B4K2xx commited on
Commit
39817da
·
verified ·
1 Parent(s): af8b41c

Upload folder using huggingface_hub

Browse files
m15/benchmark_results.json ADDED
@@ -0,0 +1,540 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m15",
3
+ "checkpoint": "results/m15/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": 413.4500055271363,
13
+ "perplexity_stderr,none": 23.383529548584082,
14
+ "acc,none": 0.19483795847079372,
15
+ "acc_stderr,none": 0.005518111913121999
16
+ },
17
+ "hellaswag": {
18
+ "name": "hellaswag",
19
+ "alias": "hellaswag",
20
+ "sample_len": 10042,
21
+ "acc,none": 0.2808205536745668,
22
+ "acc_stderr,none": 0.004484815647064684,
23
+ "acc_norm,none": 0.2944632543318064,
24
+ "acc_norm_stderr,none": 0.004548695749620993
25
+ },
26
+ "piqa": {
27
+ "name": "piqa",
28
+ "alias": "piqa",
29
+ "sample_len": 1838,
30
+ "acc,none": 0.6088139281828074,
31
+ "acc_stderr,none": 0.011386215606728629,
32
+ "acc_norm,none": 0.6077257889009793,
33
+ "acc_norm_stderr,none": 0.011391846744072072
34
+ },
35
+ "sciq": {
36
+ "name": "sciq",
37
+ "alias": "sciq",
38
+ "sample_len": 1000,
39
+ "acc,none": 0.679,
40
+ "acc_stderr,none": 0.01477082181793475,
41
+ "acc_norm,none": 0.583,
42
+ "acc_norm_stderr,none": 0.015599819048769583
43
+ },
44
+ "arc_easy": {
45
+ "name": "arc_easy",
46
+ "alias": "arc_easy",
47
+ "sample_len": 2376,
48
+ "acc,none": 0.4591750841750842,
49
+ "acc_stderr,none": 0.010225526906982679,
50
+ "acc_norm,none": 0.4078282828282828,
51
+ "acc_norm_stderr,none": 0.01008395024004126
52
+ },
53
+ "winogrande": {
54
+ "name": "winogrande",
55
+ "alias": "winogrande",
56
+ "sample_len": 1267,
57
+ "acc,none": 0.4988161010260458,
58
+ "acc_stderr,none": 0.014052446290528871
59
+ },
60
+ "blimp_adjunct_island": {
61
+ "name": "blimp_adjunct_island",
62
+ "alias": "blimp_adjunct_island",
63
+ "sample_len": 1000,
64
+ "acc,none": 0.836,
65
+ "acc_stderr,none": 0.011715000693181425
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": 0.862,
72
+ "acc_stderr,none": 0.010912152632504508
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": 0.966,
79
+ "acc_stderr,none": 0.0057338361396953854
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": 0.706,
86
+ "acc_stderr,none": 0.014414290540008265
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": 0.898,
93
+ "acc_stderr,none": 0.009575368801653944
94
+ },
95
+ "blimp_causative": {
96
+ "name": "blimp_causative",
97
+ "alias": "blimp_causative",
98
+ "sample_len": 1000,
99
+ "acc,none": 0.702,
100
+ "acc_stderr,none": 0.014470846741134585
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": 0.528,
107
+ "acc_stderr,none": 0.015794475789511517
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": 0.43,
114
+ "acc_stderr,none": 0.01566350361015521
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": 0.783,
121
+ "acc_stderr,none": 0.013041513757270706
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": 0.973,
128
+ "acc_stderr,none": 0.005128089049275305
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": 0.942,
135
+ "acc_stderr,none": 0.0073953154557929906
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": 0.878,
142
+ "acc_stderr,none": 0.010354864712936776
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": 0.923,
149
+ "acc_stderr,none": 0.00843458014024066
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": 0.921,
156
+ "acc_stderr,none": 0.00853415677333337
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": 0.862,
163
+ "acc_stderr,none": 0.010912152632504508
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": 0.892,
170
+ "acc_stderr,none": 0.009820001651345708
