B4K2xx commited on
Commit
5923142
·
verified ·
1 Parent(s): 39817da

Upload folder using huggingface_hub

Browse files
m16/benchmark_results.json ADDED
@@ -0,0 +1,540 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m16",
3
+ "checkpoint": "results/m16/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": 488.6423869861212,
13
+ "perplexity_stderr,none": 27.277809731578422,
14
+ "acc,none": 0.18183582379196583,
15
+ "acc_stderr,none": 0.005373682797180109
16
+ },
17
+ "hellaswag": {
18
+ "name": "hellaswag",
19
+ "alias": "hellaswag",
20
+ "sample_len": 10042,
21
+ "acc,none": 0.27813184624576776,
22
+ "acc_stderr,none": 0.0044716295468955935,
23
+ "acc_norm,none": 0.2913762198765186,
24
+ "acc_norm_stderr,none": 0.004534677750103097
25
+ },
26
+ "piqa": {
27
+ "name": "piqa",
28
+ "alias": "piqa",
29
+ "sample_len": 1838,
30
+ "acc,none": 0.6202393906420022,
31
+ "acc_stderr,none": 0.011323483504715666,
32
+ "acc_norm,none": 0.6082698585418934,
33
+ "acc_norm_stderr,none": 0.011389038597720108
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": 0.584,
42
+ "acc_norm_stderr,none": 0.015594460144140522
43
+ },
44
+ "arc_easy": {
45
+ "name": "arc_easy",
46
+ "alias": "arc_easy",
47
+ "sample_len": 2376,
48
+ "acc,none": 0.48484848484848486,
49
+ "acc_stderr,none": 0.01025507179453148,
50
+ "acc_norm,none": 0.43686868686868685,
51
+ "acc_norm_stderr,none": 0.0101776729281577
52
+ },
53
+ "winogrande": {
54
+ "name": "winogrande",
55
+ "alias": "winogrande",
56
+ "sample_len": 1267,
57
+ "acc,none": 0.5303867403314917,
58
+ "acc_stderr,none": 0.014026510839428848
59
+ },
60
+ "blimp_adjunct_island": {
61
+ "name": "blimp_adjunct_island",
62
+ "alias": "blimp_adjunct_island",
63
+ "sample_len": 1000,
64
+ "acc,none": 0.814,
65
+ "acc_stderr,none": 0.012310790208412926
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": 0.928,
72
+ "acc_stderr,none": 0.008178195576218687
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": 0.969,
79
+ "acc_stderr,none": 0.005483527064679181
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": 0.694,
86
+ "acc_stderr,none": 0.014580006055436898
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": 0.907,
93
+ "acc_stderr,none": 0.009188875634996669
94
+ },
95
+ "blimp_causative": {
96
+ "name": "blimp_causative",
97
+ "alias": "blimp_causative",
98
+ "sample_len": 1000,
99
+ "acc,none": 0.708,
100
+ "acc_stderr,none": 0.014385511563477433
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": 0.521,
107
+ "acc_stderr,none": 0.015805341148131185
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": 0.487,
114
+ "acc_stderr,none": 0.015813952101896636
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": 0.786,
121
+ "acc_stderr,none": 0.012975838021968847
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": 0.959,
128
+ "acc_stderr,none": 0.006273624021118764
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": 0.933,
135
+ "acc_stderr,none": 0.007910345983177575
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": 0.9,
142
+ "acc_stderr,none": 0.00949157995752499
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": 0.938,
149
+ "acc_stderr,none": 0.007629823996280269
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": 0.889,
156
+ "acc_stderr,none": 0.009938701010583716
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": 0.856,
163
+ "acc_stderr,none": 0.01110798754893916
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": 0.879,
170
+ "acc_stderr,none": 0.010318210380946179
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": 0.919,
177
+ "acc_stderr,none": 0.008632121032139986
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": 0.833,
184
+ "acc_stderr,none": 0.011800434324644593
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": 0.744,
191
+ "acc_stderr,none": 0.013807775152234145
192
+ },
193
+ "blimp_drop_argument": {
194
+ "name": "blimp_drop_argument",
195
+ "alias": "blimp_drop_argument",
196
+ "sample_len": 1000,
197
+ "acc,none": 0.761,
198
+ "acc_stderr,none": 0.013493000446937704
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": 0.817,
205
+ "acc_stderr,none": 0.012233587399477745
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": 0.832,
212
