Commit
•
564e6fe
1
Parent(s):
d048d70
Delete Tess-v2.5-Qwen2-72B-mmlu.json
Browse files- Tess-v2.5-Qwen2-72B-mmlu.json +0 -3158
Tess-v2.5-Qwen2-72B-mmlu.json
DELETED
|
@@ -1,3158 +0,0 @@
|
|
| 1 |
-
{
|
| 2 |
-
"results": {
|
| 3 |
-
"mmlu": {
|
| 4 |
-
"acc,none": 0.8439680957128615,
|
| 5 |
-
"acc_stderr,none": 0.0029499711040394372,
|
| 6 |
-
"alias": "mmlu"
|
| 7 |
-
},
|
| 8 |
-
"mmlu_humanities": {
|
| 9 |
-
"alias": " - humanities",
|
| 10 |
-
"acc,none": 0.8146652497343252,
|
| 11 |
-
"acc_stderr,none": 0.005505402478774841
|
| 12 |
-
},
|
| 13 |
-
"mmlu_formal_logic": {
|
| 14 |
-
"alias": " - formal_logic",
|
| 15 |
-
"acc,none": 0.7301587301587301,
|
| 16 |
-
"acc_stderr,none": 0.03970158273235173
|
| 17 |
-
},
|
| 18 |
-
"mmlu_high_school_european_history": {
|
| 19 |
-
"alias": " - high_school_european_history",
|
| 20 |
-
"acc,none": 0.8909090909090909,
|
| 21 |
-
"acc_stderr,none": 0.02434383813514564
|
| 22 |
-
},
|
| 23 |
-
"mmlu_high_school_us_history": {
|
| 24 |
-
"alias": " - high_school_us_history",
|
| 25 |
-
"acc,none": 0.9509803921568627,
|
| 26 |
-
"acc_stderr,none": 0.01515383934021267
|
| 27 |
-
},
|
| 28 |
-
"mmlu_high_school_world_history": {
|
| 29 |
-
"alias": " - high_school_world_history",
|
| 30 |
-
"acc,none": 0.9409282700421941,
|
| 31 |
-
"acc_stderr,none": 0.01534659746388869
|
| 32 |
-
},
|
| 33 |
-
"mmlu_international_law": {
|
| 34 |
-
"alias": " - international_law",
|
| 35 |
-
"acc,none": 0.9173553719008265,
|
| 36 |
-
"acc_stderr,none": 0.025135382356604227
|
| 37 |
-
},
|
| 38 |
-
"mmlu_jurisprudence": {
|
| 39 |
-
"alias": " - jurisprudence",
|
| 40 |
-
"acc,none": 0.8796296296296297,
|
| 41 |
-
"acc_stderr,none": 0.031457038543062525
|
| 42 |
-
},
|
| 43 |
-
"mmlu_logical_fallacies": {
|
| 44 |
-
"alias": " - logical_fallacies",
|
| 45 |
-
"acc,none": 0.901840490797546,
|
| 46 |
-
"acc_stderr,none": 0.023376180231059605
|
| 47 |
-
},
|
| 48 |
-
"mmlu_moral_disputes": {
|
| 49 |
-
"alias": " - moral_disputes",
|
| 50 |
-
"acc,none": 0.869942196531792,
|
| 51 |
-
"acc_stderr,none": 0.01810939152822133
|
| 52 |
-
},
|
| 53 |
-
"mmlu_moral_scenarios": {
|
| 54 |
-
"alias": " - moral_scenarios",
|
| 55 |
-
"acc,none": 0.829050279329609,
|
| 56 |
-
"acc_stderr,none": 0.012590873868789222
|
| 57 |
-
},
|
| 58 |
-
"mmlu_philosophy": {
|
| 59 |
-
"alias": " - philosophy",
|
| 60 |
-
"acc,none": 0.8713826366559485,
|
| 61 |
-
"acc_stderr,none": 0.01901399630412152
|
| 62 |
-
},
|
| 63 |
-
"mmlu_prehistory": {
|
| 64 |
-
"alias": " - prehistory",
|
| 65 |
-
"acc,none": 0.9104938271604939,
|
| 66 |
-
"acc_stderr,none": 0.015884141073937555
|
| 67 |
-
},
|
| 68 |
-
"mmlu_professional_law": {
|
| 69 |
-
"alias": " - professional_law",
|
| 70 |
-
"acc,none": 0.6929595827900913,
|
| 71 |
-
"acc_stderr,none": 0.011780959114513764
|
| 72 |
-
},
|
| 73 |
-
"mmlu_world_religions": {
|
| 74 |
-
"alias": " - world_religions",
|
| 75 |
-
"acc,none": 0.8888888888888888,
|
| 76 |
-
"acc_stderr,none": 0.024103384202072864
|
| 77 |
-
},
|
| 78 |
-
"mmlu_other": {
|
| 79 |
-
"alias": " - other",
|
| 80 |
-
"acc,none": 0.8625683939491471,
|
| 81 |
-
"acc_stderr,none": 0.005895325056685939
|
| 82 |
-
},
|
| 83 |
-
"mmlu_business_ethics": {
|
| 84 |
-
"alias": " - business_ethics",
|
| 85 |
-
"acc,none": 0.78,
|
| 86 |
-
"acc_stderr,none": 0.04163331998932263
|
| 87 |
-
},
|
| 88 |
-
"mmlu_clinical_knowledge": {
|
| 89 |
-
"alias": " - clinical_knowledge",
|
| 90 |
-
"acc,none": 0.8716981132075472,
|
| 91 |
-
"acc_stderr,none": 0.02058247568799185
|
| 92 |
-
},
|
| 93 |
-
"mmlu_college_medicine": {
|
| 94 |
-
"alias": " - college_medicine",
|
| 95 |
-
"acc,none": 0.8323699421965318,
|
| 96 |
-
"acc_stderr,none": 0.028481963032143395
|
| 97 |
-
},
|
| 98 |
-
"mmlu_global_facts": {
|
| 99 |
-
"alias": " - global_facts",
|
| 100 |
-
"acc,none": 0.61,
|
| 101 |
-
"acc_stderr,none": 0.04902071300001975
|
| 102 |
-
},
|
| 103 |
-
"mmlu_human_aging": {
|
| 104 |
-
"alias": " - human_aging",
|
| 105 |
-
"acc,none": 0.8565022421524664,
|
| 106 |
-
"acc_stderr,none": 0.0235293712696182
|
| 107 |
-
},
|
| 108 |
-
"mmlu_management": {
|
| 109 |
-
"alias": " - management",
|
| 110 |
-
"acc,none": 0.9223300970873787,
|
| 111 |
-
"acc_stderr,none": 0.026501440784762766
|
| 112 |
-
},
|
| 113 |
-
"mmlu_marketing": {
|
| 114 |
-
"alias": " - marketing",
|
| 115 |
-
"acc,none": 0.9487179487179487,
|
| 116 |
-
"acc_stderr,none": 0.014450181176872726
|
| 117 |
-
},
|
| 118 |
-
"mmlu_medical_genetics": {
|
| 119 |
-
"alias": " - medical_genetics",
|
| 120 |
-
"acc,none": 0.9,
|
| 121 |
-
"acc_stderr,none": 0.030151134457776348
|
| 122 |
-
},
|
| 123 |
-
"mmlu_miscellaneous": {
|
| 124 |
-
"alias": " - miscellaneous",
|
| 125 |
-
"acc,none": 0.9501915708812261,
|
| 126 |
-
"acc_stderr,none": 0.0077795348866793465
|
| 127 |
-
},
|
| 128 |
-
"mmlu_nutrition": {
|
| 129 |
-
"alias": " - nutrition",
|
| 130 |
-
"acc,none": 0.9019607843137255,
|
| 131 |
-
"acc_stderr,none": 0.017027222935582193
|
| 132 |
-
},
|
| 133 |
-
"mmlu_professional_accounting": {
|
| 134 |
-
"alias": " - professional_accounting",
|
| 135 |
-
"acc,none": 0.75177304964539,
|
| 136 |
-
"acc_stderr,none": 0.025770015644290392
|
| 137 |
-
},
|
| 138 |
-
"mmlu_professional_medicine": {
|
| 139 |
-
"alias": " - professional_medicine",
|
| 140 |
-
