lomahony commited on
Commit
6b539cf
1 Parent(s): a531b0e

Upload 6 files

Browse files
sft-160m-eval-files/dpo-pythia-160m-0shot-shelloutput.txt ADDED
@@ -0,0 +1,430 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ bootstrapping for stddev: perplexity
2
+ {
3
+ "results": {
4
+ "arc_challenge": {
5
+ "acc,none": 0.19283276450511946,
6
+ "acc_stderr,none": 0.01152905546566333,
7
+ "acc_norm,none": 0.24488054607508533,
8
+ "acc_norm_stderr,none": 0.012566273985131358
9
+ },
10
+ "arc_easy": {
11
+ "acc,none": 0.4671717171717172,
12
+ "acc_stderr,none": 0.010237645778853858,
13
+ "acc_norm,none": 0.4132996632996633,
14
+ "acc_norm_stderr,none": 0.010104361780747516
15
+ },
16
+ "boolq": {
17
+ "acc,none": 0.6149847094801223,
18
+ "acc_stderr,none": 0.00851066875102728
19
+ },
20
+ "hellaswag": {
21
+ "acc,none": 0.28958374825731925,
22
+ "acc_stderr,none": 0.004526422125860652,
23
+ "acc_norm,none": 0.3016331408086039,
24
+ "acc_norm_stderr,none": 0.004580288728196038
25
+ },
26
+ "lambada_openai": {
27
+ "perplexity,none": 40.49750927655119,
28
+ "perplexity_stderr,none": 1.9470980651595484,
29
+ "acc,none": 0.35066951290510384,
30
+ "acc_stderr,none": 0.006648045374603881
31
+ },
32
+ "openbookqa": {
33
+ "acc,none": 0.172,
34
+ "acc_stderr,none": 0.01689386887634748,
35
+ "acc_norm,none": 0.28,
36
+ "acc_norm_stderr,none": 0.020099950647503237
37
+ },
38
+ "piqa": {
39
+ "acc,none": 0.6332970620239391,
40
+ "acc_stderr,none": 0.011243625019038255,
41
+ "acc_norm,none": 0.6262241566920566,
42
+ "acc_norm_stderr,none": 0.011287972563201014
43
+ },
44
+ "sciq": {
45
+ "acc,none": 0.753,
46
+ "acc_stderr,none": 0.013644675781314128,
47
+ "acc_norm,none": 0.67,
48
+ "acc_norm_stderr,none": 0.014876872027456732
49
+ },
50
+ "wikitext": {
51
+ "word_perplexity,none": 75.24993841350984,
52
+ "byte_perplexity,none": 2.0386909133746456,
53
+ "bits_per_byte,none": 1.0276430644458452
54
+ },
55
+ "winogrande": {
56
+ "acc,none": 0.5138121546961326,
57
+ "acc_stderr,none": 0.014047122916440419
58
+ }
59
+ },
60
+ "configs": {
61
+ "arc_challenge": {
62
+ "task": "arc_challenge",
63
+ "group": [
64
+ "ai2_arc",
65
+ "multiple_choice"
66
+ ],
67
+ "dataset_path": "ai2_arc",
68
+ "dataset_name": "ARC-Challenge",
69
+ "training_split": "train",
70
+ "validation_split": "validation",
71
+ "test_split": "test",
72
+ "doc_to_text": "Question: {{question}}\nAnswer:",
73
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
74
+ "doc_to_choice": "{{choices.text}}",
75
+ "description": "",
76
+ "target_delimiter": " ",
77
+ "fewshot_delimiter": "\n\n",
78
+ "num_fewshot": 0,
79
+ "metric_list": [
80
+ {
81
+ "metric": "acc",
82
+ "aggregation": "mean",
83
+ "higher_is_better": true
84
+ },
85
+ {
86
+ "metric": "acc_norm",
87
+ "aggregation": "mean",
88
+ "higher_is_better": true
89
+ }
90
+ ],
91
+ "output_type": "multiple_choice",
92
+ "repeats": 1,
93
+ "should_decontaminate": true,
94
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
95
+ },
96
+ "arc_easy": {
97
+ "task": "arc_easy",
98
+ "group": [
99
+ "ai2_arc",
100
+ "multiple_choice"
101
+ ],
102
+ "dataset_path": "ai2_arc",
103
+ "dataset_name": "ARC-Easy",
104
+ "training_split": "train",
105
+ "validation_split": "validation",
106
+ "test_split": "test",
107
+ "doc_to_text": "Question: {{question}}\nAnswer:",
108
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
109
+ "doc_to_choice": "{{choices.text}}",
110
+ "description": "",
111
+ "target_delimiter": " ",
112
+ "fewshot_delimiter": "\n\n",
113
+ "num_fewshot": 0,
114
+ "metric_list": [
115
+ {
116
+ "metric": "acc",
117
+ "aggregation": "mean",
118
+ "higher_is_better": true
119
+ },
120
+ {
121
+ "metric": "acc_norm",
122
+ "aggregation": "mean",
123
+ "higher_is_better": true
124
+ }
125
+ ],
126
+ "output_type": "multiple_choice",
127
+ "repeats": 1,
128
+ "should_decontaminate": true,
129
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
130
+ },
131
+ "boolq": {
132
+ "task": "boolq",
133
+ "group": [
134
+ "super-glue-lm-eval-v1"
135
+ ],
136
+ "dataset_path": "super_glue",
137
+ "dataset_name": "boolq",
138
+ "training_split": "train",
139
+ "validation_split": "validation",
140
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
141
+ "doc_to_target": "label",
142
+ "doc_to_choice": [
143
+ "no",
144
+ "yes"
145
+ ],
146
+ "description": "",
147
+ "target_delimiter": " ",
148
+ "fewshot_delimiter": "\n\n",
149
+ "num_fewshot": 0,
150
+ "metric_list": [
151
+ {
152
+ "metric": "acc"
153
+ }
154
+ ],
155
+ "output_type": "multiple_choice",
156
+ "repeats": 1,
157
+ "should_decontaminate": true,
158
+ "doc_to_decontamination_query": "passage"
159
+ },
160
+ "hellaswag": {
161
+ "task": "hellaswag",
162
+ "group": [
163
+ "multiple_choice"
164
+ ],
165
+ "dataset_path": "hellaswag",
166
+ "training_split": "train",
167
+ "validation_split": "validation",
168
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
169
+ "doc_to_target": "{{label}}",
170
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
171
+ "description": "",
172
+ "target_delimiter": " ",
173
+ "fewshot_delimiter": "\n\n",
174
+ "num_fewshot": 0,
175
+ "metric_list": [
176
+ {
177
+ "metric": "acc",
178
+ "aggregation": "mean",
179
+ "higher_is_better": true
180
+ },
181
+ {
182
+ "metric": "acc_norm",
183
+ "aggregation": "mean",
184
+ "higher_is_better": true
185
+ }
186
+ ],
187
+ "output_type": "multiple_choice",
188
+ "repeats": 1,
189
+ "should_decontaminate": false
190
+ },
191
+ "lambada_openai": {
192
+ "task": "lambada_openai",
193
+ "group": [
194
+ "lambada",
195
+ "loglikelihood",
196
+ "perplexity"
197
+ ],
198
+ "dataset_path": "EleutherAI/lambada_openai",
199
+ "dataset_name": "default",
200
+ "test_split": "test",
201
+ "template_aliases": "",
202
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
203
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
204
+ "description": "",
205
+ "target_delimiter": " ",
206
+ "fewshot_delimiter": "\n\n",
207
+ "num_fewshot": 0,
208
+ "metric_list": [
209
+ {
210
+ "metric": "perplexity",
211
+ "aggregation": "perplexity",
212
+ "higher_is_better": false
213
+ },
214
+ {
215
+ "metric": "acc",
216
+ "aggregation": "mean",
217
+ "higher_is_better": true
218
+ }
219
+ ],
220
+ "output_type": "loglikelihood",
221
+ "repeats": 1,
222
+ "should_decontaminate": true,
223
+ "doc_to_decontamination_query": "{{text}}"
224
+ },
225
+ "openbookqa": {
226
+ "task": "openbookqa",
227
+ "group": [
228
+ "multiple_choice"
229
+ ],
230
+ "dataset_path": "openbookqa",
231
+ "dataset_name": "main",
232
+ "training_split": "train",
233
+ "validation_split": "validation",
234
+ "test_split": "test",
235
+ "doc_to_text": "question_stem",
236
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
237
+ "doc_to_choice": "{{choices.text}}",
238
+ "description": "",
239
+ "target_delimiter": " ",
240
+ "fewshot_delimiter": "\n\n",
241
+ "num_fewshot": 0,
242
+ "metric_list": [
243
+ {
244
+ "metric": "acc",
245
+ "aggregation": "mean",
246
+ "higher_is_better": true
247
+ },
248
+ {
249
+ "metric": "acc_norm",
250
+ "aggregation": "mean",
251
+ "higher_is_better": true
252
+ }
253
+ ],
254
+ "output_type": "multiple_choice",
255
+ "repeats": 1,
256
+ "should_decontaminate": true,
257
+ "doc_to_decontamination_query": "question_stem"
258
+ },
259
+ "piqa": {
260
+ "task": "piqa",
261
+ "group": [
262
+ "multiple_choice"
263
+ ],
264
+ "dataset_path": "piqa",
265
+ "training_split": "train",
266
+ "validation_split": "validation",
267
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
268
+ "doc_to_target": "label",
269
+ "doc_to_choice": "{{[sol1, sol2]}}",
270
+ "description": "",
271
+ "target_delimiter": " ",
272
+ "fewshot_delimiter": "\n\n",
273
+ "num_fewshot": 0,
274
+ "metric_list": [
275
+ {
276
+ "metric": "acc",
277
+ "aggregation": "mean",
278
+ "higher_is_better": true
279
+ },
280
+ {
281
+ "metric": "acc_norm",
282
+ "aggregation": "mean",
283
+ "higher_is_better": true
284
+ }
285
+ ],
286
+ "output_type": "multiple_choice",
287
+ "repeats": 1,
288
+ "should_decontaminate": true,
289
+ "doc_to_decontamination_query": "goal"
290
+ },
291
+ "sciq": {
292
+ "task": "sciq",
293
+ "group": [
294
+ "multiple_choice"
295
+ ],
296
+ "dataset_path": "sciq",
297
+ "training_split": "train",
298
+ "validation_split": "validation",
299
+ "test_split": "test",
300
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
301
+ "doc_to_target": 3,
302
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
303
+ "description": "",
304
+ "target_delimiter": " ",
305
+ "fewshot_delimiter": "\n\n",
306
+ "num_fewshot": 0,
307
+ "metric_list": [
308
+ {
309
+ "metric": "acc",
310
+ "aggregation": "mean",
311
+ "higher_is_better": true
312
+ },
313
+ {
314
+ "metric": "acc_norm",
315
+ "aggregation": "mean",
316
+ "higher_is_better": true
317
+ }
318
+ ],
319
+ "output_type": "multiple_choice",
320
+ "repeats": 1,
321
+ "should_decontaminate": true,
322
+ "doc_to_decontamination_query": "{{support}} {{question}}"
323
+ },
324
+ "wikitext": {
325
