B4K2xx commited on
Commit
af8b41c
·
verified ·
1 Parent(s): 59c0c04

Upload folder using huggingface_hub

Browse files
m14/benchmark_results.json ADDED
@@ -0,0 +1,540 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m14",
3
+ "checkpoint": "results/m14/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": 404.5905237231641,
13
+ "perplexity_stderr,none": 21.19147790393626,
14
+ "acc,none": 0.18105957694546865,
15
+ "acc_stderr,none": 0.005364743691197477
16
+ },
17
+ "hellaswag": {
18
+ "name": "hellaswag",
19
+ "alias": "hellaswag",
20
+ "sample_len": 10042,
21
+ "acc,none": 0.2801234813782115,
22
+ "acc_stderr,none": 0.004481416180499928,
23
+ "acc_norm,none": 0.293168691495718,
24
+ "acc_norm_stderr,none": 0.0045428479356060215
25
+ },
26
+ "piqa": {
27
+ "name": "piqa",
28
+ "alias": "piqa",
29
+ "sample_len": 1838,
30
+ "acc,none": 0.6093579978237215,
31
+ "acc_stderr,none": 0.011383377760053567,
32
+ "acc_norm,none": 0.6066376496191512,
33
+ "acc_norm_stderr,none": 0.011397418546689315
34
+ },
35
+ "sciq": {
36
+ "name": "sciq",
37
+ "alias": "sciq",
38
+ "sample_len": 1000,
39
+ "acc,none": 0.685,
40
+ "acc_stderr,none": 0.014696631960792617,
41
+ "acc_norm,none": 0.594,
42
+ "acc_norm_stderr,none": 0.01553722643863458
43
+ },
44
+ "arc_easy": {
45
+ "name": "arc_easy",
46
+ "alias": "arc_easy",
47
+ "sample_len": 2376,
48
+ "acc,none": 0.4654882154882155,
49
+ "acc_stderr,none": 0.010235314238969277,
50
+ "acc_norm,none": 0.4313973063973064,
51
+ "acc_norm_stderr,none": 0.010162752847747605
52
+ },
53
+ "winogrande": {
54
+ "name": "winogrande",
55
+ "alias": "winogrande",
56
+ "sample_len": 1267,
57
+ "acc,none": 0.5122336227308603,
58
+ "acc_stderr,none": 0.014048278820405729
59
+ },
60
+ "blimp_adjunct_island": {
61
+ "name": "blimp_adjunct_island",
62
+ "alias": "blimp_adjunct_island",
63
+ "sample_len": 1000,
64
+ "acc,none": 0.804,
65
+ "acc_stderr,none": 0.012559527926707347
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": 0.962,
72
+ "acc_stderr,none": 0.0060491811505849775
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": 0.984,
79
+ "acc_stderr,none": 0.003969856390319398
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": 0.719,
86
+ "acc_stderr,none": 0.01422115470843494
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": 0.749,
100
+ "acc_stderr,none": 0.013718133516888775
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": 0.522,
107
+ "acc_stderr,none": 0.01580397942816194
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": 0.511,
114
+ "acc_stderr,none": 0.01581547119529257
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": 0.816,
121
+ "acc_stderr,none": 0.01225945734093864
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": 0.976,
128
+ "acc_stderr,none": 0.004842256441727084
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": 0.954,
135
+ "acc_stderr,none": 0.006627814717380679
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": 0.903,
142
+ "acc_stderr,none": 0.009363689373248062
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": 0.931,
149
+ "acc_stderr,none": 0.008018934050315124
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": 0.927,
156
+ "acc_stderr,none": 0.008230354715243976
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": 0.878,
163
+ "acc_stderr,none": 0.010354864712936776
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": 0.891,
170
+ "acc_stderr,none": 0.009859828407037124
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": 0.934,
177
+ "acc_stderr,none": 0.007855297938697532
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": 0.757,
184
+ "acc_stderr,none": 0.013569640199177543
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": 0.617,
191
+ "acc_stderr,none": 0.0153801023256527
192
+ },
193
+ "blimp_drop_argument": {
194
+ "name": "blimp_drop_argument",
195
+ "alias": "blimp_drop_argument",
196
+ "sample_len": 1000,
197
+ "acc,none": 0.771,
198
+ "acc_stderr,none": 0.013294199326613684
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": 0.801,
205
+ "acc_stderr,none": 0.01263164908309911
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": 0.838,
212
+ "acc_stderr,none": 0.011657267771304384