"acc,none": 0.8897058823529411,
|
| 141 |
-
"acc_stderr,none": 0.019028947191474497
|
| 142 |
-
},
|
| 143 |
-
"mmlu_virology": {
|
| 144 |
-
"alias": " - virology",
|
| 145 |
-
"acc,none": 0.5662650602409639,
|
| 146 |
-
"acc_stderr,none": 0.03858158940685517
|
| 147 |
-
},
|
| 148 |
-
"mmlu_social_sciences": {
|
| 149 |
-
"alias": " - social_sciences",
|
| 150 |
-
"acc,none": 0.9038024049398765,
|
| 151 |
-
"acc_stderr,none": 0.005221504585802578
|
| 152 |
-
},
|
| 153 |
-
"mmlu_econometrics": {
|
| 154 |
-
"alias": " - econometrics",
|
| 155 |
-
"acc,none": 0.7280701754385965,
|
| 156 |
-
"acc_stderr,none": 0.041857744240220554
|
| 157 |
-
},
|
| 158 |
-
"mmlu_high_school_geography": {
|
| 159 |
-
"alias": " - high_school_geography",
|
| 160 |
-
"acc,none": 0.9393939393939394,
|
| 161 |
-
"acc_stderr,none": 0.016999994927421606
|
| 162 |
-
},
|
| 163 |
-
"mmlu_high_school_government_and_politics": {
|
| 164 |
-
"alias": " - high_school_government_and_politics",
|
| 165 |
-
"acc,none": 0.9896373056994818,
|
| 166 |
-
"acc_stderr,none": 0.007308424386792201
|
| 167 |
-
},
|
| 168 |
-
"mmlu_high_school_macroeconomics": {
|
| 169 |
-
"alias": " - high_school_macroeconomics",
|
| 170 |
-
"acc,none": 0.8897435897435897,
|
| 171 |
-
"acc_stderr,none": 0.015880331261056115
|
| 172 |
-
},
|
| 173 |
-
"mmlu_high_school_microeconomics": {
|
| 174 |
-
"alias": " - high_school_microeconomics",
|
| 175 |
-
"acc,none": 0.9411764705882353,
|
| 176 |
-
"acc_stderr,none": 0.015283995352038402
|
| 177 |
-
},
|
| 178 |
-
"mmlu_high_school_psychology": {
|
| 179 |
-
"alias": " - high_school_psychology",
|
| 180 |
-
"acc,none": 0.9357798165137615,
|
| 181 |
-
"acc_stderr,none": 0.010510494713201424
|
| 182 |
-
},
|
| 183 |
-
"mmlu_human_sexuality": {
|
| 184 |
-
"alias": " - human_sexuality",
|
| 185 |
-
"acc,none": 0.9083969465648855,
|
| 186 |
-
"acc_stderr,none": 0.025300035578642965
|
| 187 |
-
},
|
| 188 |
-
"mmlu_professional_psychology": {
|
| 189 |
-
"alias": " - professional_psychology",
|
| 190 |
-
"acc,none": 0.8970588235294118,
|
| 191 |
-
"acc_stderr,none": 0.012293751200845176
|
| 192 |
-
},
|
| 193 |
-
"mmlu_public_relations": {
|
| 194 |
-
"alias": " - public_relations",
|
| 195 |
-
"acc,none": 0.7454545454545455,
|
| 196 |
-
"acc_stderr,none": 0.041723430387053825
|
| 197 |
-
},
|
| 198 |
-
"mmlu_security_studies": {
|
| 199 |
-
"alias": " - security_studies",
|
| 200 |
-
"acc,none": 0.8408163265306122,
|
| 201 |
-
"acc_stderr,none": 0.023420972069166365
|
| 202 |
-
},
|
| 203 |
-
"mmlu_sociology": {
|
| 204 |
-
"alias": " - sociology",
|
| 205 |
-
"acc,none": 0.945273631840796,
|
| 206 |
-
"acc_stderr,none": 0.016082815796263254
|
| 207 |
-
},
|
| 208 |
-
"mmlu_us_foreign_policy": {
|
| 209 |
-
"alias": " - us_foreign_policy",
|
| 210 |
-
"acc,none": 0.94,
|
| 211 |
-
"acc_stderr,none": 0.02386832565759419
|
| 212 |
-
},
|
| 213 |
-
"mmlu_stem": {
|
| 214 |
-
"alias": " - stem",
|
| 215 |
-
"acc,none": 0.8109736758642563,
|
| 216 |
-
"acc_stderr,none": 0.0067376135296805745
|
| 217 |
-
},
|
| 218 |
-
"mmlu_abstract_algebra": {
|
| 219 |
-
"alias": " - abstract_algebra",
|
| 220 |
-
"acc,none": 0.66,
|
| 221 |
-
"acc_stderr,none": 0.04760952285695237
|
| 222 |
-
},
|
| 223 |
-
"mmlu_anatomy": {
|
| 224 |
-
"alias": " - anatomy",
|
| 225 |
-
"acc,none": 0.7925925925925926,
|
| 226 |
-
"acc_stderr,none": 0.03502553170678317
|
| 227 |
-
},
|
| 228 |
-
"mmlu_astronomy": {
|
| 229 |
-
"alias": " - astronomy",
|
| 230 |
-
"acc,none": 0.9276315789473685,
|
| 231 |
-
"acc_stderr,none": 0.021085011261884112
|
| 232 |
-
},
|
| 233 |
-
"mmlu_college_biology": {
|
| 234 |
-
"alias": " - college_biology",
|
| 235 |
-
"acc,none": 0.9444444444444444,
|
| 236 |
-
"acc_stderr,none": 0.01915507853243362
|
| 237 |
-
},
|
| 238 |
-
"mmlu_college_chemistry": {
|
| 239 |
-
"alias": " - college_chemistry",
|
| 240 |
-
"acc,none": 0.58,
|
| 241 |
-
"acc_stderr,none": 0.049604496374885836
|
| 242 |
-
},
|
| 243 |
-
"mmlu_college_computer_science": {
|
| 244 |
-
"alias": " - college_computer_science",
|
| 245 |
-
"acc,none": 0.8,
|
| 246 |
-
"acc_stderr,none": 0.040201512610368445
|
| 247 |
-
},
|
| 248 |
-
"mmlu_college_mathematics": {
|
| 249 |
-
"alias": " - college_mathematics",
|
| 250 |
-
"acc,none": 0.63,
|
| 251 |
-
"acc_stderr,none": 0.04852365870939099
|
| 252 |
-
},
|
| 253 |
-
"mmlu_college_physics": {
|
| 254 |
-
"alias": " - college_physics",
|
| 255 |
-
"acc,none": 0.6470588235294118,
|
| 256 |
-
"acc_stderr,none": 0.04755129616062947
|
| 257 |
-
},
|
| 258 |
-
"mmlu_computer_security": {
|
| 259 |
-
"alias": " - computer_security",
|
| 260 |
-
"acc,none": 0.83,
|
| 261 |
-
"acc_stderr,none": 0.0377525168068637
|
| 262 |
-
},
|
| 263 |
-
"mmlu_conceptual_physics": {
|
| 264 |
-
"alias": " - conceptual_physics",
|
| 265 |
-
"acc,none": 0.8893617021276595,
|
| 266 |
-
"acc_stderr,none": 0.020506145099008433
|
| 267 |
-
},
|
| 268 |
-
"mmlu_electrical_engineering": {
|
| 269 |
-
"alias": " - electrical_engineering",
|
| 270 |
-
"acc,none": 0.8275862068965517,
|
| 271 |
-
"acc_stderr,none": 0.03147830790259575
|
| 272 |
-
},
|
| 273 |
-
"mmlu_elementary_mathematics": {
|
| 274 |
-
"alias": " - elementary_mathematics",
|
| 275 |
-
"acc,none": 0.8888888888888888,
|
| 276 |
-
"acc_stderr,none": 0.01618571201620511
|
| 277 |
-
},
|
| 278 |
-