+ "task": "wikitext",
326
+ "group": [
327
+ "perplexity",
328
+ "loglikelihood_rolling"
329
+ ],
330
+ "dataset_path": "EleutherAI/wikitext_document_level",
331
+ "dataset_name": "wikitext-2-raw-v1",
332
+ "training_split": "train",
333
+ "validation_split": "validation",
334
+ "test_split": "test",
335
+ "template_aliases": "",
336
+ "doc_to_text": "",
337
+ "doc_to_target": "<function wikitext_detokenizer at 0x7fd1a1c44040>",
338
+ "description": "",
339
+ "target_delimiter": " ",
340
+ "fewshot_delimiter": "\n\n",
341
+ "num_fewshot": 0,
342
+ "metric_list": [
343
+ {
344
+ "metric": "word_perplexity"
345
+ },
346
+ {
347
+ "metric": "byte_perplexity"
348
+ },
349
+ {
350
+ "metric": "bits_per_byte"
351
+ }
352
+ ],
353
+ "output_type": "loglikelihood_rolling",
354
+ "repeats": 1,
355
+ "should_decontaminate": true,
356
+ "doc_to_decontamination_query": "{{page}}"
357
+ },
358
+ "winogrande": {
359
+ "task": "winogrande",
360
+ "dataset_path": "winogrande",
361
+ "dataset_name": "winogrande_xl",
362
+ "training_split": "train",
363
+ "validation_split": "validation",
364
+ "doc_to_text": "<function doc_to_text at 0x7fd1a1c16ef0>",
365
+ "doc_to_target": "<function doc_to_target at 0x7fd1a1c17370>",
366
+ "doc_to_choice": "<function doc_to_choice at 0x7fd1a1c175b0>",
367
+ "description": "",
368
+ "target_delimiter": " ",
369
+ "fewshot_delimiter": "\n\n",
370
+ "num_fewshot": 0,
371
+ "metric_list": [
372
+ {
373
+ "metric": "acc",
374
+ "aggregation": "mean",
375
+ "higher_is_better": true
376
+ }
377
+ ],
378
+ "output_type": "multiple_choice",
379
+ "repeats": 1,
380
+ "should_decontaminate": false
381
+ }
382
+ },
383
+ "versions": {
384
+ "arc_challenge": "Yaml",
385
+ "arc_easy": "Yaml",
386
+ "boolq": "Yaml",
387
+ "hellaswag": "Yaml",
388
+ "lambada_openai": "Yaml",
389
+ "openbookqa": "Yaml",
390
+ "piqa": "Yaml",
391
+ "sciq": "Yaml",
392
+ "wikitext": "Yaml",
393
+ "winogrande": "Yaml"
394
+ },
395
+ "config": {
396
+ "model": "hf",
397
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-dpo",
398
+ "num_fewshot": 0,
399
+ "batch_size": 16,
400
+ "batch_sizes": [],
401
+ "device": "cuda:0",
402
+ "use_cache": null,
403
+ "limit": null,
404
+ "bootstrap_iters": 100000
405
+ },
406
+ "git_hash": "4e44f0a"
407
+ }
408
+ hf (pretrained=lomahony/eleuther-pythia160m-hh-dpo), limit: None, num_fewshot: 0, batch_size: 16
409
+ | Task |Version|Filter| Metric | Value | |Stderr|
410
+ |--------------|-------|------|---------------|------:|---|-----:|
411
+ |arc_challenge |Yaml |none |acc | 0.1928|± |0.0115|
412
+ | | |none |acc_norm | 0.2449|± |0.0126|
413
+ |arc_easy |Yaml |none |acc | 0.4672|± |0.0102|
414
+ | | |none |acc_norm | 0.4133|± |0.0101|
415
+ |boolq |Yaml |none |acc | 0.6150|± |0.0085|
416
+ |hellaswag |Yaml |none |acc | 0.2896|± |0.0045|
417
+ | | |none |acc_norm | 0.3016|± |0.0046|
418
+ |lambada_openai|Yaml |none |perplexity |40.4975|± |1.9471|
419
+ | | |none |acc | 0.3507|± |0.0066|
420
+ |openbookqa |Yaml |none |acc | 0.1720|± |0.0169|
421
+ | | |none |acc_norm | 0.2800|± |0.0201|
422
+ |piqa |Yaml |none |acc | 0.6333|± |0.0112|
423
+ | | |none |acc_norm | 0.6262|± |0.0113|
424
+ |sciq |Yaml |none |acc | 0.7530|± |0.0136|
425
+ | | |none |acc_norm | 0.6700|± |0.0149|
426
+ |wikitext |Yaml |none |word_perplexity|75.2499| | |
427
+ | | |none |byte_perplexity| 2.0387| | |
428
+ | | |none |bits_per_byte | 1.0276| | |
429
+ |winogrande |Yaml |none |acc | 0.5138|± |0.0140|
430
+
sft-160m-eval-files/dpo-pythia-160m-5shot-shelloutput.txt ADDED
@@ -0,0 +1,430 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ bootstrapping for stddev: perplexity
2
+ {
3
+ "results": {
4
+ "arc_challenge": {
5
+ "acc,none": 0.20563139931740615,
6
+ "acc_stderr,none": 0.011810745260742547,
7
+ "acc_norm,none": 0.24744027303754265,
8
+ "acc_norm_stderr,none": 0.01261035266329267
9
+ },
10
+ "arc_easy": {
11
+ "acc,none": 0.46380471380471383,
12
+ "acc_stderr,none": 0.01023286555034674,
13
+ "acc_norm,none": 0.4385521885521885,
14
+ "acc_norm_stderr,none": 0.010182010275471116
15
+ },
16
+ "boolq": {
17
+ "acc,none": 0.6055045871559633,
18
+ "acc_stderr,none": 0.008548152025770934
19
+ },
20
+ "hellaswag": {
21
+ "acc,none": 0.2885879306910974,
22
+ "acc_stderr,none": 0.004521798577922137,
23
+ "acc_norm,none": 0.3088030272854013,
24
+ "acc_norm_stderr,none": 0.004610554974411229
25
+ },
26
+ "lambada_openai": {
27
+ "perplexity,none": 68.78788187981594,
28
+ "perplexity_stderr,none": 3.3418985414978897,
29
+ "acc,none": 0.2815835435668543,
30
+ "acc_stderr,none": 0.006266194106395877
31
+ },
32
+ "openbookqa": {
33
+ "acc,none": 0.158,
34
+ "acc_stderr,none": 0.01632804980457984,
35
+ "acc_norm,none": 0.254,
36
+ "acc_norm_stderr,none": 0.019486596801643368
37
+ },
38
+ "piqa": {
39
+ "acc,none": 0.6284004352557128,
40
+ "acc_stderr,none": 0.011274603006724757,
41
+ "acc_norm,none": 0.6332970620239391,
42
+ "acc_norm_stderr,none": 0.01124362501903826
43
+ },
44
+ "sciq": {
45
+ "acc,none": 0.76,
46
+ "acc_stderr,none": 0.013512312258920836,
47
+ "acc_norm,none": 0.737,
48
+ "acc_norm_stderr,none": 0.013929286594259724
49
+ },
50
+ "wikitext": {
51
+ "word_perplexity,none": 75.24993841350984,
52
+ "byte_perplexity,none": 2.0386909133746456,
53
+ "bits_per_byte,none": 1.0276430644458452
54
+ },
55
+ "winogrande": {
56
+ "acc,none": 0.5138121546961326,
57
+ "acc_stderr,none": 0.014047122916440419
58
+ }
59
+ },
60
+ "configs": {
61
+ "arc_challenge": {
62
+ "task": "arc_challenge",
63
+ "group": [
64
+ "ai2_arc",
65
+ "multiple_choice"
66
+ ],
67
+ "dataset_path": "ai2_arc",
68
+ "dataset_name": "ARC-Challenge",
69
+ "training_split": "train",
70
+ "validation_split": "validation",
71
+ "test_split": "test",
72
+ "doc_to_text": "Question: {{question}}\nAnswer:",
73
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
74
+ "doc_to_choice": "{{choices.text}}",
75
+ "description": "",
76
+ "target_delimiter": " ",
77
+ "fewshot_delimiter": "\n\n",
78
+ "num_fewshot": 5,
79
+ "metric_list": [
80
+ {
81
+ "metric": "acc",
82
+ "aggregation": "mean",
83
+ "higher_is_better": true
84
+ },
85
+ {
86
+ "metric": "acc_norm",
87
+ "aggregation": "mean",
88
+ "higher_is_better": true
89
+ }
90
+ ],
91
+ "output_type": "multiple_choice",
92
+ "repeats": 1,
93
+ "should_decontaminate": true,
94
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
95
+ },
96
+ "arc_easy": {
97
+ "task": "arc_easy",
98
+ "group": [
99
+ "ai2_arc",
100
+ "multiple_choice"
101
+ ],
102
+ "dataset_path": "ai2_arc",
103
+ "dataset_name": "ARC-Easy",
104
+ "training_split": "train",
105
+ "validation_split": "validation",
106
+ "test_split": "test",
107
+ "doc_to_text": "Question: {{question}}\nAnswer:",
108
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
109
+ "doc_to_choice": "{{choices.text}}",
110
+ "description": "",
111
+ "target_delimiter": " ",
112
+ "fewshot_delimiter": "\n\n",
113
+ "num_fewshot": 5,
114
+ "metric_list": [
115
+ {
116
+ "metric": "acc",
117
+ "aggregation": "mean",
118
+ "higher_is_better": true
119
+ },
120
+ {
121
+ "metric": "acc_norm",
122
+ "aggregation": "mean",
123
+ "higher_is_better": true
124
+ }
125
+ ],
126
+ "output_type": "multiple_choice",
127
+ "repeats": 1,
128
+ "should_decontaminate": true,
129
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
130
+ },
131
+ "boolq": {
132
+ "task": "boolq",
133
+ "group": [
134
+ "super-glue-lm-eval-v1"
135
+ ],
136
+ "dataset_path": "super_glue",
137
+ "dataset_name": "boolq",
138
+ "training_split": "train",
139
+ "validation_split": "validation",
140
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
141
+ "doc_to_target": "label",
142
+ "doc_to_choice": [
143
+ "no",
144
+ "yes"
145
+ ],
146
+ "description": "",
147
+ "target_delimiter": " ",
148
+ "fewshot_delimiter": "\n\n",
149
+ "num_fewshot": 5,
150
+ "metric_list": [
151
+ {
152
+ "metric": "acc"
153
+ }
154
+ ],
155
+ "output_type": "multiple_choice",
156
+ "repeats": 1,
157
+ "should_decontaminate": true,
158
+ "doc_to_decontamination_query": "passage"
159
+ },
160
+ "hellaswag": {
161
+ "task": "hellaswag",
162
+ "group": [
163
+ "multiple_choice"
164
+ ],
165
+ "dataset_path": "hellaswag",
166
+ "training_split": "train",
167
+ "validation_split": "validation",
168
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
169
+ "doc_to_target": "{{label}}",
170
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
171
+ "description": "",
172
+ "target_delimiter": " ",
173
+ "fewshot_delimiter": "\n\n",
174
+ "num_fewshot": 5,
175
+ "metric_list": [
176
+ {
177
+ "metric": "acc",
178
+ "aggregation": "mean",
179
+ "higher_is_better": true
180
+ },
181
+ {
182
+ "metric": "acc_norm",
183
+ "aggregation": "mean",
184
+ "higher_is_better": true
185
+ }
186
+ ],
187
+ "output_type": "multiple_choice",
188
+ "repeats": 1,
189
+ "should_decontaminate": false
190
+ },
191
+ "lambada_openai": {
192
+ "task": "lambada_openai",
193
+ "group": [
194
+ "lambada",
195
+ "loglikelihood",
196
+ "perplexity"
197
+ ],
198
+ "dataset_path": "EleutherAI/lambada_openai",
199