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": 0.774,
219
+ "acc_stderr,none": 0.013232501619085452
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": 0.97,
226
+ "acc_stderr,none": 0.0053971408290991564
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": 0.326,
233
+ "acc_stderr,none": 0.014830507204541047
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": 0.845,
240
+ "acc_stderr,none": 0.011450157470799522
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": 0.791,
247
+ "acc_stderr,none": 0.012864077288499337
248
+ },
249
+ "blimp_inchoative": {
250
+ "name": "blimp_inchoative",
251
+ "alias": "blimp_inchoative",
252
+ "sample_len": 1000,
253
+ "acc,none": 0.691,
254
+ "acc_stderr,none": 0.014619600977206347
255
+ },
256
+ "blimp_intransitive": {
257
+ "name": "blimp_intransitive",
258
+ "alias": "blimp_intransitive",
259
+ "sample_len": 1000,
260
+ "acc,none": 0.782,
261
+ "acc_stderr,none": 0.013063179040595235
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": 0.934,
268
+ "acc_stderr,none": 0.007855297938697532
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": 0.9,
275
+ "acc_stderr,none": 0.00949157995752499
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": 0.911,
282
+ "acc_stderr,none": 0.009008893392651516
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": 0.896,
289
+ "acc_stderr,none": 0.009658016218524242
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": 0.576,
296
+ "acc_stderr,none": 0.015635487471405186
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": 0.615,
303
+ "acc_stderr,none": 0.015395194445410945
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": 0.521,
310
+ "acc_stderr,none": 0.015805341148131185
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": 0.614,
317
+ "acc_stderr,none": 0.015402637476784335
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": 0.691,
324
+ "acc_stderr,none": 0.014619600977206347
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": 0.976,
331
+ "acc_stderr,none": 0.004842256441727084
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": 0.582,
338
+ "acc_stderr,none": 0.015605111967541904
339
+ },
340
+ "blimp_passive_1": {
341
+ "name": "blimp_passive_1",
342
+ "alias": "blimp_passive_1",
343
+ "sample_len": 1000,
344
+ "acc,none": 0.87,
345
+ "acc_stderr,none": 0.010640169792499236
346
+ },
347
+ "blimp_passive_2": {
348
+ "name": "blimp_passive_2",
349
+ "alias": "blimp_passive_2",
350
+ "sample_len": 1000,
351
+ "acc,none": 0.825,
352
+ "acc_stderr,none": 0.012021627157731998
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": 0.58,
359
+ "acc_stderr,none": 0.015615500115072966
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": 0.915,
373
+ "acc_stderr,none": 0.008823426366942316
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": 0.877,
380
+ "acc_stderr,none": 0.010391293421849803
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": 0.764,
387
+ "acc_stderr,none": 0.013434451402438602
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": 0.637,
394
+ "acc_stderr,none": 0.01521389044467136
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": 0.51,
401
+ "acc_stderr,none": 0.015816135752773183
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": 0.92,
408
+ "acc_stderr,none": 0.00858333697775368
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": 0.878,
415
+ "acc_stderr,none": 0.010354864712936776
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": 0.987,
422
+ "acc_stderr,none": 0.003583830889403662
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": 0.652,
429
+ "acc_stderr,none": 0.015070604603768328
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": 0.506,
436
+ "acc_stderr,none": 0.015818160898606836
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": 0.934,
443
+ "acc_stderr,none": 0.007855297938697532
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": 0.683,
450
+ "acc_stderr,none": 0.014721675438880174
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": 0.673,
457
+ "acc_stderr,none": 0.014842213153411162
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": 0.876,
464
+ "acc_stderr,none": 0.010427498872343873
465
+ },
466
+ "blimp_transitive": {
467
+ "name": "blimp_transitive",
468