"mmlu_high_school_biology": {
|
| 279 |
-
"alias": " - high_school_biology",
|
| 280 |
-
"acc,none": 0.9419354838709677,
|
| 281 |
-
"acc_stderr,none": 0.01330413811280927
|
| 282 |
-
},
|
| 283 |
-
"mmlu_high_school_chemistry": {
|
| 284 |
-
"alias": " - high_school_chemistry",
|
| 285 |
-
"acc,none": 0.7980295566502463,
|
| 286 |
-
"acc_stderr,none": 0.028247350122180243
|
| 287 |
-
},
|
| 288 |
-
"mmlu_high_school_computer_science": {
|
| 289 |
-
"alias": " - high_school_computer_science",
|
| 290 |
-
"acc,none": 0.91,
|
| 291 |
-
"acc_stderr,none": 0.028762349126466115
|
| 292 |
-
},
|
| 293 |
-
"mmlu_high_school_mathematics": {
|
| 294 |
-
"alias": " - high_school_mathematics",
|
| 295 |
-
"acc,none": 0.6777777777777778,
|
| 296 |
-
"acc_stderr,none": 0.028493465091028597
|
| 297 |
-
},
|
| 298 |
-
"mmlu_high_school_physics": {
|
| 299 |
-
"alias": " - high_school_physics",
|
| 300 |
-
"acc,none": 0.7284768211920529,
|
| 301 |
-
"acc_stderr,none": 0.03631329803969654
|
| 302 |
-
},
|
| 303 |
-
"mmlu_high_school_statistics": {
|
| 304 |
-
"alias": " - high_school_statistics",
|
| 305 |
-
"acc,none": 0.7824074074074074,
|
| 306 |
-
"acc_stderr,none": 0.028139689444859676
|
| 307 |
-
},
|
| 308 |
-
"mmlu_machine_learning": {
|
| 309 |
-
"alias": " - machine_learning",
|
| 310 |
-
"acc,none": 0.7589285714285714,
|
| 311 |
-
"acc_stderr,none": 0.04059867246952685
|
| 312 |
-
}
|
| 313 |
-
},
|
| 314 |
-
"groups": {
|
| 315 |
-
"mmlu": {
|
| 316 |
-
"acc,none": 0.8439680957128615,
|
| 317 |
-
"acc_stderr,none": 0.0029499711040394372,
|
| 318 |
-
"alias": "mmlu"
|
| 319 |
-
},
|
| 320 |
-
"mmlu_humanities": {
|
| 321 |
-
"alias": " - humanities",
|
| 322 |
-
"acc,none": 0.8146652497343252,
|
| 323 |
-
"acc_stderr,none": 0.005505402478774841
|
| 324 |
-
},
|
| 325 |
-
"mmlu_other": {
|
| 326 |
-
"alias": " - other",
|
| 327 |
-
"acc,none": 0.8625683939491471,
|
| 328 |
-
"acc_stderr,none": 0.005895325056685939
|
| 329 |
-
},
|
| 330 |
-
"mmlu_social_sciences": {
|
| 331 |
-
"alias": " - social_sciences",
|
| 332 |
-
"acc,none": 0.9038024049398765,
|
| 333 |
-
"acc_stderr,none": 0.005221504585802578
|
| 334 |
-
},
|
| 335 |
-
"mmlu_stem": {
|
| 336 |
-
"alias": " - stem",
|
| 337 |
-
"acc,none": 0.8109736758642563,
|
| 338 |
-
"acc_stderr,none": 0.0067376135296805745
|
| 339 |
-
}
|
| 340 |
-
},
|
| 341 |
-
"group_subtasks": {
|
| 342 |
-
"mmlu_stem": [
|
| 343 |
-
"mmlu_college_biology",
|
| 344 |
-
"mmlu_high_school_computer_science",
|
| 345 |
-
"mmlu_elementary_mathematics",
|
| 346 |
-
"mmlu_astronomy",
|
| 347 |
-
"mmlu_machine_learning",
|
| 348 |
-
"mmlu_high_school_mathematics",
|
| 349 |
-
"mmlu_electrical_engineering",
|
| 350 |
-
"mmlu_college_chemistry",
|
| 351 |
-
"mmlu_college_mathematics",
|
| 352 |
-
"mmlu_high_school_statistics",
|
| 353 |
-
"mmlu_high_school_biology",
|
| 354 |
-
"mmlu_abstract_algebra",
|
| 355 |
-
"mmlu_college_physics",
|
| 356 |
-
"mmlu_conceptual_physics",
|
| 357 |
-
"mmlu_computer_security",
|
| 358 |
-
"mmlu_anatomy",
|
| 359 |
-
"mmlu_college_computer_science",
|
| 360 |
-
"mmlu_high_school_physics",
|
| 361 |
-
"mmlu_high_school_chemistry"
|
| 362 |
-
],
|
| 363 |
-
"mmlu_other": [
|
| 364 |
-
"mmlu_marketing",
|
| 365 |
-
"mmlu_professional_accounting",
|
| 366 |
-
"mmlu_clinical_knowledge",
|
| 367 |
-
"mmlu_college_medicine",
|
| 368 |
-
"mmlu_miscellaneous",
|
| 369 |
-
"mmlu_virology",
|
| 370 |
-
"mmlu_business_ethics",
|
| 371 |
-
"mmlu_professional_medicine",
|
| 372 |
-
"mmlu_global_facts",
|
| 373 |
-
"mmlu_nutrition",
|
| 374 |
-
"mmlu_human_aging",
|
| 375 |
-
"mmlu_management",
|
| 376 |
-
"mmlu_medical_genetics"
|
| 377 |
-
],
|
| 378 |
-
"mmlu_social_sciences": [
|
| 379 |
-
"mmlu_high_school_psychology",
|
| 380 |
-
"mmlu_high_school_geography",
|
| 381 |
-
"mmlu_high_school_macroeconomics",
|
| 382 |
-
"mmlu_public_relations",
|
| 383 |
-
"mmlu_security_studies",
|
| 384 |
-
"mmlu_high_school_microeconomics",
|
| 385 |
-
"mmlu_human_sexuality",
|
| 386 |
-
"mmlu_sociology",
|
| 387 |
-
"mmlu_professional_psychology",
|
| 388 |
-
"mmlu_econometrics",
|
| 389 |
-
"mmlu_us_foreign_policy",
|
| 390 |
-
"mmlu_high_school_government_and_politics"
|
| 391 |
-
],
|
| 392 |
-
"mmlu_humanities": [
|
| 393 |
-
"mmlu_moral_scenarios",
|
| 394 |
-
"mmlu_high_school_us_history",
|
| 395 |
-
"mmlu_high_school_world_history",
|
| 396 |
-
"mmlu_world_religions",
|
| 397 |
-
"mmlu_formal_logic",
|
| 398 |
-
"mmlu_moral_disputes",
|
| 399 |
-
"mmlu_prehistory",
|
| 400 |
-
"mmlu_international_law",
|
| 401 |
-
"mmlu_logical_fallacies",
|
| 402 |
-
"mmlu_professional_law",
|
| 403 |
-
"mmlu_philosophy",
|
| 404 |
-
"mmlu_high_school_european_history",
|
| 405 |
-
"mmlu_jurisprudence"
|
| 406 |
-
],
|
| 407 |
-
"mmlu": [
|
| 408 |
-
"mmlu_humanities",
|
| 409 |
-
"mmlu_social_sciences",
|
| 410 |
-
"mmlu_other",
|
| 411 |
-
"mmlu_stem"
|
| 412 |
-
]
|
| 413 |
-
},
|
| 414 |
-
"configs": {
|
| 415 |
-
"mmlu_abstract_algebra": {
|
| 416 |
-
"task": "mmlu_abstract_algebra",
|
| 417 |
-
"task_alias": "abstract_algebra",
|
| 418 |
-
"group": "mmlu_stem",
|
| 419 |
-
"group_alias": "stem",
|
| 420 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 421 |
-
"dataset_name": "abstract_algebra",
|
| 422 |
-
"test_split": "test",
|
| 423 |
-
"fewshot_split": "dev",
|
| 424 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 425 |
