
491
+ "acc,none": 0.911,
492
+ "acc_stderr,none": 0.009008893392651516
493
+ },
494
+ "blimp_wh_questions_subject_gap_long_distance": {
495
+ "name": "blimp_wh_questions_subject_gap_long_distance",
496
+ "alias": "blimp_wh_questions_subject_gap_long_distance",
497
+ "sample_len": 1000,
498
+ "acc,none": 0.807,
499
+ "acc_stderr,none": 0.012486268734370044
500
+ },
501
+ "blimp_wh_vs_that_no_gap": {
502
+ "name": "blimp_wh_vs_that_no_gap",
503
+ "alias": "blimp_wh_vs_that_no_gap",
504
+ "sample_len": 1000,
505
+ "acc,none": 0.966,
506
+ "acc_stderr,none": 0.0057338361396953854
507
+ },
508
+ "blimp_wh_vs_that_no_gap_long_distance": {
509
+ "name": "blimp_wh_vs_that_no_gap_long_distance",
510
+ "alias": "blimp_wh_vs_that_no_gap_long_distance",
511
+ "sample_len": 1000,
512
+ "acc,none": 0.958,
513
+ "acc_stderr,none": 0.006346359293033784
514
+ },
515
+ "blimp_wh_vs_that_with_gap": {
516
+ "name": "blimp_wh_vs_that_with_gap",
517
+ "alias": "blimp_wh_vs_that_with_gap",
518
+ "sample_len": 1000,
519
+ "acc,none": 0.38,
520
+ "acc_stderr,none": 0.015356947477797658
521
+ },
522
+ "blimp_wh_vs_that_with_gap_long_distance": {
523
+ "name": "blimp_wh_vs_that_with_gap_long_distance",
524
+ "alias": "blimp_wh_vs_that_with_gap_long_distance",
525
+ "sample_len": 1000,
526
+ "acc,none": 0.2,
527
+ "acc_stderr,none": 0.012655439943366655
528
+ },
529
+ "blimp": {
530
+ "alias": "blimp",
531
+ "name": "blimp",
532
+ "sample_len": 67000,
533
+ "acc,none": 0.7324029850746269,
534
+ "acc_stderr,none": 0.0015343189725241936,
535
+ "sample_count": {
536
+ "acc,none": 67000
537
+ }
538
+ }
539
+ }
540
+ }
m28/config.json ADDED
@@ -0,0 +1,49 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m28",
3
+ "run_name": "m28-gqa-rope-dense-muonv2-lr01",
4
+ "data_dir": "data/fineweb-edu",
5
+ "vocab_size": 50257,
6
+ "block_size": 1024,
7
+ "d_model": 512,
8
+ "num_layers": 12,
9
+ "num_heads": 8,
10
+ "ffn_mult": 4,
11
+ "dropout": 0.0,
12
+ "bias": false,
13
+ "norm_type": "rmsnorm",
14
+ "tie_embeddings": true,
15
+ "attention_type": "gqa",
16
+ "gqa_num_kv_heads": 2,
17
+ "mla_latent_dim": 64,
18
+ "pos_type": "rope",
19
+ "rope_base": 10000,
20
+ "ffn_type": "dense",
21
+ "moe_num_experts": 8,
22
+ "moe_top_k": 2,
23
+ "optimizer": "muon_v2",
24
+ "lr": 0.0006,
25
+ "min_lr": 6e-05,
26
+ "weight_decay": 0.1,
27
+ "beta1": 0.9,
28
+ "beta2": 0.95,
29
+ "grad_clip": 1.0,
30
+ "muon_lr": 0.01,
31
+ "muon_momentum": 0.95,
32
+ "training_tokens": 1000000000,
33
+ "warmup_tokens": 6000000,
34
+ "lr_schedule": "cosine",
35
+ "batch_size": 4,
36
+ "grad_accum_steps": 16,
37
+ "device": "cuda",
38
+ "dtype": "bfloat16",
39
+ "compile": true,
40
+ "seed": 1337,
41
+ "eval_interval": 250,
42
+ "eval_iters": 100,
43
+ "log_interval": 10,
44
+ "checkpoint_interval": 1000,
45
+ "results_dir": "results",
46
+ "mlflow_experiment": "slm-ablation",
47
+ "mlflow_tracking_uri": "sqlite:///mlflow.db",
48
+ "mlflow_artifact_location": "./mlartifacts"
49
+ }
m28/eval_results.json ADDED
@@ -0,0 +1,212 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m28",
3
+ "checkpoint": "/home/akshat/codemindv2/results/m28/final.pt",
4
+ "trained_step": 15258,
5
+ "total_params": 58774528,
6
+ "active_params": 58774528,
7
+ "flops_per_forward": 117549056.0,
8
+ "val_loss": 3.4364620834589004,
9
+ "val_perplexity": 31.076816264456788,
10
+ "bits_per_token": 4.9577668060091105,
11
+ "ppl_tokens_evaluated": 655360,
12
+ "prefill_tok_per_sec": 191174.46259175235,
13
+ "inference_tok_per_sec": 215.11173153691215,
