B4K2xx commited on
Commit
d13b4ca
·
verified ·
1 Parent(s): ae1dbd2

Upload folder using huggingface_hub

Browse files
m19/benchmark_results.json ADDED
@@ -0,0 +1,540 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m19",
3
+ "checkpoint": "results/m19/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": 920.0575260263694,
13
+ "perplexity_stderr,none": 49.12608073743694,
14
+ "acc,none": 0.11643702697457792,
15
+ "acc_stderr,none": 0.004468650413126095
16
+ },
17
+ "hellaswag": {
18
+ "name": "hellaswag",
19
+ "alias": "hellaswag",
20
+ "sample_len": 10042,
21
+ "acc,none": 0.27185819557857,
22
+ "acc_stderr,none": 0.0044400791732768724,
23
+ "acc_norm,none": 0.27454690300736906,
24
+ "acc_norm_stderr,none": 0.004453735900948104
25
+ },
26
+ "piqa": {
27
+ "name": "piqa",
28
+ "alias": "piqa",
29
+ "sample_len": 1838,
30
+ "acc,none": 0.5870511425462459,
31
+ "acc_stderr,none": 0.011487658725078972,
32
+ "acc_norm,none": 0.6011969532100109,
33
+ "acc_norm_stderr,none": 0.011424390545037256
34
+ },
35
+ "sciq": {
36
+ "name": "sciq",
37
+ "alias": "sciq",
38
+ "sample_len": 1000,
39
+ "acc,none": 0.64,
40
+ "acc_stderr,none": 0.015186527932039985,
41
+ "acc_norm,none": 0.578,
42
+ "acc_norm_stderr,none": 0.015625625112620622
43
+ },
44
+ "arc_easy": {
45
+ "name": "arc_easy",
46
+ "alias": "arc_easy",
47
+ "sample_len": 2376,
48
+ "acc,none": 0.44234006734006737,
49
+ "acc_stderr,none": 0.010191334444220676,
50
+ "acc_norm,none": 0.4065656565656566,
51
+ "acc_norm_stderr,none": 0.01007905641922357
52
+ },
53
+ "winogrande": {
54
+ "name": "winogrande",
55
+ "alias": "winogrande",
56
+ "sample_len": 1267,
57
+ "acc,none": 0.5232833464877664,
58
+ "acc_stderr,none": 0.0140372413095736
59
+ },
60
+ "blimp_adjunct_island": {
61
+ "name": "blimp_adjunct_island",
62
+ "alias": "blimp_adjunct_island",
63
+ "sample_len": 1000,
64
+ "acc,none": 0.83,
65
+ "acc_stderr,none": 0.011884495834541793
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": 0.952,
72
+ "acc_stderr,none": 0.006763264133666652
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": 0.733,
86
+ "acc_stderr,none": 0.013996674851796299
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": 0.913,
93
+ "acc_stderr,none": 0.008916866630745944
94
+ },
95
+ "blimp_causative": {
96
+ "name": "blimp_causative",
97
+ "alias": "blimp_causative",
98
+ "sample_len": 1000,
99
+ "acc,none": 0.705,
100
+ "acc_stderr,none": 0.014428554438445425
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": 0.559,
107
+ "acc_stderr,none": 0.015708779894242766
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": 0.476,
114
+ "acc_stderr,none": 0.01580106558665186
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": 0.821,
121
+ "acc_stderr,none": 0.012128730605719056
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": 0.967,
128
+ "acc_stderr,none": 0.005651808820452346
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": 0.944,
135
+ "acc_stderr,none": 0.007274401481697097
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": 0.909,
142
+ "acc_stderr,none": 0.009099549538400337
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": 0.937,
149
+ "acc_stderr,none": 0.007687007876286461
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": 0.902,
156
+ "acc_stderr,none": 0.009406619184621247
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": 0.895,
163
+ "acc_stderr,none": 0.00969892102602496
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": 0.898,
170
+ "acc_stderr,none": 0.009575368801653944
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": 0.933,
177
+ "acc_stderr,none": 0.007910345983177575
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": 0.781,
184
+ "acc_stderr,none": 0.013084731950262116
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": 0.68,
191
+ "acc_stderr,none": 0.014758652303574938
192
+ },
193
+ "blimp_drop_argument": {
194
+ "name": "blimp_drop_argument",
195
+ "alias": "blimp_drop_argument",
196
+ "sample_len": 1000,
197
+ "acc,none": 0.772,
198
+ "acc_stderr,none": 0.013273740700804434