-
"doc_to_target": "answer",
|
| 426 |
-
"doc_to_choice": [
|
| 427 |
-
"A",
|
| 428 |
-
"B",
|
| 429 |
-
"C",
|
| 430 |
-
"D"
|
| 431 |
-
],
|
| 432 |
-
"description": "The following are multiple choice questions (with answers) about abstract algebra.\n\n",
|
| 433 |
-
"target_delimiter": " ",
|
| 434 |
-
"fewshot_delimiter": "\n\n",
|
| 435 |
-
"fewshot_config": {
|
| 436 |
-
"sampler": "first_n"
|
| 437 |
-
},
|
| 438 |
-
"num_fewshot": 5,
|
| 439 |
-
"metric_list": [
|
| 440 |
-
{
|
| 441 |
-
"metric": "acc",
|
| 442 |
-
"aggregation": "mean",
|
| 443 |
-
"higher_is_better": true
|
| 444 |
-
}
|
| 445 |
-
],
|
| 446 |
-
"output_type": "multiple_choice",
|
| 447 |
-
"repeats": 1,
|
| 448 |
-
"should_decontaminate": false,
|
| 449 |
-
"metadata": {
|
| 450 |
-
"version": 0.0
|
| 451 |
-
}
|
| 452 |
-
},
|
| 453 |
-
"mmlu_anatomy": {
|
| 454 |
-
"task": "mmlu_anatomy",
|
| 455 |
-
"task_alias": "anatomy",
|
| 456 |
-
"group": "mmlu_stem",
|
| 457 |
-
"group_alias": "stem",
|
| 458 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 459 |
-
"dataset_name": "anatomy",
|
| 460 |
-
"test_split": "test",
|
| 461 |
-
"fewshot_split": "dev",
|
| 462 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 463 |
-
"doc_to_target": "answer",
|
| 464 |
-
"doc_to_choice": [
|
| 465 |
-
"A",
|
| 466 |
-
"B",
|
| 467 |
-
"C",
|
| 468 |
-
"D"
|
| 469 |
-
],
|
| 470 |
-
"description": "The following are multiple choice questions (with answers) about anatomy.\n\n",
|
| 471 |
-
"target_delimiter": " ",
|
| 472 |
-
"fewshot_delimiter": "\n\n",
|
| 473 |
-
"fewshot_config": {
|
| 474 |
-
"sampler": "first_n"
|
| 475 |
-
},
|
| 476 |
-
"num_fewshot": 5,
|
| 477 |
-
"metric_list": [
|
| 478 |
-
{
|
| 479 |
-
"metric": "acc",
|
| 480 |
-
"aggregation": "mean",
|
| 481 |
-
"higher_is_better": true
|
| 482 |
-
}
|
| 483 |
-
],
|
| 484 |
-
"output_type": "multiple_choice",
|
| 485 |
-
"repeats": 1,
|
| 486 |
-
"should_decontaminate": false,
|
| 487 |
-
"metadata": {
|
| 488 |
-
"version": 0.0
|
| 489 |
-
}
|
| 490 |
-
},
|
| 491 |
-
"mmlu_astronomy": {
|
| 492 |
-
"task": "mmlu_astronomy",
|
| 493 |
-
"task_alias": "astronomy",
|
| 494 |
-
"group": "mmlu_stem",
|
| 495 |
-
"group_alias": "stem",
|
| 496 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 497 |
-
"dataset_name": "astronomy",
|
| 498 |
-
"test_split": "test",
|
| 499 |
-
"fewshot_split": "dev",
|
| 500 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 501 |
-
"doc_to_target": "answer",
|
| 502 |
-
"doc_to_choice": [
|
| 503 |
-
"A",
|
| 504 |
-
"B",
|
| 505 |
-
"C",
|
| 506 |
-
"D"
|
| 507 |
-
],
|
| 508 |
-
"description": "The following are multiple choice questions (with answers) about astronomy.\n\n",
|
| 509 |
-
"target_delimiter": " ",
|
| 510 |
-
"fewshot_delimiter": "\n\n",
|
| 511 |
-
"fewshot_config": {
|
| 512 |
-
"sampler": "first_n"
|
| 513 |
-
},
|
| 514 |
-
"num_fewshot": 5,
|
| 515 |
-
"metric_list": [
|
| 516 |
-
{
|
| 517 |
-
"metric": "acc",
|
| 518 |
-
"aggregation": "mean",
|
| 519 |
-
"higher_is_better": true
|
| 520 |
-
}
|
| 521 |
-
],
|
| 522 |
-
"output_type": "multiple_choice",
|
| 523 |
-
"repeats": 1,
|
| 524 |
-
"should_decontaminate": false,
|
| 525 |
-
"metadata": {
|
| 526 |
-
"version": 0.0
|
| 527 |
-
}
|
| 528 |
-
},
|
| 529 |
-
"mmlu_business_ethics": {
|
| 530 |
-
"task": "mmlu_business_ethics",
|
| 531 |
-
"task_alias": "business_ethics",
|
| 532 |
-
"group": "mmlu_other",
|
| 533 |
-
"group_alias": "other",
|
| 534 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 535 |
-
"dataset_name": "business_ethics",
|
| 536 |
-
"test_split": "test",
|
| 537 |
-
"fewshot_split": "dev",
|
| 538 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 539 |
-
"doc_to_target": "answer",
|
| 540 |
-
"doc_to_choice": [
|
| 541 |
-
"A",
|
| 542 |
-
"B",
|
| 543 |
-
"C",
|
| 544 |
-
"D"
|
| 545 |
-
],
|
| 546 |
-
"description": "The following are multiple choice questions (with answers) about business ethics.\n\n",
|
| 547 |
-
"target_delimiter": " ",
|
| 548 |
-
"fewshot_delimiter": "\n\n",
|
| 549 |
-
"fewshot_config": {
|
| 550 |
-
"sampler": "first_n"
|
| 551 |
-
},
|
| 552 |
-
"num_fewshot": 5,
|
| 553 |
-
"metric_list": [
|
| 554 |
-
{
|
| 555 |
-
"metric": "acc",
|
| 556 |
-
"aggregation": "mean",
|
| 557 |
-
"higher_is_better": true
|
| 558 |
-
}
|
| 559 |
-
],
|
| 560 |
-
"output_type": "multiple_choice",
|
| 561 |
-
"repeats": 1,
|
| 562 |
-
"should_decontaminate": false,
|
| 563 |
-
"metadata": {
|
| 564 |
-
"version": 0.0
|
| 565 |
-
}
|
| 566 |
-
},
|
| 567 |
-
"mmlu_clinical_knowledge": {
|
| 568 |
-
"task": "mmlu_clinical_knowledge",
|
| 569 |
-
"task_alias": "clinical_knowledge",
|
| 570 |
-
"group": "mmlu_other",
|
| 571 |
-
"group_alias": "other",
|
| 572 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 573 |
-
"dataset_name": "clinical_knowledge",
|
| 574 |
-
"test_split": "test",
|
| 575 |
-
"fewshot_split": "dev",
|
| 576 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 577 |
-
"doc_to_target": "answer",
|
| 578 |
-
"doc_to_choice": [
|
| 579 |
-
"A",
|
| 580 |
-
"B",
|
| 581 |
-
"C",
|
| 582 |
-
"D"
|
| 583 |
-
],
|
| 584 |
-
"description": "The following are multiple choice questions (with answers) about clinical knowledge.\n\n",
|
| 585 |
-
"target_delimiter": " ",
|
| 586 |
-