+ "acc_stderr,none": 0.011828605831454352
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": 0.805,
219
+ "acc_stderr,none": 0.012535235623319256
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": 0.121,
233
+ "acc_stderr,none": 0.010318210380946179
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": 0.832,
240
+ "acc_stderr,none": 0.011828605831454352
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": 0.728,
247
+ "acc_stderr,none": 0.014078856992462642
248
+ },
249
+ "blimp_inchoative": {
250
+ "name": "blimp_inchoative",
251
+ "alias": "blimp_inchoative",
252
+ "sample_len": 1000,
253
+ "acc,none": 0.702,
254
+ "acc_stderr,none": 0.014470846741134585
255
+ },
256
+ "blimp_intransitive": {
257
+ "name": "blimp_intransitive",
258
+ "alias": "blimp_intransitive",
259
+ "sample_len": 1000,
260
+ "acc,none": 0.767,
261
+ "acc_stderr,none": 0.01337497251921999
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": 0.813,
268
+ "acc_stderr,none": 0.012336254828074168
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": 0.815,
275
+ "acc_stderr,none": 0.012285191326386656
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": 0.914,
282
+ "acc_stderr,none": 0.008870325962594761
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": 0.87,
289
+ "acc_stderr,none": 0.010640169792499236
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": 0.536,
296
+ "acc_stderr,none": 0.015778243024904673
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": 0.609,
303
+ "acc_stderr,none": 0.015438826294681775
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": 0.548,
310
+ "acc_stderr,none": 0.01574623586588058
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": 0.578,
317
+ "acc_stderr,none": 0.015625625112620622
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": 0.623,
324
+ "acc_stderr,none": 0.0153331701257798
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": 0.978,
331
+ "acc_stderr,none": 0.004640855259274689
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": 0.732,
338
+ "acc_stderr,none": 0.014013292702729519
339
+ },
340
+ "blimp_passive_1": {
341
+ "name": "blimp_passive_1",
342
+ "alias": "blimp_passive_1",
343
+ "sample_len": 1000,
344
+ "acc,none": 0.849,
345
+ "acc_stderr,none": 0.011328165223341657
346
+ },
347
+ "blimp_passive_2": {
348
+ "name": "blimp_passive_2",
349
+ "alias": "blimp_passive_2",
350
+ "sample_len": 1000,
351
+ "acc,none": 0.805,
352
+ "acc_stderr,none": 0.012535235623319256
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": 0.717,
359
+ "acc_stderr,none": 0.01425181090648182
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": 0.884,
373
+ "acc_stderr,none": 0.010131468138756922
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": 0.906,
380
+ "acc_stderr,none": 0.009233052000787672
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": 0.672,
387
+ "acc_stderr,none": 0.01485384248727031
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": 0.612,
394
+ "acc_stderr,none": 0.015417317979911216
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": 0.319,
401
+ "acc_stderr,none": 0.014746404865473364
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": 0.811,
408
+ "acc_stderr,none": 0.01238678458811779
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": 0.842,
415
+ "acc_stderr,none": 0.011539894677559635
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": 0.978,
422
+ "acc_stderr,none": 0.004640855259274689
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": 0.566,
429
+ "acc_stderr,none": 0.01568087656637504
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": 0.415,
436
+ "acc_stderr,none": 0.015589035185604594
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": 0.937,
443
+ "acc_stderr,none": 0.007687007876286461
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": 0.765,
450
+ "acc_stderr,none": 0.013414729030247124
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": 0.676,
457
+ "acc_stderr,none": 0.014806864733738986
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": 0.891,
464
+ "acc_stderr,none": 0.009859828407037124
465
+ },
466
+ "blimp_transitive": {
467
+ "name": "blimp_transitive",
468
+ "alias": "blimp_transitive",
469
+ "sample_len": 1000,
470