+ "dataset_name": "default",
200
+ "test_split": "test",
201
+ "template_aliases": "",
202
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
203
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
204
+ "description": "",
205
+ "target_delimiter": " ",
206
+ "fewshot_delimiter": "\n\n",
207
+ "num_fewshot": 5,
208
+ "metric_list": [
209
+ {
210
+ "metric": "perplexity",
211
+ "aggregation": "perplexity",
212
+ "higher_is_better": false
213
+ },
214
+ {
215
+ "metric": "acc",
216
+ "aggregation": "mean",
217
+ "higher_is_better": true
218
+ }
219
+ ],
220
+ "output_type": "loglikelihood",
221
+ "repeats": 1,
222
+ "should_decontaminate": true,
223
+ "doc_to_decontamination_query": "{{text}}"
224
+ },
225
+ "openbookqa": {
226
+ "task": "openbookqa",
227
+ "group": [
228
+ "multiple_choice"
229
+ ],
230
+ "dataset_path": "openbookqa",
231
+ "dataset_name": "main",
232
+ "training_split": "train",
233
+ "validation_split": "validation",
234
+ "test_split": "test",
235
+ "doc_to_text": "question_stem",
236
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
237
+ "doc_to_choice": "{{choices.text}}",
238
+ "description": "",
239
+ "target_delimiter": " ",
240
+ "fewshot_delimiter": "\n\n",
241
+ "num_fewshot": 5,
242
+ "metric_list": [
243
+ {
244
+ "metric": "acc",
245
+ "aggregation": "mean",
246
+ "higher_is_better": true
247
+ },
248
+ {
249
+ "metric": "acc_norm",
250
+ "aggregation": "mean",
251
+ "higher_is_better": true
252
+ }
253
+ ],
254
+ "output_type": "multiple_choice",
255
+ "repeats": 1,
256
+ "should_decontaminate": true,
257
+ "doc_to_decontamination_query": "question_stem"
258
+ },
259
+ "piqa": {
260
+ "task": "piqa",
261
+ "group": [
262
+ "multiple_choice"
263
+ ],
264
+ "dataset_path": "piqa",
265
+ "training_split": "train",
266
+ "validation_split": "validation",
267
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
268
+ "doc_to_target": "label",
269
+ "doc_to_choice": "{{[sol1, sol2]}}",
270
+ "description": "",
271
+ "target_delimiter": " ",
272
+ "fewshot_delimiter": "\n\n",
273
+ "num_fewshot": 5,
274
+ "metric_list": [
275
+ {
276
+ "metric": "acc",
277
+ "aggregation": "mean",
278
+ "higher_is_better": true
279
+ },
280
+ {
281
+ "metric": "acc_norm",
282
+ "aggregation": "mean",
283
+ "higher_is_better": true
284
+ }
285
+ ],
286
+ "output_type": "multiple_choice",
287
+ "repeats": 1,
288
+ "should_decontaminate": true,
289
+ "doc_to_decontamination_query": "goal"
290
+ },
291
+ "sciq": {
292
+ "task": "sciq",
293
+ "group": [
294
+ "multiple_choice"
295
+ ],
296
+ "dataset_path": "sciq",
297
+ "training_split": "train",
298
+ "validation_split": "validation",
299
+ "test_split": "test",
300
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
301
+ "doc_to_target": 3,
302
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
303
+ "description": "",
304
+ "target_delimiter": " ",
305
+ "fewshot_delimiter": "\n\n",
306
+ "num_fewshot": 5,
307
+ "metric_list": [
308
+ {
309
+ "metric": "acc",
310
+ "aggregation": "mean",
311
+ "higher_is_better": true
312
+ },
313
+ {
314
+ "metric": "acc_norm",
315
+ "aggregation": "mean",
316
+ "higher_is_better": true
317
+ }
318
+ ],
319
+ "output_type": "multiple_choice",
320
+ "repeats": 1,
321
+ "should_decontaminate": true,
322
+ "doc_to_decontamination_query": "{{support}} {{question}}"
323
+ },
324
+ "wikitext": {
325
+ "task": "wikitext",
326
+ "group": [
327
+ "perplexity",
328
+ "loglikelihood_rolling"
329
+ ],
330
+ "dataset_path": "EleutherAI/wikitext_document_level",
331
+ "dataset_name": "wikitext-2-raw-v1",
332
+ "training_split": "train",
333
+ "validation_split": "validation",
334
+ "test_split": "test",
335
+ "template_aliases": "",
336
+ "doc_to_text": "",
337
+ "doc_to_target": "<function wikitext_detokenizer at 0x7ff6ce1e4040>",
338
+ "description": "",
339
+ "target_delimiter": " ",
340
+ "fewshot_delimiter": "\n\n",
341
+ "num_fewshot": 5,
342
+ "metric_list": [
343
+ {
344
+ "metric": "word_perplexity"
345
+ },
346
+ {
347
+ "metric": "byte_perplexity"
348
+ },
349
+ {
350
+ "metric": "bits_per_byte"
351
+ }
352
+ ],
353
+ "output_type": "loglikelihood_rolling",
354
+ "repeats": 1,
355
+ "should_decontaminate": true,
356
+ "doc_to_decontamination_query": "{{page}}"
357
+ },
358
+ "winogrande": {
359
+ "task": "winogrande",
360
+ "dataset_path": "winogrande",
361
+ "dataset_name": "winogrande_xl",
362
+ "training_split": "train",
363
+ "validation_split": "validation",
364
+ "doc_to_text": "<function doc_to_text at 0x7ff6ce1b6ef0>",
365
+ "doc_to_target": "<function doc_to_target at 0x7ff6ce1b7370>",
366
+ "doc_to_choice": "<function doc_to_choice at 0x7ff6ce1b75b0>",
367
+ "description": "",
368
+ "target_delimiter": " ",
369
+ "fewshot_delimiter": "\n\n",
370
+ "num_fewshot": 5,
371
+ "metric_list": [
372
+ {
373
+ "metric": "acc",
374
+ "aggregation": "mean",
375
+ "higher_is_better": true
376
+ }
377
+ ],
378
+ "output_type": "multiple_choice",
379
+ "repeats": 1,
380
+ "should_decontaminate": false
381
+ }
382
+ },
383
+ "versions": {
384
+ "arc_challenge": "Yaml",
385
+ "arc_easy": "Yaml",
386
+ "boolq": "Yaml",
387
+ "hellaswag": "Yaml",
388
+ "lambada_openai": "Yaml",
389
+ "openbookqa": "Yaml",
390
+ "piqa": "Yaml",
391
+ "sciq": "Yaml",
392
+ "wikitext": "Yaml",
393
+ "winogrande": "Yaml"
394
+ },
395
+ "config": {
396
+ "model": "hf",
397
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-dpo",
398
+ "num_fewshot": 5,
399
+ "batch_size": 16,
400
+ "batch_sizes": [],
401
+ "device": "cuda:0",
402
+ "use_cache": null,
403
+ "limit": null,
404
+ "bootstrap_iters": 100000
405
+ },
406
+ "git_hash": "4e44f0a"
407
+ }
408
+ hf (pretrained=lomahony/eleuther-pythia160m-hh-dpo), limit: None, num_fewshot: 5, batch_size: 16
409
+ | Task |Version|Filter| Metric | Value | |Stderr|
410
+ |--------------|-------|------|---------------|------:|---|-----:|
411
+ |arc_challenge |Yaml |none |acc | 0.2056|± |0.0118|
412
+ | | |none |acc_norm | 0.2474|± |0.0126|
413
+ |arc_easy |Yaml |none |acc | 0.4638|± |0.0102|
414
+ | | |none |acc_norm | 0.4386|± |0.0102|
415
+ |boolq |Yaml |none |acc | 0.6055|± |0.0085|
416
+ |hellaswag |Yaml |none |acc | 0.2886|± |0.0045|
417
+ | | |none |acc_norm | 0.3088|± |0.0046|
418
+ |lambada_openai|Yaml |none |perplexity |68.7879|± |3.3419|
419
+ | | |none |acc | 0.2816|± |0.0063|
420
+ |openbookqa |Yaml |none |acc | 0.1580|± |0.0163|
421
+ | | |none |acc_norm | 0.2540|± |0.0195|
422
+ |piqa |Yaml |none |acc | 0.6284|± |0.0113|
423
+ | | |none |acc_norm | 0.6333|± |0.0112|
424
+ |sciq |Yaml |none |acc | 0.7600|± |0.0135|
425
+ | | |none |acc_norm | 0.7370|± |0.0139|
426
+ |wikitext |Yaml |none |word_perplexity|75.2499| | |
427
+ | | |none |byte_perplexity| 2.0387| | |
428
+ | | |none |bits_per_byte | 1.0276| | |
429
+ |winogrande |Yaml |none |acc | 0.5138|± |0.0140|
430
+
sft-160m-eval-files/sft-pythia-160m-0shot-shelloutput.txt ADDED
@@ -0,0 +1,430 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ bootstrapping for stddev: perplexity
2
+ {
3
+ "results": {
4
+ "arc_challenge": {
5
+ "acc,none": 0.1945392491467577,
6
+ "acc_stderr,none": 0.011567709174648728,
7
+ "acc_norm,none": 0.24573378839590443,
8
+ "acc_norm_stderr,none": 0.012581033453730118
9
+ },
10
+ "arc_easy": {
11
+ "acc,none": 0.4524410774410774,
12
+ "acc_stderr,none": 0.010213265860171393,
13
+ "acc_norm,none": 0.4078282828282828,
14
+ "acc_norm_stderr,none": 0.010083950240041214
15
+ },
16
+ "boolq": {
17
+ "acc,none": 0.6192660550458715,
18
+ "acc_stderr,none": 0.00849262556165622
19
+ },
20
+ "hellaswag": {
21
+ "acc,none": 0.2869946225851424,
22
+ "acc_stderr,none": 0.00451434554778033,
23
+ "acc_norm,none": 0.29685321649073887,
24
+ "acc_norm_stderr,none": 0.0045593758358059315
25
+ },
26
+ "lambada_openai": {
27
+ "perplexity,none": 33.80704913877752,
28
+ "perplexity_stderr,none": 1.4967416616346076,
29
+ "acc,none": 0.3580438579468271,
30
+ "acc_stderr,none": 0.006679329465345719
31
+ },
32
+ "openbookqa": {
33
+ "acc,none": 0.17,
34
+ "acc_stderr,none": 0.01681563353139343,
35
+ "acc_norm,none": 0.26,
36
+ "acc_norm_stderr,none": 0.019635965529725515
37
+ },
38
+ "piqa": {
39
+ "acc,none": 0.6332970620239391,
40
+ "acc_stderr,none": 0.011243625019038255,
41
+ "acc_norm,none": 0.6300326441784548,
42
+ "acc_norm_stderr,none": 0.011264415223415276
43
+ },
44
+ "sciq": {
45
+ "acc,none": 0.756,
46
+ "acc_stderr,none": 0.013588548437881431,
47
+ "acc_norm,none": 0.687,
48
+ "acc_norm_stderr,none": 0.01467127282297788
49
+ },
50
+ "wikitext": {
51
+ "word_perplexity,none": 69.5803814818009,
52
+ "byte_perplexity,none": 2.012533549921474,
53
+ "bits_per_byte,none": 1.009012833861732
54
+ },
55
+ "winogrande": {
56
+ "acc,none": 0.516179952644041,
57
+ "acc_stderr,none": 0.0140451261309786
58
+ }
59
+ },
60
+ "configs": {
61
+ "arc_challenge": {
62
+ "task": "arc_challenge",
63
+ "group": [
64
+ "ai2_arc",
65
+ "multiple_choice"
66
+ ],
67
+ "dataset_path": "ai2_arc",
68
+ "dataset_name": "ARC-Challenge",
69
+ "training_split": "train",
70
+ "validation_split": "validation",