"fewshot_delimiter": "\n\n",
|
| 587 |
-
"fewshot_config": {
|
| 588 |
-
"sampler": "first_n"
|
| 589 |
-
},
|
| 590 |
-
"num_fewshot": 5,
|
| 591 |
-
"metric_list": [
|
| 592 |
-
{
|
| 593 |
-
"metric": "acc",
|
| 594 |
-
"aggregation": "mean",
|
| 595 |
-
"higher_is_better": true
|
| 596 |
-
}
|
| 597 |
-
],
|
| 598 |
-
"output_type": "multiple_choice",
|
| 599 |
-
"repeats": 1,
|
| 600 |
-
"should_decontaminate": false,
|
| 601 |
-
"metadata": {
|
| 602 |
-
"version": 0.0
|
| 603 |
-
}
|
| 604 |
-
},
|
| 605 |
-
"mmlu_college_biology": {
|
| 606 |
-
"task": "mmlu_college_biology",
|
| 607 |
-
"task_alias": "college_biology",
|
| 608 |
-
"group": "mmlu_stem",
|
| 609 |
-
"group_alias": "stem",
|
| 610 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 611 |
-
"dataset_name": "college_biology",
|
| 612 |
-
"test_split": "test",
|
| 613 |
-
"fewshot_split": "dev",
|
| 614 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 615 |
-
"doc_to_target": "answer",
|
| 616 |
-
"doc_to_choice": [
|
| 617 |
-
"A",
|
| 618 |
-
"B",
|
| 619 |
-
"C",
|
| 620 |
-
"D"
|
| 621 |
-
],
|
| 622 |
-
"description": "The following are multiple choice questions (with answers) about college biology.\n\n",
|
| 623 |
-
"target_delimiter": " ",
|
| 624 |
-
"fewshot_delimiter": "\n\n",
|
| 625 |
-
"fewshot_config": {
|
| 626 |
-
"sampler": "first_n"
|
| 627 |
-
},
|
| 628 |
-
"num_fewshot": 5,
|
| 629 |
-
"metric_list": [
|
| 630 |
-
{
|
| 631 |
-
"metric": "acc",
|
| 632 |
-
"aggregation": "mean",
|
| 633 |
-
"higher_is_better": true
|
| 634 |
-
}
|
| 635 |
-
],
|
| 636 |
-
"output_type": "multiple_choice",
|
| 637 |
-
"repeats": 1,
|
| 638 |
-
"should_decontaminate": false,
|
| 639 |
-
"metadata": {
|
| 640 |
-
"version": 0.0
|
| 641 |
-
}
|
| 642 |
-
},
|
| 643 |
-
"mmlu_college_chemistry": {
|
| 644 |
-
"task": "mmlu_college_chemistry",
|
| 645 |
-
"task_alias": "college_chemistry",
|
| 646 |
-
"group": "mmlu_stem",
|
| 647 |
-
"group_alias": "stem",
|
| 648 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 649 |
-
"dataset_name": "college_chemistry",
|
| 650 |
-
"test_split": "test",
|
| 651 |
-
"fewshot_split": "dev",
|
| 652 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 653 |
-
"doc_to_target": "answer",
|
| 654 |
-
"doc_to_choice": [
|
| 655 |
-
"A",
|
| 656 |
-
"B",
|
| 657 |
-
"C",
|
| 658 |
-
"D"
|
| 659 |
-
],
|
| 660 |
-
"description": "The following are multiple choice questions (with answers) about college chemistry.\n\n",
|
| 661 |
-
"target_delimiter": " ",
|
| 662 |
-
"fewshot_delimiter": "\n\n",
|
| 663 |
-
"fewshot_config": {
|
| 664 |
-
"sampler": "first_n"
|
| 665 |
-
},
|
| 666 |
-
"num_fewshot": 5,
|
| 667 |
-
"metric_list": [
|
| 668 |
-
{
|
| 669 |
-
"metric": "acc",
|
| 670 |
-
"aggregation": "mean",
|
| 671 |
-
"higher_is_better": true
|
| 672 |
-
}
|
| 673 |
-
],
|
| 674 |
-
"output_type": "multiple_choice",
|
| 675 |
-
"repeats": 1,
|
| 676 |
-
"should_decontaminate": false,
|
| 677 |
-
"metadata": {
|
| 678 |
-
"version": 0.0
|
| 679 |
-
}
|
| 680 |
-
},
|
| 681 |
-
"mmlu_college_computer_science": {
|
| 682 |
-
"task": "mmlu_college_computer_science",
|
| 683 |
-
"task_alias": "college_computer_science",
|
| 684 |
-
"group": "mmlu_stem",
|
| 685 |
-
"group_alias": "stem",
|
| 686 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 687 |
-
"dataset_name": "college_computer_science",
|
| 688 |
-
"test_split": "test",
|
| 689 |
-
"fewshot_split": "dev",
|
| 690 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 691 |
-
"doc_to_target": "answer",
|
| 692 |
-
"doc_to_choice": [
|
| 693 |
-
"A",
|
| 694 |
-
"B",
|
| 695 |
-
"C",
|
| 696 |
-
"D"
|
| 697 |
-
],
|
| 698 |
-
"description": "The following are multiple choice questions (with answers) about college computer science.\n\n",
|
| 699 |
-
"target_delimiter": " ",
|
| 700 |
-
"fewshot_delimiter": "\n\n",
|
| 701 |
-
"fewshot_config": {
|
| 702 |
-
"sampler": "first_n"
|
| 703 |
-
},
|
| 704 |
-
"num_fewshot": 5,
|
| 705 |
-
"metric_list": [
|
| 706 |
-
{
|
| 707 |
-
"metric": "acc",
|
| 708 |
-
"aggregation": "mean",
|
| 709 |
-
"higher_is_better": true
|
| 710 |
-
}
|
| 711 |
-
],
|
| 712 |
-
"output_type": "multiple_choice",
|
| 713 |
-
"repeats": 1,
|
| 714 |
-
"should_decontaminate": false,
|
| 715 |
-
"metadata": {
|
| 716 |
-
"version": 0.0
|
| 717 |
-
}
|
| 718 |
-
},
|
| 719 |
-
"mmlu_college_mathematics": {
|
| 720 |
-
"task": "mmlu_college_mathematics",
|
| 721 |
-
"task_alias": "college_mathematics",
|
| 722 |
-
"group": "mmlu_stem",
|
| 723 |
-
"group_alias": "stem",
|
| 724 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 725 |
-
"dataset_name": "college_mathematics",
|
| 726 |
-
"test_split": "test",
|
| 727 |
-
"fewshot_split": "dev",
|
| 728 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 729 |
-
"doc_to_target": "answer",
|
| 730 |
-
"doc_to_choice": [
|
| 731 |
-
"A",
|
| 732 |
-
"B",
|
| 733 |
-
"C",
|
| 734 |
-
"D"
|
| 735 |
-
],
|
| 736 |
-
"description": "The following are multiple choice questions (with answers) about college mathematics.\n\n",
|
| 737 |
-
"target_delimiter": " ",
|
| 738 |
-
"fewshot_delimiter": "\n\n",
|
| 739 |
-
"fewshot_config": {
|
| 740 |
-
"sampler": "first_n"
|
| 741 |
-
},
|
| 742 |
-
"num_fewshot": 5,
|
| 743 |
-
"metric_list": [
|
| 744 |
-
{
|
| 745 |
-
"metric": "acc",
|
| 746 |
-