+ "acc,none": 0.818,
471
+ "acc_stderr,none": 0.0122075806376622
472
+ },
473
+ "blimp_wh_island": {
474
+ "name": "blimp_wh_island",
475
+ "alias": "blimp_wh_island",
476
+ "sample_len": 1000,
477
+ "acc,none": 0.758,
478
+ "acc_stderr,none": 0.013550631705556003
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": 0.749,
485
+ "acc_stderr,none": 0.013718133516888775
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": 0.905,
492
+ "acc_stderr,none": 0.009276910103103353
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": 0.838,
499
+ "acc_stderr,none": 0.011657267771304384
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": 0.956,
506
+ "acc_stderr,none": 0.006488921798427387
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": 0.927,
513
+ "acc_stderr,none": 0.008230354715243976
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": 0.609,
520
+ "acc_stderr,none": 0.015438826294681775
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": 0.374,
527
+ "acc_stderr,none": 0.015308767369006505
528
+ },
529
+ "blimp": {
530
+ "alias": "blimp",
531
+ "name": "blimp",
532
+ "sample_len": 67000,
533
+ "acc,none": 0.7700447761194029,
534
+ "acc_stderr,none": 0.0014815512338215082,
535
+ "sample_count": {
536
+ "acc,none": 67000
537
+ }
538
+ }
539
+ }
540
+ }
m16/config.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m16",
3
+ "run_name": "m16-mha-rope-fat-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_shared_ffn_mult": 8,
51
+ "moe_aux_loss_coef": 0.01
52
+ }
m16/eval_results.json ADDED
@@ -0,0 +1,223 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m16",
3
+ "checkpoint": "results/m16/final.pt",
4
+ "trained_step": 15258,
5
+ "total_params": 290034688,
6
+ "active_params": 139039744,
7
+ "flops_per_forward": 278079488.0,
8
+ "val_loss": 3.45664686858654,
9
+ "val_perplexity": 31.710468678414106,
10
+ "bits_per_token": 4.986887295414166,
11
+ "ppl_tokens_evaluated": 655360,
12
+ "prefill_tok_per_sec": 57482.46046879679,
13
+ "inference_tok_per_sec": 41.88420904730769,
14
+ "inference_tok_per_sec_kvcache": 70.19531699040982,
15
+ "kvcache_speedup": 1.6759375093158637,
16
+ "gen_tokens": 128,
17
+ "peak_infer_vram_gb": 1.754,
18
+ "peak_train_vram_gb": 4.338,
19
+ "closeness_answer_ppl": 14.486,
20
+ "closeness_hit_rate": 0.65,
21
+ "closeness_token_f1": 0.064,
22
+ "closeness_n": 20,
23
+ "closeness_items": [
24
+ {
25
+ "prompt": "The capital of France is",
26
+ "answer": " Paris",
27
+ "generated": " Paris, and the capital of",
28
+ "answer_nll": 1.6583,
29
+ "answer_ppl": 5.25,
30
+ "hit": true,
31
+ "token_f1": 0.0
32
+ },
33
+ {
34
+ "prompt": "The capital of Japan is",
35
+ "answer": " Tokyo",
36
+ "generated": " Tokyo.\nThe city is",
37
+ "answer_nll": 1.607,
38
+ "answer_ppl": 4.988,
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": " Jupiter, which is about the",
46
+ "answer_nll": 1.1213,
47
+ "answer_ppl": 3.069,
48
+ "hit": true,
49
+ "token_f1": 0.0
50
+ },
51
+ {
52
+ "prompt": "Water is made of hydrogen and",
53
+ "answer": " oxygen",
54
+ "generated": " oxygen.\nThe hydrogen is",
55
+ "answer_nll": 0.3562,
56
+ "answer_ppl": 1.428,
57
+ "hit": true,
58
+ "token_f1": 0.0
59
+ },
60
+ {
61
+ "prompt": "The opposite of hot is",
62
+ "answer": " cold",
63
+ "generated": " cold, and the cold is",
64
+ "answer_nll": 1.8448,
65
+ "answer_ppl": 6.327,
66
+ "hit": true,
67
+ "token_f1": 0.333
68
+ },
69
+ {
70
+ "prompt": "The opposite of up is",
71
+ "answer": " down",
72
+ "generated": " the fact that the two are",
73
+ "answer_nll": 7.4933,
74
+ "answer_ppl": 1795.958,
75
+ "hit": false,
76
+ "token_f1": 0.0
77
+ },
78
+ {
79
+ "prompt": "Two plus two equals",
80
+ "answer": " four",
81
+ "generated": " two.\n- The ratio",
82
+ "answer_nll": 3.8241,
83
+ "answer_ppl": 45.792,
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": 1.4976,
92
+ "answer_ppl": 4.471,
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 born in",
100
+ "answer_nll": 0.1429,
101
+ "answer_ppl": 1.154,
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.\nThe speed of",
109
+ "answer_nll": 0.0854,
110
+ "answer_ppl": 1.089,
111
+ "hit": true,
112
+ "token_f1": 0.0
113
+ },
114
+ {
115
+ "prompt": "A triangle has three",
116
+ "answer": " sides",
117