71
+ "test_split": "test",
72
+ "doc_to_text": "Question: {{question}}\nAnswer:",
73
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
74
+ "doc_to_choice": "{{choices.text}}",
75
+ "description": "",
76
+ "target_delimiter": " ",
77
+ "fewshot_delimiter": "\n\n",
78
+ "num_fewshot": 0,
79
+ "metric_list": [
80
+ {
81
+ "metric": "acc",
82
+ "aggregation": "mean",
83
+ "higher_is_better": true
84
+ },
85
+ {
86
+ "metric": "acc_norm",
87
+ "aggregation": "mean",
88
+ "higher_is_better": true
89
+ }
90
+ ],
91
+ "output_type": "multiple_choice",
92
+ "repeats": 1,
93
+ "should_decontaminate": true,
94
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
95
+ },
96
+ "arc_easy": {
97
+ "task": "arc_easy",
98
+ "group": [
99
+ "ai2_arc",
100
+ "multiple_choice"
101
+ ],
102
+ "dataset_path": "ai2_arc",
103
+ "dataset_name": "ARC-Easy",
104
+ "training_split": "train",
105
+ "validation_split": "validation",
106
+ "test_split": "test",
107
+ "doc_to_text": "Question: {{question}}\nAnswer:",
108
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
109
+ "doc_to_choice": "{{choices.text}}",
110
+ "description": "",
111
+ "target_delimiter": " ",
112
+ "fewshot_delimiter": "\n\n",
113
+ "num_fewshot": 0,
114
+ "metric_list": [
115
+ {
116
+ "metric": "acc",
117
+ "aggregation": "mean",
118
+ "higher_is_better": true
119
+ },
120
+ {
121
+ "metric": "acc_norm",
122
+ "aggregation": "mean",
123
+ "higher_is_better": true
124
+ }
125
+ ],
126
+ "output_type": "multiple_choice",
127
+ "repeats": 1,
128
+ "should_decontaminate": true,
129
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
130
+ },
131
+ "boolq": {
132
+ "task": "boolq",
133
+ "group": [
134
+ "super-glue-lm-eval-v1"
135
+ ],
136
+ "dataset_path": "super_glue",
137
+ "dataset_name": "boolq",
138
+ "training_split": "train",
139
+ "validation_split": "validation",
140
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
141
+ "doc_to_target": "label",
142
+ "doc_to_choice": [
143
+ "no",
144
+ "yes"
145
+ ],
146
+ "description": "",
147
+ "target_delimiter": " ",
148
+ "fewshot_delimiter": "\n\n",
149
+ "num_fewshot": 0,
150
+ "metric_list": [
151
+ {
152
+ "metric": "acc"
153
+ }
154
+ ],
155
+ "output_type": "multiple_choice",
156
+ "repeats": 1,
157
+ "should_decontaminate": true,
158
+ "doc_to_decontamination_query": "passage"
159
+ },
160
+ "hellaswag": {
161
+ "task": "hellaswag",
162
+ "group": [
163
+ "multiple_choice"
164
+ ],
165
+ "dataset_path": "hellaswag",
166
+ "training_split": "train",
167
+ "validation_split": "validation",
168
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
169
+ "doc_to_target": "{{label}}",
170
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
171
+ "description": "",
172
+ "target_delimiter": " ",
173
+ "fewshot_delimiter": "\n\n",
174
+ "num_fewshot": 0,
175
+ "metric_list": [
176
+ {
177
+ "metric": "acc",
178
+ "aggregation": "mean",
179
+ "higher_is_better": true
180
+ },
181
+ {
182
+ "metric": "acc_norm",
183
+ "aggregation": "mean",
184
+ "higher_is_better": true
185
+ }
186
+ ],
187
+ "output_type": "multiple_choice",
188
+ "repeats": 1,
189
+ "should_decontaminate": false
190
+ },
191
+ "lambada_openai": {
192
+ "task": "lambada_openai",
193
+ "group": [
194
+ "lambada",
195
+ "loglikelihood",
196
+ "perplexity"
197
+ ],
198
+ "dataset_path": "EleutherAI/lambada_openai",
199
+ "dataset_name": "default",
200
+ "test_split": "test",
201
+ "template_aliases": "",
202
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
203
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
204
+ "description": "",
205
+ "target_delimiter": " ",
206
+ "fewshot_delimiter": "\n\n",
207
+ "num_fewshot": 0,
208
+ "metric_list": [
209
+ {
210
+ "metric": "perplexity",
211
+ "aggregation": "perplexity",
212
+ "higher_is_better": false
213
+ },
214
+ {
215
+ "metric": "acc",
216
+ "aggregation": "mean",
217
+ "higher_is_better": true
218
+ }
219
+ ],
220
+ "output_type": "loglikelihood",
221
+ "repeats": 1,
222
+ "should_decontaminate": true,
223
+ "doc_to_decontamination_query": "{{text}}"
224
+ },
225
+ "openbookqa": {
226
+ "task": "openbookqa",
227
+ "group": [
228
+ "multiple_choice"
229
+ ],
230
+ "dataset_path": "openbookqa",
231
+ "dataset_name": "main",
232
+ "training_split": "train",
233
+ "validation_split": "validation",
234
+ "test_split": "test",
235
+ "doc_to_text": "question_stem",
236
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
237
+ "doc_to_choice": "{{choices.text}}",
238
+ "description": "",
239
+ "target_delimiter": " ",
240
+ "fewshot_delimiter": "\n\n",
241
+ "num_fewshot": 0,
242
+ "metric_list": [
243
+ {
244
+ "metric": "acc",
245
+ "aggregation": "mean",
246
+ "higher_is_better": true
247
+ },
248
+ {
249
+ "metric": "acc_norm",
250
+ "aggregation": "mean",
251
+ "higher_is_better": true
252
+ }
253
+ ],
254
+ "output_type": "multiple_choice",
255
+ "repeats": 1,
256
+ "should_decontaminate": true,
257
+ "doc_to_decontamination_query": "question_stem"
258
+ },
259
+ "piqa": {
260
+ "task": "piqa",
261
+ "group": [
262
+ "multiple_choice"
263
+ ],
264
+ "dataset_path": "piqa",
265
+ "training_split": "train",
266
+ "validation_split": "validation",
267
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
268
+ "doc_to_target": "label",
269
+ "doc_to_choice": "{{[sol1, sol2]}}",
270
+ "description": "",
271
+ "target_delimiter": " ",
272
+ "fewshot_delimiter": "\n\n",
273
+ "num_fewshot": 0,
274
+ "metric_list": [
275
+ {
276
+ "metric": "acc",
277
+ "aggregation": "mean",
278
+ "higher_is_better": true
279
+ },
280
+ {
281
+ "metric": "acc_norm",
282
+ "aggregation": "mean",
283
+ "higher_is_better": true
284
+ }
285
+ ],
286
+ "output_type": "multiple_choice",
287
+ "repeats": 1,
288
+ "should_decontaminate": true,
289
+ "doc_to_decontamination_query": "goal"
290
+ },
291
+ "sciq": {
292
+ "task": "sciq",
293
+ "group": [
294
+ "multiple_choice"
295
+ ],
296
+ "dataset_path": "sciq",
297
+ "training_split": "train",
298
+ "validation_split": "validation",
299
+ "test_split": "test",
300
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
301
+ "doc_to_target": 3,
302
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
303
+ "description": "",
304
+ "target_delimiter": " ",
305
+ "fewshot_delimiter": "\n\n",
306
+ "num_fewshot": 0,
307
+ "metric_list": [
308
+ {
309
+ "metric": "acc",
310
+ "aggregation": "mean",
311
+ "higher_is_better": true
312
+ },
313
+ {
314
+ "metric": "acc_norm",
315
+ "aggregation": "mean",
316
+ "higher_is_better": true
317
+ }
318
+ ],
319
+ "output_type": "multiple_choice",
320
+ "repeats": 1,
321
+ "should_decontaminate": true,
322
+ "doc_to_decontamination_query": "{{support}} {{question}}"
323
+ },
324
+ "wikitext": {
325
+ "task": "wikitext",
326
+ "group": [
327
+ "perplexity",
328
+ "loglikelihood_rolling"
329
+ ],
330
+ "dataset_path": "EleutherAI/wikitext_document_level",
331
+ "dataset_name": "wikitext-2-raw-v1",
332
+ "training_split": "train",
333
+ "validation_split": "validation",
334
+ "test_split": "test",
335
+ "template_aliases": "",
336
+ "doc_to_text": "",
337
+ "doc_to_target": "<function wikitext_detokenizer at 0x7f9804f4c040>",
338
+ "description": "",
339
+ "target_delimiter": " ",
340
+ "fewshot_delimiter": "\n\n",
341
+ "num_fewshot": 0,
342
+ "metric_list": [
343
+ {
344
+ "metric": "word_perplexity"
345
+ },
346
+ {
347
+ "metric": "byte_perplexity"
348
+ },
349
+ {
350
+ "metric": "bits_per_byte"
351
+ }
352
+ ],
353
+ "output_type": "loglikelihood_rolling",
354
+ "repeats": 1,
355
+ "should_decontaminate": true,
356
+ "doc_to_decontamination_query": "{{page}}"
357
+ },
358
+ "winogrande": {
359
+ "task": "winogrande",
360
+ "dataset_path": "winogrande",
361
+ "dataset_name": "winogrande_xl",
362
+ "training_split": "train",
363
+ "validation_split": "validation",
364
+ "doc_to_text": "<function doc_to_text at 0x7f9804f1eef0>",
365
+ "doc_to_target": "<function doc_to_target at 0x7f9804f1f370>",
366
+ "doc_to_choice": "<function doc_to_choice at 0x7f9804f1f5b0>",
367
+ "description": "",
368
+ "target_delimiter": " ",
369
+ "fewshot_delimiter": "\n\n",
370
+ "num_fewshot": 0,
371
+ "metric_list": [
372
+ {
373
+ "metric": "acc",
374
+ "aggregation": "mean",
375
+ "higher_is_better": true
376
+ }
377
+ ],
378
+ "output_type": "multiple_choice",
379
+ "repeats": 1,
380
+ "should_decontaminate": false
381
+ }
382
+ },
383
+ "versions": {
384
+ "arc_challenge": "Yaml",
385
+ "arc_easy": "Yaml",
386
+ "boolq": "Yaml",
387
+ "hellaswag": "Yaml",
388
+ "lambada_openai": "Yaml",
389
+ "openbookqa": "Yaml",
390
+ "piqa": "Yaml",
391
+ "sciq": "Yaml",
392
+ "wikitext": "Yaml",
393
+ "winogrande": "Yaml"
394
+ },
395
+ "config": {
396
+ "model": "hf",
397
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-sft",
398
+ "num_fewshot": 0,
399
+ "batch_size": 16,
400
+ "batch_sizes": [],
401
+ "device": "cuda:0",
402
+ "use_cache": null,
403
+ "limit": null,
404
+ "bootstrap_iters": 100000
405
+ },
406
+ "git_hash": "4e44f0a"
407
+ }
408