"aggregation": "mean",
|
| 747 |
-
"higher_is_better": true
|
| 748 |
-
}
|
| 749 |
-
],
|
| 750 |
-
"output_type": "multiple_choice",
|
| 751 |
-
"repeats": 1,
|
| 752 |
-
"should_decontaminate": false,
|
| 753 |
-
"metadata": {
|
| 754 |
-
"version": 0.0
|
| 755 |
-
}
|
| 756 |
-
},
|
| 757 |
-
"mmlu_college_medicine": {
|
| 758 |
-
"task": "mmlu_college_medicine",
|
| 759 |
-
"task_alias": "college_medicine",
|
| 760 |
-
"group": "mmlu_other",
|
| 761 |
-
"group_alias": "other",
|
| 762 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 763 |
-
"dataset_name": "college_medicine",
|
| 764 |
-
"test_split": "test",
|
| 765 |
-
"fewshot_split": "dev",
|
| 766 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 767 |
-
"doc_to_target": "answer",
|
| 768 |
-
"doc_to_choice": [
|
| 769 |
-
"A",
|
| 770 |
-
"B",
|
| 771 |
-
"C",
|
| 772 |
-
"D"
|
| 773 |
-
],
|
| 774 |
-
"description": "The following are multiple choice questions (with answers) about college medicine.\n\n",
|
| 775 |
-
"target_delimiter": " ",
|
| 776 |
-
"fewshot_delimiter": "\n\n",
|
| 777 |
-
"fewshot_config": {
|
| 778 |
-
"sampler": "first_n"
|
| 779 |
-
},
|
| 780 |
-
"num_fewshot": 5,
|
| 781 |
-
"metric_list": [
|
| 782 |
-
{
|
| 783 |
-
"metric": "acc",
|
| 784 |
-
"aggregation": "mean",
|
| 785 |
-
"higher_is_better": true
|
| 786 |
-
}
|
| 787 |
-
],
|
| 788 |
-
"output_type": "multiple_choice",
|
| 789 |
-
"repeats": 1,
|
| 790 |
-
"should_decontaminate": false,
|
| 791 |
-
"metadata": {
|
| 792 |
-
"version": 0.0
|
| 793 |
-
}
|
| 794 |
-
},
|
| 795 |
-
"mmlu_college_physics": {
|
| 796 |
-
"task": "mmlu_college_physics",
|
| 797 |
-
"task_alias": "college_physics",
|
| 798 |
-
"group": "mmlu_stem",
|
| 799 |
-
"group_alias": "stem",
|
| 800 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 801 |
-
"dataset_name": "college_physics",
|
| 802 |
-
"test_split": "test",
|
| 803 |
-
"fewshot_split": "dev",
|
| 804 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 805 |
-
"doc_to_target": "answer",
|
| 806 |
-
"doc_to_choice": [
|
| 807 |
-
"A",
|
| 808 |
-
"B",
|
| 809 |
-
"C",
|
| 810 |
-
"D"
|
| 811 |
-
],
|
| 812 |
-
"description": "The following are multiple choice questions (with answers) about college physics.\n\n",
|
| 813 |
-
"target_delimiter": " ",
|
| 814 |
-
"fewshot_delimiter": "\n\n",
|
| 815 |
-
"fewshot_config": {
|
| 816 |
-
"sampler": "first_n"
|
| 817 |
-
},
|
| 818 |
-
"num_fewshot": 5,
|
| 819 |
-
"metric_list": [
|
| 820 |
-
{
|
| 821 |
-
"metric": "acc",
|
| 822 |
-
"aggregation": "mean",
|
| 823 |
-
"higher_is_better": true
|
| 824 |
-
}
|
| 825 |
-
],
|
| 826 |
-
"output_type": "multiple_choice",
|
| 827 |
-
"repeats": 1,
|
| 828 |
-
"should_decontaminate": false,
|
| 829 |
-
"metadata": {
|
| 830 |
-
"version": 0.0
|
| 831 |
-
}
|
| 832 |
-
},
|
| 833 |
-
"mmlu_computer_security": {
|
| 834 |
-
"task": "mmlu_computer_security",
|
| 835 |
-
"task_alias": "computer_security",
|
| 836 |
-
"group": "mmlu_stem",
|
| 837 |
-
"group_alias": "stem",
|
| 838 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 839 |
-
"dataset_name": "computer_security",
|
| 840 |
-
"test_split": "test",
|
| 841 |
-
"fewshot_split": "dev",
|
| 842 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 843 |
-
"doc_to_target": "answer",
|
| 844 |
-
"doc_to_choice": [
|
| 845 |
-
"A",
|
| 846 |
-
"B",
|
| 847 |
-
"C",
|
| 848 |
-
"D"
|
| 849 |
-
],
|
| 850 |
-
"description": "The following are multiple choice questions (with answers) about computer security.\n\n",
|
| 851 |
-
"target_delimiter": " ",
|
| 852 |
-
"fewshot_delimiter": "\n\n",
|
| 853 |
-
"fewshot_config": {
|
| 854 |
-
"sampler": "first_n"
|
| 855 |
-
},
|
| 856 |
-
"num_fewshot": 5,
|
| 857 |
-
"metric_list": [
|
| 858 |
-
{
|
| 859 |
-
"metric": "acc",
|
| 860 |
-
"aggregation": "mean",
|
| 861 |
-
"higher_is_better": true
|
| 862 |
-
}
|
| 863 |
-
],
|
| 864 |
-
"output_type": "multiple_choice",
|
| 865 |
-
"repeats": 1,
|
| 866 |
-
"should_decontaminate": false,
|
| 867 |
-
"metadata": {
|
| 868 |
-
"version": 0.0
|
| 869 |
-
}
|
| 870 |
-
},
|
| 871 |
-
"mmlu_conceptual_physics": {
|
| 872 |
-
"task": "mmlu_conceptual_physics",
|
| 873 |
-
"task_alias": "conceptual_physics",
|
| 874 |
-
"group": "mmlu_stem",
|
| 875 |
-
"group_alias": "stem",
|
| 876 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 877 |
-
"dataset_name": "conceptual_physics",
|
| 878 |
-
"test_split": "test",
|
| 879 |
-
"fewshot_split": "dev",
|
| 880 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 881 |
-
"doc_to_target": "answer",
|
| 882 |
-
"doc_to_choice": [
|
| 883 |
-
"A",
|
| 884 |
-
"B",
|
| 885 |
-
"C",
|
| 886 |
-
"D"
|
| 887 |
-
],
|
| 888 |
-
"description": "The following are multiple choice questions (with answers) about conceptual physics.\n\n",
|
| 889 |
-
"target_delimiter": " ",
|
| 890 |
-
"fewshot_delimiter": "\n\n",
|
| 891 |
-
"fewshot_config": {
|
| 892 |
-
"sampler": "first_n"
|
| 893 |
-
},
|
| 894 |
-
"num_fewshot": 5,
|
| 895 |
-
"metric_list": [
|
| 896 |
-
{
|
| 897 |
-
"metric": "acc",
|
| 898 |
-
"aggregation": "mean",
|
| 899 |
-
"higher_is_better": true
|
| 900 |
-
}
|
| 901 |
-
],
|
| 902 |
-
"output_type": "multiple_choice",
|
| 903 |
-
"repeats": 1,
|
| 904 |
-
"should_decontaminate": false,
|
| 905 |
-
"metadata": {
|
| 906 |
-
"version": 0.0
|
| 907 |
-
}
|
| 908 |