+ "alias": "blimp_transitive",
469
+ "sample_len": 1000,
470
+ "acc,none": 0.809,
471
+ "acc_stderr,none": 0.012436787112179482
472
+ },
473
+ "blimp_wh_island": {
474
+ "name": "blimp_wh_island",
475
+ "alias": "blimp_wh_island",
476
+ "sample_len": 1000,
477
+ "acc,none": 0.855,
478
+ "acc_stderr,none": 0.0111399775178902
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": 0.767,
485
+ "acc_stderr,none": 0.01337497251921999
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": 0.917,
492
+ "acc_stderr,none": 0.008728527206074756
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": 0.847,
499
+ "acc_stderr,none": 0.011389500459665504
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": 0.969,
506
+ "acc_stderr,none": 0.005483527064679181
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": 0.95,
513
+ "acc_stderr,none": 0.006895472974897918
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": 0.528,
520
+ "acc_stderr,none": 0.015794475789511517
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": 0.284,
527
+ "acc_stderr,none": 0.014267009061031241
528
+ },
529
+ "blimp": {
530
+ "alias": "blimp",
531
+ "name": "blimp",
532
+ "sample_len": 67000,
533
+ "acc,none": 0.7844776119402985,
534
+ "acc_stderr,none": 0.0014535278370615976,
535
+ "sample_count": {
536
+ "acc,none": 67000
537
+ }
538
+ }
539
+ }
540
+ }
m14/config.json ADDED
@@ -0,0 +1,50 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m14",
3
+ "run_name": "m14-mha-rope-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": "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_aux_loss_coef": 0.01
50
+ }
m14/eval_results.json ADDED
@@ -0,0 +1,223 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m14",
3
+ "checkpoint": "results/m14/final.pt",
4
+ "trained_step": 15258,
5
+ "total_params": 239703040,
6
+ "active_params": 88708096,
7
+ "flops_per_forward": 177416192.0,
8
+ "val_loss": 3.4363590747118,
9
+ "val_perplexity": 31.07361524541882,
10
+ "bits_per_token": 4.9576181958005,
11
+ "ppl_tokens_evaluated": 655360,
12
+ "prefill_tok_per_sec": 80126.71676878858,
13
+ "inference_tok_per_sec": 46.77562740153019,
14
+ "inference_tok_per_sec_kvcache": 79.65645893983364,
15
+ "kvcache_speedup": 1.7029479531305616,
16
+ "gen_tokens": 128,
17
+ "peak_infer_vram_gb": 1.452,
18
+ "peak_train_vram_gb": 3.608,
19
+ "closeness_answer_ppl": 13.041,
20
+ "closeness_hit_rate": 0.55,
21
+ "closeness_token_f1": 0.068,
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": 3.202,
29
+ "answer_ppl": 24.582,
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. The",
37
+ "answer_nll": 2.5489,
38
+ "answer_ppl": 12.793,
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": 2.4559,
47
+ "answer_ppl": 11.657,
48
+ "hit": false,
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": 0.2959,
56
+ "answer_ppl": 1.344,
57
+ "hit": true,
58
+ "token_f1": 0.0
59
+ },
60
+ {
61
+ "prompt": "The opposite of hot is",
62
+ "answer": " cold",
63
+ "generated": " hot. Hot is the heat",
64
+ "answer_nll": 4.2344,
65
+ "answer_ppl": 69.019,
66
+ "hit": false,
67
+ "token_f1": 0.0
68
+ },
69
+ {
70
+ "prompt": "The opposite of up is",
71
+ "answer": " down",
72
+ "generated": " the difference between the two.",
73
+ "answer_nll": 4.0606,
74
+ "answer_ppl": 58.008,
75
+ "hit": false,
76
+ "token_f1": 0.0
77
+ },
78
+ {
79
+ "prompt": "Two plus two equals",
80
+ "answer": " four",
81
+ "generated": " two.\nThe first is",
82
+ "answer_nll": 2.1366,
83
+ "answer_ppl": 8.471,
84
+ "hit": false,
85
+ "token_f1": 0.0
86
+ },
87
+ {
88
+ "prompt": "The sun rises in the",
89
+ "answer": " east",
90
+ "generated": " morning and sets in the evening",
91
+ "answer_nll": 2.7328,
92
+ "answer_ppl": 15.375,
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.\nThe first president",
100
+ "answer_nll": 0.2571,
101
+ "answer_ppl": 1.293,
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": 0.0656,
110
+ "answer_ppl": 1.068,
111
+ "hit": true,
112
+ "token_f1": 0.0
113
+ },
114
+ {
115
+ "prompt": "A triangle has three",
116
+ "answer": " sides",
117
+ "generated": " sides, and the third side",
118
+ "answer_nll": 1.3205,
119
+ "answer_ppl": 3.745,
120
+ "hit": true,
121
+ "token_f1": 0.0
122
+ },
123
+ {
124