+ hf (pretrained=lomahony/eleuther-pythia160m-hh-sft), limit: None, num_fewshot: 0, batch_size: 16
409
+ | Task |Version|Filter| Metric | Value | |Stderr|
410
+ |--------------|-------|------|---------------|------:|---|-----:|
411
+ |arc_challenge |Yaml |none |acc | 0.1945|± |0.0116|
412
+ | | |none |acc_norm | 0.2457|± |0.0126|
413
+ |arc_easy |Yaml |none |acc | 0.4524|± |0.0102|
414
+ | | |none |acc_norm | 0.4078|± |0.0101|
415
+ |boolq |Yaml |none |acc | 0.6193|± |0.0085|
416
+ |hellaswag |Yaml |none |acc | 0.2870|± |0.0045|
417
+ | | |none |acc_norm | 0.2969|± |0.0046|
418
+ |lambada_openai|Yaml |none |perplexity |33.8070|± |1.4967|
419
+ | | |none |acc | 0.3580|± |0.0067|
420
+ |openbookqa |Yaml |none |acc | 0.1700|± |0.0168|
421
+ | | |none |acc_norm | 0.2600|± |0.0196|
422
+ |piqa |Yaml |none |acc | 0.6333|± |0.0112|
423
+ | | |none |acc_norm | 0.6300|± |0.0113|
424
+ |sciq |Yaml |none |acc | 0.7560|± |0.0136|
425
+ | | |none |acc_norm | 0.6870|± |0.0147|
426
+ |wikitext |Yaml |none |word_perplexity|69.5804| | |
427
+ | | |none |byte_perplexity| 2.0125| | |
428
+ | | |none |bits_per_byte | 1.0090| | |
429
+ |winogrande |Yaml |none |acc | 0.5162|± |0.0140|
430
+
sft-160m-eval-files/sft-pythia-160m-0shot/results.json ADDED
@@ -0,0 +1,406 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "results": {
3
+ "arc_challenge": {
4
+ "acc,none": 0.1945392491467577,
5
+ "acc_stderr,none": 0.011567709174648728,
6
+ "acc_norm,none": 0.24573378839590443,
7
+ "acc_norm_stderr,none": 0.012581033453730118
8
+ },
9
+ "arc_easy": {
10
+ "acc,none": 0.4524410774410774,
11
+ "acc_stderr,none": 0.010213265860171393,
12
+ "acc_norm,none": 0.4078282828282828,
13
+ "acc_norm_stderr,none": 0.010083950240041214
14
+ },
15
+ "boolq": {
16
+ "acc,none": 0.6192660550458715,
17
+ "acc_stderr,none": 0.00849262556165622
18
+ },
19
+ "hellaswag": {
20
+ "acc,none": 0.2869946225851424,
21
+ "acc_stderr,none": 0.00451434554778033,
22
+ "acc_norm,none": 0.29685321649073887,
23
+ "acc_norm_stderr,none": 0.0045593758358059315
24
+ },
25
+ "lambada_openai": {
26
+ "perplexity,none": 33.80704913877752,
27
+ "perplexity_stderr,none": 1.4967416616346076,
28
+ "acc,none": 0.3580438579468271,
29
+ "acc_stderr,none": 0.006679329465345719
30
+ },
31
+ "openbookqa": {
32
+ "acc,none": 0.17,
33
+ "acc_stderr,none": 0.01681563353139343,
34
+ "acc_norm,none": 0.26,
35
+ "acc_norm_stderr,none": 0.019635965529725515
36
+ },
37
+ "piqa": {
38
+ "acc,none": 0.6332970620239391,
39
+ "acc_stderr,none": 0.011243625019038255,
40
+ "acc_norm,none": 0.6300326441784548,
41
+ "acc_norm_stderr,none": 0.011264415223415276
42
+ },
43
+ "sciq": {
44
+ "acc,none": 0.756,
45
+ "acc_stderr,none": 0.013588548437881431,
46
+ "acc_norm,none": 0.687,
47
+ "acc_norm_stderr,none": 0.01467127282297788
48
+ },
49
+ "wikitext": {
50
+ "word_perplexity,none": 69.5803814818009,
51
+ "byte_perplexity,none": 2.012533549921474,
52
+ "bits_per_byte,none": 1.009012833861732
53
+ },
54
+ "winogrande": {
55
+ "acc,none": 0.516179952644041,
56
+ "acc_stderr,none": 0.0140451261309786
57
+ }
58
+ },
59
+ "configs": {
60
+ "arc_challenge": {
61
+ "task": "arc_challenge",
62
+ "group": [
63
+ "ai2_arc",
64
+ "multiple_choice"
65
+ ],
66
+ "dataset_path": "ai2_arc",
67
+ "dataset_name": "ARC-Challenge",
68
+ "training_split": "train",
69
+ "validation_split": "validation",
70
+ "test_split": "test",
71
+ "doc_to_text": "Question: {{question}}\nAnswer:",
72
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
73
+ "doc_to_choice": "{{choices.text}}",
74
+ "description": "",
75
+ "target_delimiter": " ",
76
+ "fewshot_delimiter": "\n\n",
77
+ "num_fewshot": 0,
78
+ "metric_list": [
79
+ {
80
+ "metric": "acc",
81
+ "aggregation": "mean",
82
+ "higher_is_better": true
83
+ },
84
+ {
85
+ "metric": "acc_norm",
86
+ "aggregation": "mean",
87
+ "higher_is_better": true
88
+ }
89
+ ],
90
+ "output_type": "multiple_choice",
91
+ "repeats": 1,
92
+ "should_decontaminate": true,
93
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
94
+ },
95
+ "arc_easy": {
96
+ "task": "arc_easy",
97
+ "group": [
98
+ "ai2_arc",
99
+ "multiple_choice"
100
+ ],
101
+ "dataset_path": "ai2_arc",
102
+ "dataset_name": "ARC-Easy",
103
+ "training_split": "train",
104
+ "validation_split": "validation",
105
+ "test_split": "test",
106
+ "doc_to_text": "Question: {{question}}\nAnswer:",
107
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
108
+ "doc_to_choice": "{{choices.text}}",
109
+ "description": "",
110
+ "target_delimiter": " ",
111
+ "fewshot_delimiter": "\n\n",
112
+ "num_fewshot": 0,
113
+ "metric_list": [
114
+ {
115
+ "metric": "acc",
116
+ "aggregation": "mean",
117
+ "higher_is_better": true
118
+ },
119
+ {
120
+ "metric": "acc_norm",
121
+ "aggregation": "mean",
122
+ "higher_is_better": true
123
+ }
124
+ ],
125
+ "output_type": "multiple_choice",
126
+ "repeats": 1,
127
+ "should_decontaminate": true,
128
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
129
+ },
130
+ "boolq": {
131
+ "task": "boolq",
132
+ "group": [
133
+ "super-glue-lm-eval-v1"
134
+ ],
135
+ "dataset_path": "super_glue",
136
+ "dataset_name": "boolq",
137
+ "training_split": "train",
138
+ "validation_split": "validation",
139
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
140
+ "doc_to_target": "label",
141
+ "doc_to_choice": [
142
+ "no",
143
+ "yes"
144
+ ],
145
+ "description": "",
146
+ "target_delimiter": " ",
147
+ "fewshot_delimiter": "\n\n",
148
+ "num_fewshot": 0,
149
+ "metric_list": [
150
+ {
151
+ "metric": "acc"
152
+ }
153
+ ],
154
+ "output_type": "multiple_choice",
155
+ "repeats": 1,
156
+ "should_decontaminate": true,
157
+ "doc_to_decontamination_query": "passage"
158
+ },
159
+ "hellaswag": {
160
+ "task": "hellaswag",
161
+ "group": [
162
+ "multiple_choice"
163
+ ],
164
+ "dataset_path": "hellaswag",
165
+ "training_split": "train",
166
+ "validation_split": "validation",
167
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
168
+ "doc_to_target": "{{label}}",
169
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
170
+ "description": "",
171
+ "target_delimiter": " ",
172
+ "fewshot_delimiter": "\n\n",
173
+ "num_fewshot": 0,
174
+ "metric_list": [
175
+ {
176
+ "metric": "acc",
177
+ "aggregation": "mean",
178
+ "higher_is_better": true
179
+ },
180
+ {
181
+ "metric": "acc_norm",
182
+ "aggregation": "mean",
183
+ "higher_is_better": true
184
+ }
185
+ ],
186
+ "output_type": "multiple_choice",
187
+ "repeats": 1,
188
+ "should_decontaminate": false
189
+ },
190
+ "lambada_openai": {
191
+ "task": "lambada_openai",
192
+ "group": [
193
+ "lambada",
194
+ "loglikelihood",
195
+ "perplexity"
196
+ ],
197
+ "dataset_path": "EleutherAI/lambada_openai",
198
+ "dataset_name": "default",
199
+ "test_split": "test",
200
+ "template_aliases": "",
201
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
202
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
203
+ "description": "",
204
+ "target_delimiter": " ",
205
+ "fewshot_delimiter": "\n\n",
206
+ "num_fewshot": 0,
207
+ "metric_list": [
208
+ {
209
+ "metric": "perplexity",
210
+ "aggregation": "perplexity",
211
+ "higher_is_better": false
212
+ },
213
+ {
214
+ "metric": "acc",
215
+ "aggregation": "mean",
216
+ "higher_is_better": true
217
+ }
218
+ ],
219
+ "output_type": "loglikelihood",
220
+ "repeats": 1,
221
+ "should_decontaminate": true,
222
+ "doc_to_decontamination_query": "{{text}}"
223
+ },
224
+ "openbookqa": {
225
+ "task": "openbookqa",
226
+ "group": [
227
+ "multiple_choice"
228
+ ],
229
+ "dataset_path": "openbookqa",
230
+ "dataset_name": "main",
231
+ "training_split": "train",
232
+ "validation_split": "validation",
233
+ "test_split": "test",
234
+ "doc_to_text": "question_stem",
235
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
236
+ "doc_to_choice": "{{choices.text}}",
237
+ "description": "",
238
+ "target_delimiter": " ",
239
+ "fewshot_delimiter": "\n\n",
240
+ "num_fewshot": 0,
241
+ "metric_list": [
242
+ {
243
+ "metric": "acc",
244
+ "aggregation": "mean",
245
+ "higher_is_better": true
246
+ },
247
+ {
248
+ "metric": "acc_norm",
249
+ "aggregation": "mean",
250
+ "higher_is_better": true
251
+ }
252
+ ],
253
+ "output_type": "multiple_choice",
254
+ "repeats": 1,
255
+ "should_decontaminate": true,
256
+ "doc_to_decontamination_query": "question_stem"
257
+ },
258
+ "piqa": {
259
+ "task": "piqa",
260
+ "group": [
261
+ "multiple_choice"
262
+ ],
263
+ "dataset_path": "piqa",
264
+ "training_split": "train",
265
+ "validation_split": "validation",
266
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
267
+ "doc_to_target": "label",
268
+ "doc_to_choice": "{{[sol1, sol2]}}",
269
+ "description": "",
270
+ "target_delimiter": " ",
271
+ "fewshot_delimiter": "\n\n",
272
+ "num_fewshot": 0,
273
+ "metric_list": [
274
+ {
275
+ "metric": "acc",
276
+ "aggregation": "mean",
277
+ "higher_is_better": true
278
+ },
279
+ {
280
+ "metric": "acc_norm",
281
+ "aggregation": "mean",
282
+ "higher_is_better": true
283
+ }
284
+ ],
285
+ "output_type": "multiple_choice",
286
+ "repeats": 1,
287