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": 0.799,
205
+ "acc_stderr,none": 0.012679107214617274
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": 0.804,
212
+ "acc_stderr,none": 0.012559527926707347
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": 0.775,
219
+ "acc_stderr,none": 0.013211720158614833
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": 0.981,
226
+ "acc_stderr,none": 0.004319451082910613
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": 0.318,
233
+ "acc_stderr,none": 0.014734079309311901
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": 0.865,
240
+ "acc_stderr,none": 0.010811655372416006
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": 0.714,
247
+ "acc_stderr,none": 0.014297146862518003
248
+ },
249
+ "blimp_inchoative": {
250
+ "name": "blimp_inchoative",
251
+ "alias": "blimp_inchoative",
252
+ "sample_len": 1000,
253
+ "acc,none": 0.717,
254
+ "acc_stderr,none": 0.01425181090648182
255
+ },
256
+ "blimp_intransitive": {
257
+ "name": "blimp_intransitive",
258
+ "alias": "blimp_intransitive",
259
+ "sample_len": 1000,
260
+ "acc,none": 0.787,
261
+ "acc_stderr,none": 0.012953717566737312
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": 0.861,
268
+ "acc_stderr,none": 0.010945263761042892
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": 0.776,
275
+ "acc_stderr,none": 0.013190830072364589
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": 0.915,
282
+ "acc_stderr,none": 0.008823426366942316
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": 0.894,
289
+ "acc_stderr,none": 0.00973955126578524
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": 0.498,
296
+ "acc_stderr,none": 0.01581917337430266
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": 0.62,
303
+ "acc_stderr,none": 0.015356947477797658
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": 0.595,
310
+ "acc_stderr,none": 0.01553113699045296
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": 0.602,
317
+ "acc_stderr,none": 0.015486634102859016
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": 0.597,
324
+ "acc_stderr,none": 0.015518757419066512
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": 0.996,
331
+ "acc_stderr,none": 0.001996994739098726
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": 0.706,
338
+ "acc_stderr,none": 0.014414290540008265
339
+ },
340
+ "blimp_passive_1": {
341
+ "name": "blimp_passive_1",
342
+ "alias": "blimp_passive_1",
343
+ "sample_len": 1000,
344
+ "acc,none": 0.892,
345
+ "acc_stderr,none": 0.009820001651345708
346
+ },
347
+ "blimp_passive_2": {
348
+ "name": "blimp_passive_2",
349
+ "alias": "blimp_passive_2",
350
+ "sample_len": 1000,
351
+ "acc,none": 0.88,
352
+ "acc_stderr,none": 0.010281328012747462
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": 0.625,
359
+ "acc_stderr,none": 0.015316971293620996
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": 0.953,
373
+ "acc_stderr,none": 0.006695956678163013
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": 0.9,
380
+ "acc_stderr,none": 0.00949157995752499
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": 0.757,
387
+ "acc_stderr,none": 0.013569640199177543
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": 0.632,
394
+ "acc_stderr,none": 0.015258073561521743
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": 0.386,
401
+ "acc_stderr,none": 0.015402637476784335
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": 0.893,
408
+ "acc_stderr,none": 0.009779910359847203
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": 0.893,
415
+ "acc_stderr,none": 0.009779910359847203
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": 0.99,
422
+ "acc_stderr,none": 0.003148000938676784
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": 0.682,
429
+ "acc_stderr,none": 0.014734079309311903
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": 0.538,
436
+ "acc_stderr,none": 0.015773547629015002
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": 0.871,
443
+ "acc_stderr,none": 0.010605256784796466