+ "prompt": "The chemical symbol for gold is",
125
+ "answer": " Au",
126
+ "generated": " gold. Gold is the most",
127
+ "answer_nll": 6.5487,
128
+ "answer_ppl": 698.318,
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": " about 1.5 \u00b0C",
136
+ "answer_nll": 5.9355,
137
+ "answer_ppl": 378.246,
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 play is",
145
+ "answer_nll": 0.1597,
146
+ "answer_ppl": 1.173,
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 Sun is",
154
+ "answer_nll": 0.6747,
155
+ "answer_ppl": 1.963,
156
+ "hit": true,
157
+ "token_f1": 0.333
158
+ },
159
+ {
160
+ "prompt": "The plural of mouse is",
161
+ "answer": " mice",
162
+ "generated": " the mouse.\nThe word",
163
+ "answer_nll": 5.1476,
164
+ "answer_ppl": 172.017,
165
+ "hit": false,
166
+ "token_f1": 0.0
167
+ },
168
+ {
169
+ "prompt": "Photosynthesis occurs in the",
170
+ "answer": " leaves",
171
+ "generated": " leaves of the plant.\n",
172
+ "answer_nll": 3.5603,
173
+ "answer_ppl": 35.174,
174
+ "hit": true,
175
+ "token_f1": 0.4
176
+ },
177
+ {
178
+ "prompt": "The human heart pumps",
179
+ "answer": " blood",
180
+ "generated": " blood to the rest of the",
181
+ "answer_nll": 1.1786,
182
+ "answer_ppl": 3.25,
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": " dollar. The dollar is the",
190
+ "answer_nll": 2.5244,
191
+ "answer_ppl": 12.483,
192
+ "hit": true,
193
+ "token_f1": 0.333
194
+ },
195
+ {
196
+ "prompt": "The tallest animal in the world is the",
197
+ "answer": " giraffe",
198
+ "generated": " lion.\nThe lion is",
199
+ "answer_nll": 2.3226,
200
+ "answer_ppl": 10.202,
201
+ "hit": false,
202
+ "token_f1": 0.0
203
+ }
204
+ ],
205
+ "moe/aux_loss": 0.12053199759684503,
206
+ "moe/routing_entropy": 0.9948502505819002,
207
+ "moe/max_load_frac": 0.1416552225748698,
208
+ "moe/expert_0_prob": 0.12510733157396317,
209
+ "moe/expert_1_prob": 0.1254948280751705,
210
+ "moe/expert_2_prob": 0.130660605058074,
211
+ "moe/expert_3_prob": 0.12439362742006779,
212
+ "moe/expert_4_prob": 0.12149989195168018,
213
+ "moe/expert_5_prob": 0.13262441903352737,
214
+ "moe/expert_6_prob": 0.12078449092805385,
215
+ "moe/expert_7_prob": 0.11943480558693409,
216
+ "generation_samples": {
217
+ "The most important thing about sleep is": "The most important thing about sleep is that it\u2019s a good idea to prepare yourself and your family for the day ahead. If you find that you get into bed, you might be craving for the night. You might be feeling a bit hungry later, and also feeling the energy back to sleep.\nWhat happens if you can",
218
+ "Scientists have discovered that": "Scientists have discovered that, in many ways, the human body is a complex machine with four chambers, eight organs, five nerve paths, one central nerve, two peripheral nerve pathways, three peripheral nerve pathways, and a small number of other systems.\nThe human body is a complex machine with four distinct layers. The outer",
219
+ "In the history of mathematics,": "In the history of mathematics, the term \"curriculum\" is often used to describe any subject that is relevant to the science or mathematics curriculum. The term \"curriculum\" is often used in relation to the broader subject of mathematics curriculum.\nThe term \"curriculum\" is used more widely in the United States",
220
+ "Python code for prime number": "Python code for prime number of elements for free. However, the program is not capable to compute integers and can only compute integers, which is a requirement for the first and last elements of the Java package.<|endoftext|>The name Gogol is derived from the Greek word for \"horn\",\nwhere the word \"horn\" means",
221
+ "The best way to learn a new language is": "The best way to learn a new language is to learn about its culture, culture, and language. Learning a new language is something that can be enjoyed by everyone. If you have ever wondered how the language works, you know that it\u2019s used in different cultures. Learning the language is a beautiful language because it allows you to express yourself"
222
+ }
223
+ }
m14/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:bd1ca02cd62bf3170452324a914b4d51064a30bf6569a6676d6bcabbdc88dcda
3
+ size 958841736