+ "should_decontaminate": true,
288
+ "doc_to_decontamination_query": "goal"
289
+ },
290
+ "sciq": {
291
+ "task": "sciq",
292
+ "group": [
293
+ "multiple_choice"
294
+ ],
295
+ "dataset_path": "sciq",
296
+ "training_split": "train",
297
+ "validation_split": "validation",
298
+ "test_split": "test",
299
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
300
+ "doc_to_target": 3,
301
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
302
+ "description": "",
303
+ "target_delimiter": " ",
304
+ "fewshot_delimiter": "\n\n",
305
+ "num_fewshot": 0,
306
+ "metric_list": [
307
+ {
308
+ "metric": "acc",
309
+ "aggregation": "mean",
310
+ "higher_is_better": true
311
+ },
312
+ {
313
+ "metric": "acc_norm",
314
+ "aggregation": "mean",
315
+ "higher_is_better": true
316
+ }
317
+ ],
318
+ "output_type": "multiple_choice",
319
+ "repeats": 1,
320
+ "should_decontaminate": true,
321
+ "doc_to_decontamination_query": "{{support}} {{question}}"
322
+ },
323
+ "wikitext": {
324
+ "task": "wikitext",
325
+ "group": [
326
+ "perplexity",
327
+ "loglikelihood_rolling"
328
+ ],
329
+ "dataset_path": "EleutherAI/wikitext_document_level",
330
+ "dataset_name": "wikitext-2-raw-v1",
331
+ "training_split": "train",
332
+ "validation_split": "validation",
333
+ "test_split": "test",
334
+ "template_aliases": "",
335
+ "doc_to_text": "",
336
+ "doc_to_target": "<function wikitext_detokenizer at 0x7f9804f4c040>",
337
+ "description": "",
338
+ "target_delimiter": " ",
339
+ "fewshot_delimiter": "\n\n",
340
+ "num_fewshot": 0,
341
+ "metric_list": [
342
+ {
343
+ "metric": "word_perplexity"
344
+ },
345
+ {
346
+ "metric": "byte_perplexity"
347
+ },
348
+ {
349
+ "metric": "bits_per_byte"
350
+ }
351
+ ],
352
+ "output_type": "loglikelihood_rolling",
353
+ "repeats": 1,
354
+ "should_decontaminate": true,
355
+ "doc_to_decontamination_query": "{{page}}"
356
+ },
357
+ "winogrande": {
358
+ "task": "winogrande",
359
+ "dataset_path": "winogrande",
360
+ "dataset_name": "winogrande_xl",
361
+ "training_split": "train",
362
+ "validation_split": "validation",
363
+ "doc_to_text": "<function doc_to_text at 0x7f9804f1eef0>",
364
+ "doc_to_target": "<function doc_to_target at 0x7f9804f1f370>",
365
+ "doc_to_choice": "<function doc_to_choice at 0x7f9804f1f5b0>",
366
+ "description": "",
367
+ "target_delimiter": " ",
368
+ "fewshot_delimiter": "\n\n",
369
+ "num_fewshot": 0,
370
+ "metric_list": [
371
+ {
372
+ "metric": "acc",
373
+ "aggregation": "mean",
374
+ "higher_is_better": true
375
+ }
376
+ ],
377
+ "output_type": "multiple_choice",
378
+ "repeats": 1,
379
+ "should_decontaminate": false
380
+ }
381
+ },
382
+ "versions": {
383
+ "arc_challenge": "Yaml",
384
+ "arc_easy": "Yaml",
385
+ "boolq": "Yaml",
386
+ "hellaswag": "Yaml",
387
+ "lambada_openai": "Yaml",
388
+ "openbookqa": "Yaml",
389
+ "piqa": "Yaml",
390
+ "sciq": "Yaml",
391
+ "wikitext": "Yaml",
392
+ "winogrande": "Yaml"
393
+ },
394
+ "config": {
395
+ "model": "hf",
396
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-sft",
397
+ "num_fewshot": 0,
398
+ "batch_size": 16,
399
+ "batch_sizes": [],
400
+ "device": "cuda:0",
401
+ "use_cache": null,
402
+ "limit": null,
403
+ "bootstrap_iters": 100000
404
+ },
405
+ "git_hash": "4e44f0a"
406
+ }
sft-160m-eval-files/sft-pythia-160m-5shot-shelloutput.txt ADDED
@@ -0,0 +1,430 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ bootstrapping for stddev: perplexity
2
+ {
3
+ "results": {
4
+ "arc_challenge": {
5
+ "acc,none": 0.18686006825938567,
6
+ "acc_stderr,none": 0.011391015649694396,
7
+ "acc_norm,none": 0.24146757679180889,
8
+ "acc_norm_stderr,none": 0.012506564839739432
9
+ },
10
+ "arc_easy": {
11
+ "acc,none": 0.45707070707070707,
12
+ "acc_stderr,none": 0.010221897564256042,
13
+ "acc_norm,none": 0.4335016835016835,
14
+ "acc_norm_stderr,none": 0.010168640625454107
15
+ },
16
+ "boolq": {
17
+ "acc,none": 0.5938837920489297,
18
+ "acc_stderr,none": 0.008589510943787407
19
+ },
20
+ "hellaswag": {
21
+ "acc,none": 0.2861979685321649,
22
+ "acc_stderr,none": 0.004510593395289897,
23
+ "acc_norm,none": 0.3037243576976698,
24
+ "acc_norm_stderr,none": 0.00458925224362782
25
+ },
26
+ "lambada_openai": {
27
+ "perplexity,none": 60.10678259260095,
28
+ "perplexity_stderr,none": 2.70306190425745,
29
+ "acc,none": 0.2829419755482243,
30
+ "acc_stderr,none": 0.006275349431343604
31
+ },
32
+ "openbookqa": {
33
+ "acc,none": 0.164,
34
+ "acc_stderr,none": 0.016575811142446713,
35
+ "acc_norm,none": 0.256,
36
+ "acc_norm_stderr,none": 0.019536923574747598
37
+ },
38
+ "piqa": {
39
+ "acc,none": 0.6349292709466812,
40
+ "acc_stderr,none": 0.011233021830554834,
41
+ "acc_norm,none": 0.6305767138193689,
42
+ "acc_norm_stderr,none": 0.011260988628572333
43
+ },
44
+ "sciq": {
45
+ "acc,none": 0.77,
46
+ "acc_stderr,none": 0.013314551335935952,
47
+ "acc_norm,none": 0.747,
48
+ "acc_norm_stderr,none": 0.01375427861358708
49
+ },
50
+ "wikitext": {
51
+ "word_perplexity,none": 69.5803814818009,
52
+ "byte_perplexity,none": 2.012533549921474,
53
+ "bits_per_byte,none": 1.009012833861732
54
+ },
55
+ "winogrande": {
56
+ "acc,none": 0.516179952644041,
57
+ "acc_stderr,none": 0.0140451261309786
58
+ }
59
+ },
60
+ "configs": {
61
+ "arc_challenge": {
62
+ "task": "arc_challenge",
63
+ "group": [
64
+ "ai2_arc",
65
+ "multiple_choice"
66
+ ],
67
+ "dataset_path": "ai2_arc",
68
+ "dataset_name": "ARC-Challenge",
69
+ "training_split": "train",
70
+ "validation_split": "validation",
71
+ "test_split": "test",
72
+ "doc_to_text": "Question: {{question}}\nAnswer:",
73
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
74
+ "doc_to_choice": "{{choices.text}}",
75
+ "description": "",
76
+ "target_delimiter": " ",
77
+ "fewshot_delimiter": "\n\n",
78
+ "num_fewshot": 5,
79
+ "metric_list": [
80
+ {
81
+ "metric": "acc",
82
+ "aggregation": "mean",
83
+ "higher_is_better": true
84
+ },
85
+ {
86
+ "metric": "acc_norm",
87
+ "aggregation": "mean",
88
+ "higher_is_better": true
89
+ }
90
+ ],
91
+ "output_type": "multiple_choice",
92
+ "repeats": 1,
93
+ "should_decontaminate": true,
94
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
95
+ },
96
+ "arc_easy": {
97
+ "task": "arc_easy",
98
+ "group": [
99
+ "ai2_arc",
100
+ "multiple_choice"
101
+ ],
102
+ "dataset_path": "ai2_arc",
103
+ "dataset_name": "ARC-Easy",
104
+ "training_split": "train",
105
+ "validation_split": "validation",
106
+ "test_split": "test",
107
+ "doc_to_text": "Question: {{question}}\nAnswer:",
108
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
109
+ "doc_to_choice": "{{choices.text}}",
110
+ "description": "",
111
+ "target_delimiter": " ",
112
+ "fewshot_delimiter": "\n\n",
113
+ "num_fewshot": 5,
114
+ "metric_list": [
115
+ {
116
+ "metric": "acc",
117
+ "aggregation": "mean",
118
+ "higher_is_better": true
119
+ },
120
+ {
121
+ "metric": "acc_norm",
122
+ "aggregation": "mean",
123
+ "higher_is_better": true
124
+ }
125
+ ],
126
+ "output_type": "multiple_choice",
127
+ "repeats": 1,
128
+ "should_decontaminate": true,
129
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
130
+ },
131
+ "boolq": {
132
+ "task": "boolq",
133
+ "group": [
134
+ "super-glue-lm-eval-v1"
135
+ ],
136
+ "dataset_path": "super_glue",
137
+ "dataset_name": "boolq",
138
+ "training_split": "train",
139
+ "validation_split": "validation",
140
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
141
+ "doc_to_target": "label",
142
+ "doc_to_choice": [
143
+ "no",
144
+ "yes"
145
+ ],
146
+ "description": "",
147
+ "target_delimiter": " ",
148
+ "fewshot_delimiter": "\n\n",
149
+ "num_fewshot": 5,
150
+ "metric_list": [
151
+ {
152
+ "metric": "acc"
153
+ }
154
+ ],
155
+ "output_type": "multiple_choice",
156
+ "repeats": 1,
157
+ "should_decontaminate": true,
158
+ "doc_to_decontamination_query": "passage"
159
+ },
160
+ "hellaswag": {
161
+ "task": "hellaswag",
162
+ "group": [
163
+ "multiple_choice"
164
+ ],
165
+ "dataset_path": "hellaswag",
166
+ "training_split": "train",
167
+ "validation_split": "validation",
168
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
169
+ "doc_to_target": "{{label}}",
170
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
171
+ "description": "",
172
+ "target_delimiter": " ",
173
+ "fewshot_delimiter": "\n\n",
174
+ "num_fewshot": 5,
175
+ "metric_list": [
176
+ {
177
+ "metric": "acc",
178
+ "aggregation": "mean",
179
+ "higher_is_better": true
180
+ },
181
+ {
182
+ "metric": "acc_norm",
183
+ "aggregation": "mean",
184
+ "higher_is_better": true
185
+ }
186
+ ],
187
+ "output_type": "multiple_choice",
188
+ "repeats": 1,
189
+ "should_decontaminate": false
190
+ },
191
+ "lambada_openai": {
192
+ "task": "lambada_openai",
193
+ "group": [
194
+ "lambada",
195
+ "loglikelihood",
196
+ "perplexity"
197
+ ],
198
+ "dataset_path": "EleutherAI/lambada_openai",
199
+ "dataset_name": "default",
200
+ "test_split": "test",
201
+ "template_aliases": "",
202
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
203
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
204