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": 0.659,
450
+ "acc_stderr,none": 0.014998131348402837
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": 0.661,
457
+ "acc_stderr,none": 0.014976758771620224
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": 0.869,
464
+ "acc_stderr,none": 0.0106748748448379
465
+ },
466
+ "blimp_transitive": {
467
+ "name": "blimp_transitive",
468
+ "alias": "blimp_transitive",
469
+ "sample_len": 1000,
470
+ "acc,none": 0.81,
471
+ "acc_stderr,none": 0.012411851354816254
472
+ },
473
+ "blimp_wh_island": {
474
+ "name": "blimp_wh_island",
475
+ "alias": "blimp_wh_island",
476
+ "sample_len": 1000,
477
+ "acc,none": 0.782,
478
+ "acc_stderr,none": 0.013063179040595235
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": 0.809,
485
+ "acc_stderr,none": 0.012436787112179482
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": 0.943,
492
+ "acc_stderr,none": 0.007335175853706905
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": 0.857,
499
+ "acc_stderr,none": 0.011075814808567074
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": 0.98,
506
+ "acc_stderr,none": 0.004429403980178329
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": 0.974,
513
+ "acc_stderr,none": 0.005034813735318254
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": 0.506,
520
+ "acc_stderr,none": 0.015818160898606836
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": 0.274,
527
+ "acc_stderr,none": 0.01411109928825969
528
+ },
529
+ "blimp": {
530
+ "alias": "blimp",
531
+ "name": "blimp",
532
+ "sample_len": 67000,
533
+ "acc,none": 0.7824925373134328,
534
+ "acc_stderr,none": 0.0014535747039117115,
535
+ "sample_count": {
536
+ "acc,none": 67000
537
+ }
538
+ }
539
+ }
540
+ }
m19/config.json ADDED
@@ -0,0 +1,52 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m19",
3
+ "run_name": "m19-kda-rope-dense-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": "kda",
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",
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": 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
+ "kda_num_heads": 8,
50
+ "kda_head_dim": 64,
51
+ "kda_conv_size": 4
52
+ }
m19/eval_results.json ADDED
@@ -0,0 +1,212 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "m19",
3
+ "checkpoint": "/home/akshat/codemindv2/results/m19/final.pt",
4
+ "trained_step": 15258,
5
+ "total_params": 65202016,
6
+ "active_params": 65202016,
7
+ "flops_per_forward": 130404032.0,
8
+ "val_loss": 3.513864967226982,
9
+ "val_perplexity": 33.577794389533295,
10
+ "bits_per_token": 5.069435562571827,
11
+ "ppl_tokens_evaluated": 655360,
12
+ "prefill_tok_per_sec": 129983.98214409001,
13
+ "inference_tok_per_sec": 51.43506844999894,
14
+ "inference_tok_per_sec_kvcache": null,
15
+ "kvcache_speedup": null,
16
+ "gen_tokens": 128,
17
+ "peak_infer_vram_gb": 0.406,
18
+ "peak_train_vram_gb": 4.68,
19
+ "closeness_answer_ppl": 17.717,
20
+ "closeness_hit_rate": 0.45,
21
+ "closeness_token_f1": 0.05,
22
+ "closeness_n": 20,
23
+ "closeness_items": [
24
+ {
25
+ "prompt": "The capital of France is",
26
+ "answer": " Paris",
27
+ "generated": " the capital of the French Republic",
28
+ "answer_nll": 2.3599,
29
+ "answer_ppl": 10.59,
30
+ "hit": false,
31
+ "token_f1": 0.0
32
+ },
33
+ {
34
+ "prompt": "The capital of Japan is",
35
+ "answer": " Tokyo",
36
+ "generated": " Tokyo, and the capital of",
37
+ "answer_nll": 1.6333,
38
+ "answer_ppl": 5.121,
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": 2.8595,
47
+ "answer_ppl": 17.454,
48
+ "hit": false,
49
+ "token_f1": 0.0
50
+ },
51
+ {
52
+ "prompt": "Water is made of hydrogen and",
53
+ "answer": " oxygen",
54
+ "generated": " oxygen. The water is then",
55
+ "answer_nll": 0.9142,
56
+ "answer_ppl": 2.495,
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": 3.0011,
65
+ "answer_ppl": 20.108,
66
+ "hit": false,
67
+ "token_f1": 0.0
68
+ },
69
+ {
70
+ "prompt": "The opposite of up is",
71
+ "answer": " down",
72
+ "generated": " the \u201cs\u201d",
73
+ "answer_nll": 8.2403,