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": 0.829,
184
+ "acc_stderr,none": 0.011912216456264588
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": 0.737,
191
+ "acc_stderr,none": 0.013929286594259776
192
+ },
193
+ "blimp_drop_argument": {
194
+ "name": "blimp_drop_argument",
195
+ "alias": "blimp_drop_argument",
196
+ "sample_len": 1000,
197
+ "acc,none": 0.744,
198
+ "acc_stderr,none": 0.013807775152234145
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": 0.816,
205
+ "acc_stderr,none": 0.01225945734093864
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": 0.839,
212
+ "acc_stderr,none": 0.011628164696727263
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": 0.726,
219
+ "acc_stderr,none": 0.01411109928825969
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": 0.969,
226
+ "acc_stderr,none": 0.005483527064679181
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": 0.187,
233
+ "acc_stderr,none": 0.012336254828074166
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": 0.872,
240
+ "acc_stderr,none": 0.010570133761108595
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": 0.685,
247
+ "acc_stderr,none": 0.014696631960792617
248
+ },
249
+ "blimp_inchoative": {
250
+ "name": "blimp_inchoative",
251
+ "alias": "blimp_inchoative",
252
+ "sample_len": 1000,
253
+ "acc,none": 0.665,
254
+ "acc_stderr,none": 0.014933117490932629
255
+ },
256
+ "blimp_intransitive": {
257
+ "name": "blimp_intransitive",
258
+ "alias": "blimp_intransitive",
259
+ "sample_len": 1000,
260
+ "acc,none": 0.748,
261
+ "acc_stderr,none": 0.013736254390651213
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": 0.723,
268
+ "acc_stderr,none": 0.014158794845306273
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": 0.839,
275
+ "acc_stderr,none": 0.011628164696727263
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": 0.905,
282
+ "acc_stderr,none": 0.009276910103103353
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": 0.82,
289
+ "acc_stderr,none": 0.012155153135512022
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": 0.618,
296
+ "acc_stderr,none": 0.015372453034968502
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": 0.595,
303
+ "acc_stderr,none": 0.01553113699045296
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": 0.532,
310
+ "acc_stderr,none": 0.015786868759359002
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": 0.665,
317
+ "acc_stderr,none": 0.014933117490932629
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": 0.723,
324
+ "acc_stderr,none": 0.014158794845306273
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": 0.993,
331
+ "acc_stderr,none": 0.002637794146243781
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": 0.712,
338
+ "acc_stderr,none": 0.01432694179723158
339
+ },
340
+ "blimp_passive_1": {
341
+ "name": "blimp_passive_1",
342
+ "alias": "blimp_passive_1",
343
+ "sample_len": 1000,
344
+ "acc,none": 0.862,
345
+ "acc_stderr,none": 0.010912152632504508
346
+ },
347
+ "blimp_passive_2": {
348
+ "name": "blimp_passive_2",
349
+ "alias": "blimp_passive_2",
350
+ "sample_len": 1000,
351
+ "acc,none": 0.842,
352
+ "acc_stderr,none": 0.011539894677559635
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": 0.706,
359
+ "acc_stderr,none": 0.014414290540008265
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": 0.998,
366
+ "acc_stderr,none": 0.0014135055705578234
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": 0.908,
373
+ "acc_stderr,none": 0.009144376393151129
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": 0.905,
380
+ "acc_stderr,none": 0.009276910103103353