14
+ "inference_tok_per_sec_kvcache": 205.6556695510074,
15
+ "kvcache_speedup": 0.9560411609429952,
16
+ "gen_tokens": 128,
17
+ "peak_infer_vram_gb": 0.366,
18
+ "peak_train_vram_gb": 3.993,
19
+ "closeness_answer_ppl": 38.273,
20
+ "closeness_hit_rate": 0.35,
21
+ "closeness_token_f1": 0.02,
22
+ "closeness_n": 20,
23
+ "closeness_items": [
24
+ {
25
+ "prompt": "The capital of France is",
26
+ "answer": " Paris",
27
+ "generated": " the city of Paris. The",
28
+ "answer_nll": 2.9576,
29
+ "answer_ppl": 19.252,
30
+ "hit": true,
31
+ "token_f1": 0.0
32
+ },
33
+ {
34
+ "prompt": "The capital of Japan is",
35
+ "answer": " Tokyo",
36
+ "generated": " located in the heart of the",
37
+ "answer_nll": 4.5123,
38
+ "answer_ppl": 91.132,
39
+ "hit": false,
40
+ "token_f1": 0.0
41
+ },
42
+ {
43
+ "prompt": "The largest planet in the solar system is",
44
+ "answer": " Jupiter",
45
+ "generated": " the sun.\nThe sun",
46
+ "answer_nll": 3.5618,
47
+ "answer_ppl": 35.228,
48
+ "hit": false,
49
+ "token_f1": 0.0
50
+ },
51
+ {
52
+ "prompt": "Water is made of hydrogen and",
53
+ "answer": " oxygen",
54
+ "generated": " helium.\nThe hydrogen is",
55
+ "answer_nll": 1.0301,
56
+ "answer_ppl": 2.801,
57
+ "hit": false,
58
+ "token_f1": 0.0
59
+ },
60
+ {
61
+ "prompt": "The opposite of hot is",
62
+ "answer": " cold",
63
+ "generated": " the heat of the sun.",
64
+ "answer_nll": 5.4039,
65
+ "answer_ppl": 222.282,
66
+ "hit": false,
67
+ "token_f1": 0.0
68
+ },
69
+ {
70
+ "prompt": "The opposite of up is",
71
+ "answer": " down",
72
+ "generated": " the fact that the two of",
73
+ "answer_nll": 8.4353,
74
+ "answer_ppl": 4606.82,
75
+ "hit": false,
76
+ "token_f1": 0.0
77
+ },
78
+ {
79
+ "prompt": "Two plus two equals",
80
+ "answer": " four",
81
+ "generated": " 1.\n- The first",
82
+ "answer_nll": 4.2668,
83
+ "answer_ppl": 71.295,
84
+ "hit": false,
85
+ "token_f1": 0.0
86
+ },
87
+ {
88
+ "prompt": "The sun rises in the",
89
+ "answer": " east",
90
+ "generated": " sky, and the sun is",
91
+ "answer_nll": 3.1628,
92
+ "answer_ppl": 23.637,
93
+ "hit": false,
94
+ "token_f1": 0.0
95
+ },
96
+ {
97
+ "prompt": "The first president of the United States was George",
98
+ "answer": " Washington",
99
+ "generated": " Washington. He was a man",
100
+ "answer_nll": 0.8406,
101
+ "answer_ppl": 2.318,
102
+ "hit": true,
103
+ "token_f1": 0.0
104
+ },
105
+ {
106
+ "prompt": "The speed of light is approximately 300,000 kilometers per",
107
+ "answer": " second",
108
+ "generated": " second. The speed of light",
109
+ "answer_nll": 0.229,
110
+ "answer_ppl": 1.257,
111
+ "hit": true,
112
+ "token_f1": 0.0
113
+ },
114
+ {
115
+ "prompt": "A triangle has three",
116
+ "answer": " sides",
117
+ "generated": " sides, and the other two",
118
+ "answer_nll": 1.4465,
119
+ "answer_ppl": 4.248,
120
+ "hit": true,
121
+ "token_f1": 0.0
122
+ },
123
+ {
124
+ "prompt": "The chemical symbol for gold is",
125
+ "answer": " Au",
126
+ "generated": " the same as the symbol for",
127
+ "answer_nll": 8.2115,
128
+ "answer_ppl": 3683.169,
129
+ "hit": false,
130
+ "token_f1": 0.0
131
+ },
132
+ {
133
+ "prompt": "The freezing point of water in Celsius is",
134
+ "answer": " zero",
135
+ "generated": " the temperature at which the water",
136
+ "answer_nll": 5.6623,
137
+ "answer_ppl": 287.822,
138
+ "hit": false,
139
+ "token_f1": 0.0
140
+ },
141
+ {
142
+ "prompt": "The author of Romeo and Juliet is William",
143
+ "answer": " Shakespeare",
144
+ "generated": " Shakespeare.\nThe author of",