74
+ "answer_ppl": 3790.537,
75
+ "hit": false,
76
+ "token_f1": 0.0
77
+ },
78
+ {
79
+ "prompt": "Two plus two equals",
80
+ "answer": " four",
81
+ "generated": " two plus two plus two plus",
82
+ "answer_nll": 2.5678,
83
+ "answer_ppl": 13.038,
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": 2.6056,
92
+ "answer_ppl": 13.539,
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": 0.9871,
101
+ "answer_ppl": 2.683,
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": 0.096,
110
+ "answer_ppl": 1.101,
111
+ "hit": true,
112
+ "token_f1": 0.0
113
+ },
114
+ {
115
+ "prompt": "A triangle has three",
116
+ "answer": " sides",
117
+ "generated": " sides, and the two sides",
118
+ "answer_nll": 1.193,
119
+ "answer_ppl": 3.297,
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 the most",
127
+ "answer_nll": 7.6968,
128
+ "answer_ppl": 2201.402,
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 same as the freezing point",
136
+ "answer_nll": 4.8671,
137
+ "answer_ppl": 129.943,
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. He is the author",
145
+ "answer_nll": 0.4312,
146
+ "answer_ppl": 1.539,
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": 0.7212,
155
+ "answer_ppl": 2.057,
156
+ "hit": true,
157
+ "token_f1": 0.333
158
+ },
159
+ {
160
+ "prompt": "The plural of mouse is",
161
+ "answer": " mice",
162
+ "generated": " the mouse, and the word",
163
+ "answer_nll": 8.3891,
164
+ "answer_ppl": 4398.878,
165
+ "hit": false,
166
+ "token_f1": 0.0
167
+ },
168
+ {
169
+ "prompt": "Photosynthesis occurs in the",
170
+ "answer": " leaves",
171
+ "generated": " soil, and the plant uses",
172
+ "answer_nll": 3.4986,
173
+ "answer_ppl": 33.068,
174
+ "hit": false,
175
+ "token_f1": 0.0
176
+ },
177
+ {
178
+ "prompt": "The human heart pumps",
179
+ "answer": " blood",
180
+ "generated": " blood to the brain, where",
181
+ "answer_nll": 1.2565,
182
+ "answer_ppl": 3.513,
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": 2.1939,
191
+ "answer_ppl": 8.97,
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": " giant tortoise.\nThe",
199
+ "answer_nll": 1.9782,
200
+ "answer_ppl": 7.23,
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 makes you feel better.\nSleep is important for your brain and nervous system. And that\u2019s why it is important to stay alert and alert in sleep.\nSleep is crucial for your brain function. And it is important for your brain to use the whole body to find it.",
207
+ "Scientists have discovered that": "Scientists have discovered that when their team looked at the same fish, they found that, on average, they saw only a partial decrease in fish populations when compared with the same groups on that fish.\n\u201cIf you look at the data, you\u2019ve only seen the changes in the fish populations and they\ufffd",
208
+ "In the history of mathematics,": "In the history of mathematics, the notion of number as a mathematical principle is a familiar problem, and therefore it is extremely difficult to be certain that it was invented without the help of mathematics.\nIn the history of mathematics, the foundations of mathematics, including the concepts of space, time, space and time are known.\nModern",
209
+ "Python code for prime number": "Python code for prime number, value, and power.\n- Add the code to the file using the same tool in the variable file.\n- In the example above, use the same tool for data in the variable file.\n- The code in the variable file can be changed in all the functions.\n- In",
210
+ "The best way to learn a new language is": "The best way to learn a new language is to read and write, and to recognize how to express oneself. In this way, you can easily recognize, and use, the beauty of a new language.\nIt is important to understand the difference between an understanding and a meaningful understanding. The more you comprehend, the more you will learn.\n"
211
+ }
212
+ }
m19/model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3cc0d3130a221557efbe07611a1836be4ac45e0ac243b7777b38e493ce42517c
3
+ size 260832368