+ "generated": " sides, and the sides of",
118
+ "answer_nll": 0.6984,
119
+ "answer_ppl": 2.011,
120
+ "hit": true,
121
+ "token_f1": 0.333
122
+ },
123
+ {
124
+ "prompt": "The chemical symbol for gold is",
125
+ "answer": " Au",
126
+ "generated": " gold. Gold is a precious",
127
+ "answer_nll": 5.97,
128
+ "answer_ppl": 391.516,
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 freezing point of water in",
136
+ "answer_nll": 6.6724,
137
+ "answer_ppl": 790.297,
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": 0.7232,
146
+ "answer_ppl": 2.061,
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 Earth is",
154
+ "answer_nll": 0.5369,
155
+ "answer_ppl": 1.711,
156
+ "hit": true,
157
+ "token_f1": 0.0
158
+ },
159
+ {
160
+ "prompt": "The plural of mouse is",
161
+ "answer": " mice",
162
+ "generated": " a word that is used to",
163
+ "answer_nll": 5.9564,
164
+ "answer_ppl": 386.222,
165
+ "hit": false,
166
+ "token_f1": 0.0
167
+ },
168
+ {
169
+ "prompt": "Photosynthesis occurs in the",
170
+ "answer": " leaves",
171
+ "generated": " form of photosynthesis.\n",
172
+ "answer_nll": 5.4259,
173
+ "answer_ppl": 227.222,
174
+ "hit": false,
175
+ "token_f1": 0.0
176
+ },
177
+ {
178
+ "prompt": "The human heart pumps",
179
+ "answer": " blood",
180
+ "generated": " blood through the body, and",
181
+ "answer_nll": 1.5057,
182
+ "answer_ppl": 4.507,
183
+ "hit": true,
184
+ "token_f1": 0.333
185
+ },
186
+ {
187
+ "prompt": "The currency used in the United States is the",
188
+ "answer": " dollar",
189
+ "generated": " U.S. dollar.",
190
+ "answer_nll": 3.8917,
191
+ "answer_ppl": 48.992,
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": " elephant.\nThe elephant is",
199
+ "answer_nll": 2.4518,
200
+ "answer_ppl": 11.609,
201
+ "hit": false,
202
+ "token_f1": 0.0
203
+ }
204
+ ],
205
+ "moe/aux_loss": 0.12057399628683925,
206
+ "moe/routing_entropy": 0.9932721691826979,
207
+ "moe/max_load_frac": 0.1440912882486979,
208
+ "moe/expert_0_prob": 0.12709741964936255,
209
+ "moe/expert_1_prob": 0.12377416417002678,
210
+ "moe/expert_2_prob": 0.12426500096917152,
211
+ "moe/expert_3_prob": 0.10939359031617642,
212
+ "moe/expert_4_prob": 0.13168378099799155,
213
+ "moe/expert_5_prob": 0.13501964434981345,
214
+ "moe/expert_6_prob": 0.12264941483736039,
215
+ "moe/expert_7_prob": 0.1261169958859682,
216
+ "generation_samples": {
217
+ "The most important thing about sleep is": "The most important thing about sleep is that it\u2019s a combination of many things.\n- It\u2019s not a matter of a single event.\n- It\u2019s about getting the right amount of sleep each night.\n- It\u2019s about the right amount of sleep each night.\n- It can",
218
+ "Scientists have discovered that": "Scientists have discovered that, in many ways, the human brain is actually a computer that is \"wired\" to \"learn\" to become a computer.\n\"These findings are the first of many to have a positive impact on how we think about the way we think about computers, how we understand how computers work, and",
219
+ "In the history of mathematics,": "In the history of mathematics, the term arithmetic began to be used as a synonym for arithmetic.\nThe word \u201calgebra\u201d has its origins in the Greek word \u201calgebra,\u201d (plus, minus) meaning problem solving. The term algebra is from the Greek word algebra, or \u201cal",
220
+ "Python code for prime number": "Python code for prime number to 5 decimal number. However, the program does have two solutions for prime and prime operations. One is for prime numbers (25) and the other for prime numbers (50).\nThe program can be run in almost any programming language. It includes both the\n- Python library (which allows for",
221
+ "The best way to learn a new language is": "The best way to learn a new language is to learn it in a fun and fun way. When you are learning a language, you should have a wide variety of skills and tools. You should be able to read a language without having to master the language in your head. Learning a new language is a great way to learn new things.\n"
222
+ }
223
+ }
m16/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:44293dbf2ffd26d5705179b90163359d4b5ef39e7a0f3750844a73d086fe7bf2
3
+ size 1160171192