145
+ "answer_nll": 0.5891,
146
+ "answer_ppl": 1.802,
147
+ "hit": true,
148
+ "token_f1": 0.0
149
+ },
150
+ {
151
+ "prompt": "The Earth revolves around the",
152
+ "answer": " Sun",
153
+ "generated": " sun.\nThe sun is",
154
+ "answer_nll": 1.6502,
155
+ "answer_ppl": 5.208,
156
+ "hit": true,
157
+ "token_f1": 0.4
158
+ },
159
+ {
160
+ "prompt": "The plural of mouse is",
161
+ "answer": " mice",
162
+ "generated": " a mouse, which is a",
163
+ "answer_nll": 5.9685,
164
+ "answer_ppl": 390.937,
165
+ "hit": false,
166
+ "token_f1": 0.0
167
+ },
168
+ {
169
+ "prompt": "Photosynthesis occurs in the",
170
+ "answer": " leaves",
171
+ "generated": " body.\nThe body is",
172
+ "answer_nll": 4.1835,
173
+ "answer_ppl": 65.594,
174
+ "hit": false,
175
+ "token_f1": 0.0
176
+ },
177
+ {
178
+ "prompt": "The human heart pumps",
179
+ "answer": " blood",
180
+ "generated": " are not the only ones that",
181
+ "answer_nll": 4.1331,
182
+ "answer_ppl": 62.371,
183
+ "hit": false,
184
+ "token_f1": 0.0
185
+ },
186
+ {
187
+ "prompt": "The currency used in the United States is the",
188
+ "answer": " dollar",
189
+ "generated": " U.S. dollar.",
190
+ "answer_nll": 3.7576,
191
+ "answer_ppl": 42.845,
192
+ "hit": true,
193
+ "token_f1": 0.0
194
+ },
195
+ {
196
+ "prompt": "The tallest animal in the world is the",
197
+ "answer": " giraffe",
198
+ "generated": " bull-like bull-like",
199
+ "answer_nll": 2.8921,
200
+ "answer_ppl": 18.031,
201
+ "hit": false,
202
+ "token_f1": 0.0
203
+ }
204
+ ],
205
+ "generation_samples": {
206
+ "The most important thing about sleep is": "The most important thing about sleep is that it\u2019s a good time to prepare yourself and maintain a calm and relaxed lifestyle. If we all enjoy the sun, we don\u2019t want to be alone. Our daily routine is a reflection of ourselves, and also our physical health.\nMost of us have probably heard of \ufffd",
207
+ "Scientists have discovered that": "Scientists have discovered that, in many ways, the human body is actually a computer system. \"We've discovered that when you put a computer in a box, you're going to turn the human body into a computer,\" says Dr. John O'Brien, one of the authors of this piece. \"You can't",
208
+ "In the history of mathematics,": "In the history of mathematics, the term has been used to describe the practice of using the metric system to measure the relationship between the two measures of the metric system. The metric system, on the other hand, refers to the measurement data of the meter which can be combined into a metric system. The metric system was developed in the",
209
+ "Python code for prime number": "Python code for prime number of objects.\nEach step in the program is numbered like a number, and a symbol. The function will use the number of the first digits to represent the second digit of the sequence of numbers. The result is the following:\nThis is the function of\n- the first digit of the sequence",
210
+ "The best way to learn a new language is": "The best way to learn a new language is to learn about it. So if you have an understanding of a language, you will benefit from reading a list of words to learn a new language.\n- Linguistic studies help students learn what the language in which they speak is. Linguistic methods are the study of language with a particular"
211
+ }
212
+ }
m28/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3bcc3570b24485f5844f5ac7cf129270e9c14e402fdb6ba4dde1e79f6b43ff9c
3
+ size 235107624