+ "description": "",
205
+ "target_delimiter": " ",
206
+ "fewshot_delimiter": "\n\n",
207
+ "num_fewshot": 5,
208
+ "metric_list": [
209
+ {
210
+ "metric": "perplexity",
211
+ "aggregation": "perplexity",
212
+ "higher_is_better": false
213
+ },
214
+ {
215
+ "metric": "acc",
216
+ "aggregation": "mean",
217
+ "higher_is_better": true
218
+ }
219
+ ],
220
+ "output_type": "loglikelihood",
221
+ "repeats": 1,
222
+ "should_decontaminate": true,
223
+ "doc_to_decontamination_query": "{{text}}"
224
+ },
225
+ "openbookqa": {
226
+ "task": "openbookqa",
227
+ "group": [
228
+ "multiple_choice"
229
+ ],
230
+ "dataset_path": "openbookqa",
231
+ "dataset_name": "main",
232
+ "training_split": "train",
233
+ "validation_split": "validation",
234
+ "test_split": "test",
235
+ "doc_to_text": "question_stem",
236
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
237
+ "doc_to_choice": "{{choices.text}}",
238
+ "description": "",
239
+ "target_delimiter": " ",
240
+ "fewshot_delimiter": "\n\n",
241
+ "num_fewshot": 5,
242
+ "metric_list": [
243
+ {
244
+ "metric": "acc",
245
+ "aggregation": "mean",
246
+ "higher_is_better": true
247
+ },
248
+ {
249
+ "metric": "acc_norm",
250
+ "aggregation": "mean",
251
+ "higher_is_better": true
252
+ }
253
+ ],
254
+ "output_type": "multiple_choice",
255
+ "repeats": 1,
256
+ "should_decontaminate": true,
257
+ "doc_to_decontamination_query": "question_stem"
258
+ },
259
+ "piqa": {
260
+ "task": "piqa",
261
+ "group": [
262
+ "multiple_choice"
263
+ ],
264
+ "dataset_path": "piqa",
265
+ "training_split": "train",
266
+ "validation_split": "validation",
267
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
268
+ "doc_to_target": "label",
269
+ "doc_to_choice": "{{[sol1, sol2]}}",
270
+ "description": "",
271
+ "target_delimiter": " ",
272
+ "fewshot_delimiter": "\n\n",
273
+ "num_fewshot": 5,
274
+ "metric_list": [
275
+ {
276
+ "metric": "acc",
277
+ "aggregation": "mean",
278
+ "higher_is_better": true
279
+ },
280
+ {
281
+ "metric": "acc_norm",
282
+ "aggregation": "mean",
283
+ "higher_is_better": true
284
+ }
285
+ ],
286
+ "output_type": "multiple_choice",
287
+ "repeats": 1,
288
+ "should_decontaminate": true,
289
+ "doc_to_decontamination_query": "goal"
290
+ },
291
+ "sciq": {
292
+ "task": "sciq",
293
+ "group": [
294
+ "multiple_choice"
295
+ ],
296
+ "dataset_path": "sciq",
297
+ "training_split": "train",
298
+ "validation_split": "validation",
299
+ "test_split": "test",
300
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
301
+ "doc_to_target": 3,
302
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
303
+ "description": "",
304
+ "target_delimiter": " ",
305
+ "fewshot_delimiter": "\n\n",
306
+ "num_fewshot": 5,
307
+ "metric_list": [
308
+ {
309
+ "metric": "acc",
310
+ "aggregation": "mean",
311
+ "higher_is_better": true
312
+ },
313
+ {
314
+ "metric": "acc_norm",
315
+ "aggregation": "mean",
316
+ "higher_is_better": true
317
+ }
318
+ ],
319
+ "output_type": "multiple_choice",
320
+ "repeats": 1,
321
+ "should_decontaminate": true,
322
+ "doc_to_decontamination_query": "{{support}} {{question}}"
323
+ },
324
+ "wikitext": {
325
+ "task": "wikitext",
326
+ "group": [
327
+ "perplexity",
328
+ "loglikelihood_rolling"
329
+ ],
330
+ "dataset_path": "EleutherAI/wikitext_document_level",
331
+ "dataset_name": "wikitext-2-raw-v1",
332
+ "training_split": "train",
333
+ "validation_split": "validation",
334
+ "test_split": "test",
335
+ "template_aliases": "",
336
+ "doc_to_text": "",
337
+ "doc_to_target": "<function wikitext_detokenizer at 0x7f50b9b54040>",
338
+ "description": "",
339
+ "target_delimiter": " ",
340
+ "fewshot_delimiter": "\n\n",
341
+ "num_fewshot": 5,
342
+ "metric_list": [
343
+ {
344
+ "metric": "word_perplexity"
345
+ },
346
+ {
347
+ "metric": "byte_perplexity"
348
+ },
349
+ {
350
+ "metric": "bits_per_byte"
351
+ }
352
+ ],
353
+ "output_type": "loglikelihood_rolling",
354
+ "repeats": 1,
355
+ "should_decontaminate": true,
356
+ "doc_to_decontamination_query": "{{page}}"
357
+ },
358
+ "winogrande": {
359
+ "task": "winogrande",
360
+ "dataset_path": "winogrande",
361
+ "dataset_name": "winogrande_xl",
362
+ "training_split": "train",
363
+ "validation_split": "validation",
364
+ "doc_to_text": "<function doc_to_text at 0x7f50b9b26ef0>",
365
+ "doc_to_target": "<function doc_to_target at 0x7f50b9b27370>",
366
+ "doc_to_choice": "<function doc_to_choice at 0x7f50b9b275b0>",
367
+ "description": "",
368
+ "target_delimiter": " ",
369
+ "fewshot_delimiter": "\n\n",
370
+ "num_fewshot": 5,
371
+ "metric_list": [
372
+ {
373
+ "metric": "acc",
374
+ "aggregation": "mean",
375
+ "higher_is_better": true
376
+ }
377
+ ],
378
+ "output_type": "multiple_choice",
379
+ "repeats": 1,
380
+ "should_decontaminate": false
381
+ }
382
+ },
383
+ "versions": {
384
+ "arc_challenge": "Yaml",
385
+ "arc_easy": "Yaml",
386
+ "boolq": "Yaml",
387
+ "hellaswag": "Yaml",
388
+ "lambada_openai": "Yaml",
389
+ "openbookqa": "Yaml",
390
+ "piqa": "Yaml",
391
+ "sciq": "Yaml",
392
+ "wikitext": "Yaml",
393
+ "winogrande": "Yaml"
394
+ },
395
+ "config": {
396
+ "model": "hf",
397
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-sft",
398
+ "num_fewshot": 5,
399
+ "batch_size": 16,
400
+ "batch_sizes": [],
401
+ "device": "cuda:0",
402
+ "use_cache": null,
403
+ "limit": null,
404
+ "bootstrap_iters": 100000
405
+ },
406
+ "git_hash": "4e44f0a"
407
+ }
408
+ hf (pretrained=lomahony/eleuther-pythia160m-hh-sft), limit: None, num_fewshot: 5, batch_size: 16
409
+ | Task |Version|Filter| Metric | Value | |Stderr|
410
+ |--------------|-------|------|---------------|------:|---|-----:|
411
+ |arc_challenge |Yaml |none |acc | 0.1869|± |0.0114|
412
+ | | |none |acc_norm | 0.2415|± |0.0125|
413
+ |arc_easy |Yaml |none |acc | 0.4571|± |0.0102|
414
+ | | |none |acc_norm | 0.4335|± |0.0102|
415
+ |boolq |Yaml |none |acc | 0.5939|± |0.0086|
416
+ |hellaswag |Yaml |none |acc | 0.2862|± |0.0045|
417
+ | | |none |acc_norm | 0.3037|± |0.0046|
418
+ |lambada_openai|Yaml |none |perplexity |60.1068|± |2.7031|
419
+ | | |none |acc | 0.2829|± |0.0063|
420
+ |openbookqa |Yaml |none |acc | 0.1640|± |0.0166|
421
+ | | |none |acc_norm | 0.2560|± |0.0195|
422
+ |piqa |Yaml |none |acc | 0.6349|± |0.0112|
423
+ | | |none |acc_norm | 0.6306|± |0.0113|
424
+ |sciq |Yaml |none |acc | 0.7700|± |0.0133|
425
+ | | |none |acc_norm | 0.7470|± |0.0138|
426
+ |wikitext |Yaml |none |word_perplexity|69.5804| | |
427
+ | | |none |byte_perplexity| 2.0125| | |
428
+ | | |none |bits_per_byte | 1.0090| | |
429
+ |winogrande |Yaml |none |acc | 0.5162|± |0.0140|
430
+
sft-160m-eval-files/sft-pythia-160m-5shot/results.json ADDED
@@ -0,0 +1,406 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "results": {
3
+ "arc_challenge": {
4
+ "acc,none": 0.18686006825938567,
5
+ "acc_stderr,none": 0.011391015649694396,
6
+ "acc_norm,none": 0.24146757679180889,
7
+ "acc_norm_stderr,none": 0.012506564839739432
8
+ },
9
+ "arc_easy": {
10
+ "acc,none": 0.45707070707070707,
11
+ "acc_stderr,none": 0.010221897564256042,
12
+ "acc_norm,none": 0.4335016835016835,
13
+ "acc_norm_stderr,none": 0.010168640625454107
14
+ },
15
+ "boolq": {
16
+ "acc,none": 0.5938837920489297,
17
+ "acc_stderr,none": 0.008589510943787407
18
+ },
19
+ "hellaswag": {
20
+ "acc,none": 0.2861979685321649,
21
+ "acc_stderr,none": 0.004510593395289897,
22
+ "acc_norm,none": 0.3037243576976698,
23
+ "acc_norm_stderr,none": 0.00458925224362782
24
+ },
25
+ "lambada_openai": {
26
+ "perplexity,none": 60.10678259260095,
27
+ "perplexity_stderr,none": 2.70306190425745,
28
+ "acc,none": 0.2829419755482243,
29
+ "acc_stderr,none": 0.006275349431343604
30
+ },
31
+ "openbookqa": {
32
+ "acc,none": 0.164,
33
+ "acc_stderr,none": 0.016575811142446713,
34
+ "acc_norm,none": 0.256,
35
+ "acc_norm_stderr,none": 0.019536923574747598
36
+ },
37
+ "piqa": {
38
+ "acc,none": 0.6349292709466812,
39
+ "acc_stderr,none": 0.011233021830554834,
40
+ "acc_norm,none": 0.6305767138193689,
41
+ "acc_norm_stderr,none": 0.011260988628572333
42
+ },
43
+ "sciq": {
44
+ "acc,none": 0.77,
45
+ "acc_stderr,none": 0.013314551335935952,
46
+ "acc_norm,none": 0.747,
47
+ "acc_norm_stderr,none": 0.01375427861358708
48
+ },
49
+ "wikitext": {
50
+ "word_perplexity,none": 69.5803814818009,
51
+ "byte_perplexity,none": 2.012533549921474,
52
+ "bits_per_byte,none": 1.009012833861732
53
+ },
54
+ "winogrande": {
55
+ "acc,none": 0.516179952644041,
56
+ "acc_stderr,none": 0.0140451261309786
57
+ }
58
+ },
59
+ "configs": {
60
+ "arc_challenge": {
61
+ "task": "arc_challenge",
62
+ "group": [
63
+ "ai2_arc",
64
+ "multiple_choice"
65
+ ],
66
+ "dataset_path": "ai2_arc",
67
+ "dataset_name": "ARC-Challenge",
68
+ "training_split": "train",
69
+ "validation_split": "validation",
70
+ "test_split": "test",
71
+ "doc_to_text": "Question: {{question}}\nAnswer:",
72
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
73
+ "doc_to_choice": "{{choices.text}}",
74
+ "description": "",
75
+ "target_delimiter": " ",
76
+ "fewshot_delimiter": "\n\n",