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": 0.704,
387
+ "acc_stderr,none": 0.01444273494157504
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": 0.647,
394
+ "acc_stderr,none": 0.015120172605483696
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": 0.425,
401
+ "acc_stderr,none": 0.01564032031704017
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": 0.864,
408
+ "acc_stderr,none": 0.01084535023047304
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": 0.869,
415
+ "acc_stderr,none": 0.0106748748448379
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": 0.985,
422
+ "acc_stderr,none": 0.003845749574502972
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": 0.632,
429
+ "acc_stderr,none": 0.015258073561521743
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": 0.528,
436
+ "acc_stderr,none": 0.015794475789511517
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": 0.832,
443
+ "acc_stderr,none": 0.011828605831454352
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": 0.663,
450
+ "acc_stderr,none": 0.01495508791865346
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": 0.66,
457
+ "acc_stderr,none": 0.014987482264363972
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": 0.848,
464
+ "acc_stderr,none": 0.01135891830347525
465
+ },
466
+ "blimp_transitive": {
467
+ "name": "blimp_transitive",
468
+ "alias": "blimp_transitive",
469
+ "sample_len": 1000,
470
+ "acc,none": 0.797,
471
+ "acc_stderr,none": 0.01272607374459825
472
+ },
473
+ "blimp_wh_island": {
474
+ "name": "blimp_wh_island",
475
+ "alias": "blimp_wh_island",
476
+ "sample_len": 1000,
477
+ "acc,none": 0.685,
478
+ "acc_stderr,none": 0.014696631960792617
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": 0.795,
485
+ "acc_stderr,none": 0.012772554096113201
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": 0.931,
492
+ "acc_stderr,none": 0.008018934050315124
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": 0.828,
499
+ "acc_stderr,none": 0.011939788882495308
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": 0.968,
506
+ "acc_stderr,none": 0.005568393575081328
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": 0.969,
513
+ "acc_stderr,none": 0.005483527064679181
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": 0.486,
520
+ "acc_stderr,none": 0.01581309754773093
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": 0.258,
527
+ "acc_stderr,none": 0.013842963108656601
528
+ },
529
+ "blimp": {
530
+ "alias": "blimp",
531
+ "name": "blimp",
532
+ "sample_len": 67000,
533
+ "acc,none": 0.7696865671641792,
534
+ "acc_stderr,none": 0.0014888534910081543,
535
+ "sample_count": {
536
+ "acc,none": 67000
537
+ }
538
+ }
539
+ }
540
+ }
m15/config.json ADDED
@@ -0,0 +1,51 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m15",
3
+ "run_name": "m15-mha-rope-shared-moe-muon",
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": "mha",
16
+ "gqa_num_kv_heads": 2,
17
+ "mla_latent_dim": 64,
18
+ "pos_type": "rope",
19
+ "rope_base": 10000,
20
+ "ffn_type": "shared_moe",
21
+ "moe_num_experts": 8,
22
+ "moe_top_k": 2,
23
+ "optimizer": "muon",
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": 0.02,
31
+ "muon_momentum": 0.95,
32
+ "training_tokens": 1000000000,
33
+ "warmup_tokens": 6000000,
34
+ "lr_schedule": "cosine",
35
+ "batch_size": 2,
36
+ "grad_accum_steps": 32,
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
+ "moe_num_shared_experts": 1,
50
+ "moe_aux_loss_coef": 0.01
51
+ }
m15/eval_results.json ADDED
@@ -0,0 +1,223 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m15",
3
+ "checkpoint": "results/m15/final.pt",
4
+ "trained_step": 15258,