-
},
|
| 909 |
-
"mmlu_econometrics": {
|
| 910 |
-
"task": "mmlu_econometrics",
|
| 911 |
-
"task_alias": "econometrics",
|
| 912 |
-
"group": "mmlu_social_sciences",
|
| 913 |
-
"group_alias": "social_sciences",
|
| 914 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 915 |
-
"dataset_name": "econometrics",
|
| 916 |
-
"test_split": "test",
|
| 917 |
-
"fewshot_split": "dev",
|
| 918 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 919 |
-
"doc_to_target": "answer",
|
| 920 |
-
"doc_to_choice": [
|
| 921 |
-
"A",
|
| 922 |
-
"B",
|
| 923 |
-
"C",
|
| 924 |
-
"D"
|
| 925 |
-
],
|
| 926 |
-
"description": "The following are multiple choice questions (with answers) about econometrics.\n\n",
|
| 927 |
-
"target_delimiter": " ",
|
| 928 |
-
"fewshot_delimiter": "\n\n",
|
| 929 |
-
"fewshot_config": {
|
| 930 |
-
"sampler": "first_n"
|
| 931 |
-
},
|
| 932 |
-
"num_fewshot": 5,
|
| 933 |
-
"metric_list": [
|
| 934 |
-
{
|
| 935 |
-
"metric": "acc",
|
| 936 |
-
"aggregation": "mean",
|
| 937 |
-
"higher_is_better": true
|
| 938 |
-
}
|
| 939 |
-
],
|
| 940 |
-
"output_type": "multiple_choice",
|
| 941 |
-
"repeats": 1,
|
| 942 |
-
"should_decontaminate": false,
|
| 943 |
-
"metadata": {
|
| 944 |
-
"version": 0.0
|
| 945 |
-
}
|
| 946 |
-
},
|
| 947 |
-
"mmlu_electrical_engineering": {
|
| 948 |
-
"task": "mmlu_electrical_engineering",
|
| 949 |
-
"task_alias": "electrical_engineering",
|
| 950 |
-
"group": "mmlu_stem",
|
| 951 |
-
"group_alias": "stem",
|
| 952 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 953 |
-
"dataset_name": "electrical_engineering",
|
| 954 |
-
"test_split": "test",
|
| 955 |
-
"fewshot_split": "dev",
|
| 956 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 957 |
-
"doc_to_target": "answer",
|
| 958 |
-
"doc_to_choice": [
|
| 959 |
-
"A",
|
| 960 |
-
"B",
|
| 961 |
-
"C",
|
| 962 |
-
"D"
|
| 963 |
-
],
|
| 964 |
-
"description": "The following are multiple choice questions (with answers) about electrical engineering.\n\n",
|
| 965 |
-
"target_delimiter": " ",
|
| 966 |
-
"fewshot_delimiter": "\n\n",
|
| 967 |
-
"fewshot_config": {
|
| 968 |
-
"sampler": "first_n"
|
| 969 |
-
},
|
| 970 |
-
"num_fewshot": 5,
|
| 971 |
-
"metric_list": [
|
| 972 |
-
{
|
| 973 |
-
"metric": "acc",
|
| 974 |
-
"aggregation": "mean",
|
| 975 |
-
"higher_is_better": true
|
| 976 |
-
}
|
| 977 |
-
],
|
| 978 |
-
"output_type": "multiple_choice",
|
| 979 |
-
"repeats": 1,
|
| 980 |
-
"should_decontaminate": false,
|
| 981 |
-
"metadata": {
|
| 982 |
-
"version": 0.0
|
| 983 |
-
}
|
| 984 |
-
},
|
| 985 |
-
"mmlu_elementary_mathematics": {
|
| 986 |
-
"task": "mmlu_elementary_mathematics",
|
| 987 |
-
"task_alias": "elementary_mathematics",
|
| 988 |
-
"group": "mmlu_stem",
|
| 989 |
-
"group_alias": "stem",
|
| 990 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 991 |
-
"dataset_name": "elementary_mathematics",
|
| 992 |
-
"test_split": "test",
|
| 993 |
-
"fewshot_split": "dev",
|
| 994 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 995 |
-
"doc_to_target": "answer",
|
| 996 |
-
"doc_to_choice": [
|
| 997 |
-
"A",
|
| 998 |
-
"B",
|
| 999 |
-
"C",
|
| 1000 |
-
"D"
|
| 1001 |
-
],
|
| 1002 |
-
"description": "The following are multiple choice questions (with answers) about elementary mathematics.\n\n",
|
| 1003 |
-
"target_delimiter": " ",
|
| 1004 |
-
"fewshot_delimiter": "\n\n",
|
| 1005 |
-
"fewshot_config": {
|
| 1006 |
-
"sampler": "first_n"
|
| 1007 |
-
},
|
| 1008 |
-
"num_fewshot": 5,
|
| 1009 |
-
"metric_list": [
|
| 1010 |
-
{
|
| 1011 |
-
"metric": "acc",
|
| 1012 |
-
"aggregation": "mean",
|
| 1013 |
-
"higher_is_better": true
|
| 1014 |
-
}
|
| 1015 |
-
],
|
| 1016 |
-
"output_type": "multiple_choice",
|
| 1017 |
-
"repeats": 1,
|
| 1018 |
-
"should_decontaminate": false,
|
| 1019 |
-
"metadata": {
|
| 1020 |
-
"version": 0.0
|
| 1021 |
-
}
|
| 1022 |
-
},
|
| 1023 |
-
"mmlu_formal_logic": {
|
| 1024 |
-
"task": "mmlu_formal_logic",
|
| 1025 |
-
"task_alias": "formal_logic",
|
| 1026 |
-
"group": "mmlu_humanities",
|
| 1027 |
-
"group_alias": "humanities",
|
| 1028 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1029 |
-
"dataset_name": "formal_logic",
|
| 1030 |
-
"test_split": "test",
|
| 1031 |
-
"fewshot_split": "dev",
|
| 1032 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1033 |
-
"doc_to_target": "answer",
|
| 1034 |
-
"doc_to_choice": [
|
| 1035 |
-
"A",
|
| 1036 |
-
"B",
|
| 1037 |
-
"C",
|
| 1038 |
-
"D"
|
| 1039 |
-
],
|
| 1040 |
-
"description": "The following are multiple choice questions (with answers) about formal logic.\n\n",
|
| 1041 |
-
"target_delimiter": " ",
|
| 1042 |
-
"fewshot_delimiter": "\n\n",
|
| 1043 |
-
"fewshot_config": {
|
| 1044 |
-
"sampler": "first_n"
|
| 1045 |
-
},
|
| 1046 |
-
"num_fewshot": 5,
|
| 1047 |
-
"metric_list": [
|
| 1048 |
-
{
|
| 1049 |
-
"metric": "acc",
|
| 1050 |
-
"aggregation": "mean",
|
| 1051 |
-
"higher_is_better": true
|
| 1052 |
-
}
|
| 1053 |
-
],
|
| 1054 |
-
"output_type": "multiple_choice",
|
| 1055 |
-
"repeats": 1,
|
| 1056 |
-
"should_decontaminate": false,
|
| 1057 |
-
"metadata": {
|
| 1058 |
-
"version": 0.0
|
| 1059 |
-
}
|
| 1060 |
-
},
|
| 1061 |
-
"mmlu_global_facts": {
|
| 1062 |
-
"task": "mmlu_global_facts",
|
| 1063 |
-
"task_alias": "global_facts",
|
| 1064 |
-
"group": "mmlu_other",