77
+ "num_fewshot": 5,
78
+ "metric_list": [
79
+ {
80
+ "metric": "acc",
81
+ "aggregation": "mean",
82
+ "higher_is_better": true
83
+ },
84
+ {
85
+ "metric": "acc_norm",
86
+ "aggregation": "mean",
87
+ "higher_is_better": true
88
+ }
89
+ ],
90
+ "output_type": "multiple_choice",
91
+ "repeats": 1,
92
+ "should_decontaminate": true,
93
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
94
+ },
95
+ "arc_easy": {
96
+ "task": "arc_easy",
97
+ "group": [
98
+ "ai2_arc",
99
+ "multiple_choice"
100
+ ],
101
+ "dataset_path": "ai2_arc",
102
+ "dataset_name": "ARC-Easy",
103
+ "training_split": "train",
104
+ "validation_split": "validation",
105
+ "test_split": "test",
106
+ "doc_to_text": "Question: {{question}}\nAnswer:",
107
+ "doc_to_target": "{{choices.label.index(answerKey)}}",
108
+ "doc_to_choice": "{{choices.text}}",
109
+ "description": "",
110
+ "target_delimiter": " ",
111
+ "fewshot_delimiter": "\n\n",
112
+ "num_fewshot": 5,
113
+ "metric_list": [
114
+ {
115
+ "metric": "acc",
116
+ "aggregation": "mean",
117
+ "higher_is_better": true
118
+ },
119
+ {
120
+ "metric": "acc_norm",
121
+ "aggregation": "mean",
122
+ "higher_is_better": true
123
+ }
124
+ ],
125
+ "output_type": "multiple_choice",
126
+ "repeats": 1,
127
+ "should_decontaminate": true,
128
+ "doc_to_decontamination_query": "Question: {{question}}\nAnswer:"
129
+ },
130
+ "boolq": {
131
+ "task": "boolq",
132
+ "group": [
133
+ "super-glue-lm-eval-v1"
134
+ ],
135
+ "dataset_path": "super_glue",
136
+ "dataset_name": "boolq",
137
+ "training_split": "train",
138
+ "validation_split": "validation",
139
+ "doc_to_text": "{{passage}}\nQuestion: {{question}}?\nAnswer:",
140
+ "doc_to_target": "label",
141
+ "doc_to_choice": [
142
+ "no",
143
+ "yes"
144
+ ],
145
+ "description": "",
146
+ "target_delimiter": " ",
147
+ "fewshot_delimiter": "\n\n",
148
+ "num_fewshot": 5,
149
+ "metric_list": [
150
+ {
151
+ "metric": "acc"
152
+ }
153
+ ],
154
+ "output_type": "multiple_choice",
155
+ "repeats": 1,
156
+ "should_decontaminate": true,
157
+ "doc_to_decontamination_query": "passage"
158
+ },
159
+ "hellaswag": {
160
+ "task": "hellaswag",
161
+ "group": [
162
+ "multiple_choice"
163
+ ],
164
+ "dataset_path": "hellaswag",
165
+ "training_split": "train",
166
+ "validation_split": "validation",
167
+ "doc_to_text": "{% set text = activity_label ~ ': ' ~ ctx_a ~ ' ' ~ ctx_b.capitalize() %}{{text|trim|replace(' [title]', '. ')|regex_replace('\\[.*?\\]', '')|replace(' ', ' ')}}",
168
+ "doc_to_target": "{{label}}",
169
+ "doc_to_choice": "{{endings|map('trim')|map('replace', ' [title]', '. ')|map('regex_replace', '\\[.*?\\]', '')|map('replace', ' ', ' ')|list}}",
170
+ "description": "",
171
+ "target_delimiter": " ",
172
+ "fewshot_delimiter": "\n\n",
173
+ "num_fewshot": 5,
174
+ "metric_list": [
175
+ {
176
+ "metric": "acc",
177
+ "aggregation": "mean",
178
+ "higher_is_better": true
179
+ },
180
+ {
181
+ "metric": "acc_norm",
182
+ "aggregation": "mean",
183
+ "higher_is_better": true
184
+ }
185
+ ],
186
+ "output_type": "multiple_choice",
187
+ "repeats": 1,
188
+ "should_decontaminate": false
189
+ },
190
+ "lambada_openai": {
191
+ "task": "lambada_openai",
192
+ "group": [
193
+ "lambada",
194
+ "loglikelihood",
195
+ "perplexity"
196
+ ],
197
+ "dataset_path": "EleutherAI/lambada_openai",
198
+ "dataset_name": "default",
199
+ "test_split": "test",
200
+ "template_aliases": "",
201
+ "doc_to_text": "{{text.split(' ')[:-1]|join(' ')}}",
202
+ "doc_to_target": "{{' '+text.split(' ')[-1]}}",
203
+ "description": "",
204
+ "target_delimiter": " ",
205
+ "fewshot_delimiter": "\n\n",
206
+ "num_fewshot": 5,
207
+ "metric_list": [
208
+ {
209
+ "metric": "perplexity",
210
+ "aggregation": "perplexity",
211
+ "higher_is_better": false
212
+ },
213
+ {
214
+ "metric": "acc",
215
+ "aggregation": "mean",
216
+ "higher_is_better": true
217
+ }
218
+ ],
219
+ "output_type": "loglikelihood",
220
+ "repeats": 1,
221
+ "should_decontaminate": true,
222
+ "doc_to_decontamination_query": "{{text}}"
223
+ },
224
+ "openbookqa": {
225
+ "task": "openbookqa",
226
+ "group": [
227
+ "multiple_choice"
228
+ ],
229
+ "dataset_path": "openbookqa",
230
+ "dataset_name": "main",
231
+ "training_split": "train",
232
+ "validation_split": "validation",
233
+ "test_split": "test",
234
+ "doc_to_text": "question_stem",
235
+ "doc_to_target": "{{choices.label.index(answerKey.lstrip())}}",
236
+ "doc_to_choice": "{{choices.text}}",
237
+ "description": "",
238
+ "target_delimiter": " ",
239
+ "fewshot_delimiter": "\n\n",
240
+ "num_fewshot": 5,
241
+ "metric_list": [
242
+ {
243
+ "metric": "acc",
244
+ "aggregation": "mean",
245
+ "higher_is_better": true
246
+ },
247
+ {
248
+ "metric": "acc_norm",
249
+ "aggregation": "mean",
250
+ "higher_is_better": true
251
+ }
252
+ ],
253
+ "output_type": "multiple_choice",
254
+ "repeats": 1,
255
+ "should_decontaminate": true,
256
+ "doc_to_decontamination_query": "question_stem"
257
+ },
258
+ "piqa": {
259
+ "task": "piqa",
260
+ "group": [
261
+ "multiple_choice"
262
+ ],
263
+ "dataset_path": "piqa",
264
+ "training_split": "train",
265
+ "validation_split": "validation",
266
+ "doc_to_text": "Question: {{goal}}\nAnswer:",
267
+ "doc_to_target": "label",
268
+ "doc_to_choice": "{{[sol1, sol2]}}",
269
+ "description": "",
270
+ "target_delimiter": " ",
271
+ "fewshot_delimiter": "\n\n",
272
+ "num_fewshot": 5,
273
+ "metric_list": [
274
+ {
275
+ "metric": "acc",
276
+ "aggregation": "mean",
277
+ "higher_is_better": true
278
+ },
279
+ {
280
+ "metric": "acc_norm",
281
+ "aggregation": "mean",
282
+ "higher_is_better": true
283
+ }
284
+ ],
285
+ "output_type": "multiple_choice",
286
+ "repeats": 1,
287
+ "should_decontaminate": true,
288
+ "doc_to_decontamination_query": "goal"
289
+ },
290
+ "sciq": {
291
+ "task": "sciq",
292
+ "group": [
293
+ "multiple_choice"
294
+ ],
295
+ "dataset_path": "sciq",
296
+ "training_split": "train",
297
+ "validation_split": "validation",
298
+ "test_split": "test",
299
+ "doc_to_text": "{{support.lstrip()}}\nQuestion: {{question}}\nAnswer:",
300
+ "doc_to_target": 3,
301
+ "doc_to_choice": "{{[distractor1, distractor2, distractor3, correct_answer]}}",
302
+ "description": "",
303
+ "target_delimiter": " ",
304
+ "fewshot_delimiter": "\n\n",
305
+ "num_fewshot": 5,
306
+ "metric_list": [
307
+ {
308
+ "metric": "acc",
309
+ "aggregation": "mean",
310
+ "higher_is_better": true
311
+ },
312
+ {
313
+ "metric": "acc_norm",
314
+ "aggregation": "mean",
315
+ "higher_is_better": true
316
+ }
317
+ ],
318
+ "output_type": "multiple_choice",
319
+ "repeats": 1,
320
+ "should_decontaminate": true,
321
+ "doc_to_decontamination_query": "{{support}} {{question}}"
322
+ },
323
+ "wikitext": {
324
+ "task": "wikitext",
325
+ "group": [
326
+ "perplexity",
327
+ "loglikelihood_rolling"
328
+ ],
329
+ "dataset_path": "EleutherAI/wikitext_document_level",
330
+ "dataset_name": "wikitext-2-raw-v1",
331
+ "training_split": "train",
332
+ "validation_split": "validation",
333
+ "test_split": "test",
334
+ "template_aliases": "",
335
+ "doc_to_text": "",
336
+ "doc_to_target": "<function wikitext_detokenizer at 0x7f50b9b54040>",
337
+ "description": "",
338
+ "target_delimiter": " ",
339
+ "fewshot_delimiter": "\n\n",
340
+ "num_fewshot": 5,
341
+ "metric_list": [
342
+ {
343
+ "metric": "word_perplexity"
344
+ },
345
+ {
346
+ "metric": "byte_perplexity"
347
+ },
348
+ {
349
+ "metric": "bits_per_byte"
350
+ }
351
+ ],
352
+ "output_type": "loglikelihood_rolling",
353
+ "repeats": 1,
354
+ "should_decontaminate": true,
355
+ "doc_to_decontamination_query": "{{page}}"
356
+ },
357
+ "winogrande": {
358
+ "task": "winogrande",
359
+ "dataset_path": "winogrande",
360
+ "dataset_name": "winogrande_xl",
361
+ "training_split": "train",
362
+ "validation_split": "validation",
363
+ "doc_to_text": "<function doc_to_text at 0x7f50b9b26ef0>",
364
+ "doc_to_target": "<function doc_to_target at 0x7f50b9b27370>",
365
+ "doc_to_choice": "<function doc_to_choice at 0x7f50b9b275b0>",
366
+ "description": "",
367
+ "target_delimiter": " ",
368
+ "fewshot_delimiter": "\n\n",
369
+ "num_fewshot": 5,
370
+ "metric_list": [
371
+ {
372
+ "metric": "acc",
373
+ "aggregation": "mean",
374
+ "higher_is_better": true
375
+ }
376
+ ],
377
+ "output_type": "multiple_choice",
378
+ "repeats": 1,
379
+ "should_decontaminate": false
380
+ }
381
+ },
382
+ "versions": {
383
+ "arc_challenge": "Yaml",
384
+ "arc_easy": "Yaml",
385
+ "boolq": "Yaml",
386
+ "hellaswag": "Yaml",
387
+ "lambada_openai": "Yaml",
388
+ "openbookqa": "Yaml",
389
+ "piqa": "Yaml",
390
+ "sciq": "Yaml",
391
+ "wikitext": "Yaml",
392
+ "winogrande": "Yaml"
393
+ },
394
+ "config": {
395
+ "model": "hf",
396
+ "model_args": "pretrained=lomahony/eleuther-pythia160m-hh-sft",
397
+ "num_fewshot": 5,
398
+ "batch_size": 16,
399
+ "batch_sizes": [],
400
+ "device": "cuda:0",
401
+ "use_cache": null,
402
+ "limit": null,
403
+ "bootstrap_iters": 100000
404
+ },
405
+ "git_hash": "4e44f0a"
406
+ }