5
+ "total_params": 264868864,
6
+ "active_params": 113873920,
7
+ "flops_per_forward": 227747840.0,
8
+ "val_loss": 3.4677844196558,
9
+ "val_perplexity": 32.06561972800647,
10
+ "bits_per_token": 5.002955385109434,
11
+ "ppl_tokens_evaluated": 655360,
12
+ "prefill_tok_per_sec": 61298.66085312115,
13
+ "inference_tok_per_sec": 41.6292336332186,
14
+ "inference_tok_per_sec_kvcache": 67.26176681140403,
15
+ "kvcache_speedup": 1.6157339672410307,
16
+ "gen_tokens": 128,
17
+ "peak_infer_vram_gb": 1.599,
18
+ "peak_train_vram_gb": 3.981,
19
+ "closeness_answer_ppl": 12.372,
20
+ "closeness_hit_rate": 0.55,
21
+ "closeness_token_f1": 0.029,
22
+ "closeness_n": 20,
23
+ "closeness_items": [
24
+ {
25
+ "prompt": "The capital of France is",
26
+ "answer": " Paris",
27
+ "generated": " Paris.\nThe city is",
28
+ "answer_nll": 2.0906,
29
+ "answer_ppl": 8.09,
30
+ "hit": true,
31
+ "token_f1": 0.0
32
+ },
33
+ {
34
+ "prompt": "The capital of Japan is",
35
+ "answer": " Tokyo",
36
+ "generated": " the city of Tokyo.\n",
37
+ "answer_nll": 3.1428,
38
+ "answer_ppl": 23.169,
39
+ "hit": true,
40
+ "token_f1": 0.0
41
+ },
42
+ {
43
+ "prompt": "The largest planet in the solar system is",
44
+ "answer": " Jupiter",
45
+ "generated": " the Sun, which is the",
46
+ "answer_nll": 1.6507,
47
+ "answer_ppl": 5.211,
48
+ "hit": false,
49
+ "token_f1": 0.0
50
+ },
51
+ {
52
+ "prompt": "Water is made of hydrogen and",
53
+ "answer": " oxygen",
54
+ "generated": " oxygen, which are both carbon",
55
+ "answer_nll": 0.2688,
56
+ "answer_ppl": 1.308,
57
+ "hit": true,
58
+ "token_f1": 0.0
59
+ },
60
+ {
61
+ "prompt": "The opposite of hot is",
62
+ "answer": " cold",
63
+ "generated": " the fact that the temperature of",
64
+ "answer_nll": 3.2884,
65
+ "answer_ppl": 26.799,
66
+ "hit": false,
67
+ "token_f1": 0.0
68
+ },
69
+ {
70
+ "prompt": "The opposite of up is",
71
+ "answer": " down",
72
+ "generated": " the word \u201cfoss",
73
+ "answer_nll": 6.3577,
74
+ "answer_ppl": 576.902,
75
+ "hit": false,
76
+ "token_f1": 0.0
77
+ },
78
+ {
79
+ "prompt": "Two plus two equals",
80
+ "answer": " four",
81
+ "generated": " the sum of the two.",
82
+ "answer_nll": 3.4839,
83
+ "answer_ppl": 32.588,
84
+ "hit": false,
85
+ "token_f1": 0.0
86
+ },
87
+ {
88
+ "prompt": "The sun rises in the",
89
+ "answer": " east",
90
+ "generated": " east and sets in the west",
91
+ "answer_nll": 0.7555,
92
+ "answer_ppl": 2.129,
93
+ "hit": true,
94
+ "token_f1": 0.286
95
+ },
96
+ {
97
+ "prompt": "The first president of the United States was George",
98
+ "answer": " Washington",
99
+ "generated": " Washington, who was a member",
100
+ "answer_nll": 0.4636,
101
+ "answer_ppl": 1.59,
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": 0.1067,
110
+ "answer_ppl": 1.113,
111
+ "hit": true,
112
+ "token_f1": 0.0
113
+ },
114
+ {
115
+ "prompt": "A triangle has three",
116
+ "answer": " sides",
117
+ "generated": " sides. The first is the",
118
+ "answer_nll": 1.2746,
119
+ "answer_ppl": 3.577,
120
+ "hit": true,
121
+ "token_f1": 0.0
122
+ },
123
+ {
124
+ "prompt": "The chemical symbol for gold is",
125
+ "answer": " Au",
126
+ "generated": " the gold leaf.\nThe",
127
+ "answer_nll": 6.585,
128
+ "answer_ppl": 724.124,
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": " -2.5\u00b0C",
136
+ "answer_nll": 5.1008,
137
+ "answer_ppl": 164.146,
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, a professor of English",
145
+ "answer_nll": 0.989,
146
+ "answer_ppl": 2.688,
147
+ "hit": true,
148
+ "token_f1": 0.0
149
+ },
150
+ {
151
+ "prompt": "The Earth revolves around the",
152
+ "answer": " Sun",
153
+ "generated": " sun, and the moon revolves",
154
+ "answer_nll": 1.0905,
155
+ "answer_ppl": 2.976,
156
+ "hit": true,
157
+ "token_f1": 0.0
158
+ },
159
+ {
160
+ "prompt": "The plural of mouse is",
161
+ "answer": " mice",
162
+ "generated": " mouse.\n- mouse =",
163
+ "answer_nll": 4.5976,
164
+ "answer_ppl": 99.242,
165
+ "hit": false,
166
+ "token_f1": 0.0
167
+ },
168
+ {
169
+ "prompt": "Photosynthesis occurs in the",
170
+ "answer": " leaves",
171
+ "generated": " chloroplast, which is",
172
+ "answer_nll": 3.1152,
173
+ "answer_ppl": 22.538,
174
+ "hit": false,
175
+ "token_f1": 0.0
176
+ },
177
+ {
178
+ "prompt": "The human heart pumps",
179
+ "answer": " blood",
180
+ "generated": " blood through the heart and lungs",
181
+ "answer_nll": 1.616,
182
+ "answer_ppl": 5.033,
183
+ "hit": true,
184
+ "token_f1": 0.286
185
+ },
186
+ {
187
+ "prompt": "The currency used in the United States is the",
188
+ "answer": " dollar",
189
+ "generated": " U.S. dollar.",
190
+ "answer_nll": 2.0529,
191
+ "answer_ppl": 7.79,
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": " African elephant.\nThe elephant",
199
+ "answer_nll": 2.2793,
200
+ "answer_ppl": 9.77,
201
+ "hit": false,
202
+ "token_f1": 0.0
203
+ }
204
+ ],
205
+ "moe/aux_loss": 0.12051594359800219,
206
+ "moe/routing_entropy": 0.9923138906558353,
207
+ "moe/max_load_frac": 0.14490839640299483,
208
+ "moe/expert_0_prob": 0.13816578015685083,
209
+ "moe/expert_1_prob": 0.12321405634284019,
210
+ "moe/expert_2_prob": 0.12382426597177983,
211
+ "moe/expert_3_prob": 0.11399608328938485,
212
+ "moe/expert_4_prob": 0.11689293012022972,
213
+ "moe/expert_5_prob": 0.1181701436638832,
214
+ "moe/expert_6_prob": 0.13688957169651986,
215
+ "moe/expert_7_prob": 0.12884716577827932,
216
+ "generation_samples": {
217
+ "The most important thing about sleep is": "The most important thing about sleep is that it\u2019s not just about eating. It is also about maintaining a healthy body and a healthy personality.\n- Sleep is a huge focus for all of us. Let\u2019s face it: sleep is our first resort.\n- Sleep is the best thing to have for you.",
218
+ "Scientists have discovered that": "Scientists have discovered that, when in fact the earth was not the same size as Earth in the past.\nBut when you put the earth in its place, with all its surface, the earth could have a circumference of a little over one-quarter of its entire surface. So the moon, the moon, this huge",
219
+ "In the history of mathematics,": "In the history of mathematics, the term has been used to describe a set of principles and practices which are in some sense similar to traditional methods of research. It is a special case of the mathematical sciences of the 19th Century, but the term has been used in different ways by different researchers. The traditional methods of research have been",
220
+ "Python code for prime number": "Python code for prime number in the above example. However, the string is not a string itself but a string in the string itself.\nA simple example would be a prime number that is also the prime number of the base $n$N$n$N$n$n\nTo check if a prime number is",
221
+ "The best way to learn a new language is": "The best way to learn a new language is to learn about its culture, geography, economy, technology, history, philosophy, culture, values, etc.\n- If you have some knowledge of the language, you need to think of the culture of that language, and that culture.\n- If you are new to the language, you need"
222
+ }
223
+ }
m15/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:150978df2850d9eae1ab6d89c412c4189b4f5dc70446ab6edf1a5f12850c9ba3
3
+ size 1059507864