|
| 1065 |
-
"group_alias": "other",
|
| 1066 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1067 |
-
"dataset_name": "global_facts",
|
| 1068 |
-
"test_split": "test",
|
| 1069 |
-
"fewshot_split": "dev",
|
| 1070 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1071 |
-
"doc_to_target": "answer",
|
| 1072 |
-
"doc_to_choice": [
|
| 1073 |
-
"A",
|
| 1074 |
-
"B",
|
| 1075 |
-
"C",
|
| 1076 |
-
"D"
|
| 1077 |
-
],
|
| 1078 |
-
"description": "The following are multiple choice questions (with answers) about global facts.\n\n",
|
| 1079 |
-
"target_delimiter": " ",
|
| 1080 |
-
"fewshot_delimiter": "\n\n",
|
| 1081 |
-
"fewshot_config": {
|
| 1082 |
-
"sampler": "first_n"
|
| 1083 |
-
},
|
| 1084 |
-
"num_fewshot": 5,
|
| 1085 |
-
"metric_list": [
|
| 1086 |
-
{
|
| 1087 |
-
"metric": "acc",
|
| 1088 |
-
"aggregation": "mean",
|
| 1089 |
-
"higher_is_better": true
|
| 1090 |
-
}
|
| 1091 |
-
],
|
| 1092 |
-
"output_type": "multiple_choice",
|
| 1093 |
-
"repeats": 1,
|
| 1094 |
-
"should_decontaminate": false,
|
| 1095 |
-
"metadata": {
|
| 1096 |
-
"version": 0.0
|
| 1097 |
-
}
|
| 1098 |
-
},
|
| 1099 |
-
"mmlu_high_school_biology": {
|
| 1100 |
-
"task": "mmlu_high_school_biology",
|
| 1101 |
-
"task_alias": "high_school_biology",
|
| 1102 |
-
"group": "mmlu_stem",
|
| 1103 |
-
"group_alias": "stem",
|
| 1104 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1105 |
-
"dataset_name": "high_school_biology",
|
| 1106 |
-
"test_split": "test",
|
| 1107 |
-
"fewshot_split": "dev",
|
| 1108 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1109 |
-
"doc_to_target": "answer",
|
| 1110 |
-
"doc_to_choice": [
|
| 1111 |
-
"A",
|
| 1112 |
-
"B",
|
| 1113 |
-
"C",
|
| 1114 |
-
"D"
|
| 1115 |
-
],
|
| 1116 |
-
"description": "The following are multiple choice questions (with answers) about high school biology.\n\n",
|
| 1117 |
-
"target_delimiter": " ",
|
| 1118 |
-
"fewshot_delimiter": "\n\n",
|
| 1119 |
-
"fewshot_config": {
|
| 1120 |
-
"sampler": "first_n"
|
| 1121 |
-
},
|
| 1122 |
-
"num_fewshot": 5,
|
| 1123 |
-
"metric_list": [
|
| 1124 |
-
{
|
| 1125 |
-
"metric": "acc",
|
| 1126 |
-
"aggregation": "mean",
|
| 1127 |
-
"higher_is_better": true
|
| 1128 |
-
}
|
| 1129 |
-
],
|
| 1130 |
-
"output_type": "multiple_choice",
|
| 1131 |
-
"repeats": 1,
|
| 1132 |
-
"should_decontaminate": false,
|
| 1133 |
-
"metadata": {
|
| 1134 |
-
"version": 0.0
|
| 1135 |
-
}
|
| 1136 |
-
},
|
| 1137 |
-
"mmlu_high_school_chemistry": {
|
| 1138 |
-
"task": "mmlu_high_school_chemistry",
|
| 1139 |
-
"task_alias": "high_school_chemistry",
|
| 1140 |
-
"group": "mmlu_stem",
|
| 1141 |
-
"group_alias": "stem",
|
| 1142 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1143 |
-
"dataset_name": "high_school_chemistry",
|
| 1144 |
-
"test_split": "test",
|
| 1145 |
-
"fewshot_split": "dev",
|
| 1146 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1147 |
-
"doc_to_target": "answer",
|
| 1148 |
-
"doc_to_choice": [
|
| 1149 |
-
"A",
|
| 1150 |
-
"B",
|
| 1151 |
-
"C",
|
| 1152 |
-
"D"
|
| 1153 |
-
],
|
| 1154 |
-
"description": "The following are multiple choice questions (with answers) about high school chemistry.\n\n",
|
| 1155 |
-
"target_delimiter": " ",
|
| 1156 |
-
"fewshot_delimiter": "\n\n",
|
| 1157 |
-
"fewshot_config": {
|
| 1158 |
-
"sampler": "first_n"
|
| 1159 |
-
},
|
| 1160 |
-
"num_fewshot": 5,
|
| 1161 |
-
"metric_list": [
|
| 1162 |
-
{
|
| 1163 |
-
"metric": "acc",
|
| 1164 |
-
"aggregation": "mean",
|
| 1165 |
-
"higher_is_better": true
|
| 1166 |
-
}
|
| 1167 |
-
],
|
| 1168 |
-
"output_type": "multiple_choice",
|
| 1169 |
-
"repeats": 1,
|
| 1170 |
-
"should_decontaminate": false,
|
| 1171 |
-
"metadata": {
|
| 1172 |
-
"version": 0.0
|
| 1173 |
-
}
|
| 1174 |
-
},
|
| 1175 |
-
"mmlu_high_school_computer_science": {
|
| 1176 |
-
"task": "mmlu_high_school_computer_science",
|
| 1177 |
-
"task_alias": "high_school_computer_science",
|
| 1178 |
-
"group": "mmlu_stem",
|
| 1179 |
-
"group_alias": "stem",
|
| 1180 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1181 |
-
"dataset_name": "high_school_computer_science",
|
| 1182 |
-
"test_split": "test",
|
| 1183 |
-
"fewshot_split": "dev",
|
| 1184 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1185 |
-
"doc_to_target": "answer",
|
| 1186 |
-
"doc_to_choice": [
|
| 1187 |
-
"A",
|
| 1188 |
-
"B",
|
| 1189 |
-
"C",
|
| 1190 |
-
"D"
|
| 1191 |
-
],
|
| 1192 |
-
"description": "The following are multiple choice questions (with answers) about high school computer science.\n\n",
|
| 1193 |
-
"target_delimiter": " ",
|
| 1194 |
-
"fewshot_delimiter": "\n\n",
|
| 1195 |
-
"fewshot_config": {
|
| 1196 |
-
"sampler": "first_n"
|
| 1197 |
-
},
|
| 1198 |
-
"num_fewshot": 5,
|
| 1199 |
-
"metric_list": [
|
| 1200 |
-
{
|
| 1201 |
-
"metric": "acc",
|
| 1202 |
-
"aggregation": "mean",
|
| 1203 |
-
"higher_is_better": true
|
| 1204 |
-
}
|
| 1205 |
-
],
|
| 1206 |
-
"output_type": "multiple_choice",
|
| 1207 |
-
"repeats": 1,
|
| 1208 |
-
"should_decontaminate": false,
|
| 1209 |
-
"metadata": {
|
| 1210 |
-
"version": 0.0
|
| 1211 |
-
}
|
| 1212 |
-
},
|
| 1213 |
-
"mmlu_high_school_european_history": {
|
| 1214 |
-
"task": "mmlu_high_school_european_history",
|
| 1215 |
-
"task_alias": "high_school_european_history",
|
| 1216 |
-
"group": "mmlu_humanities",
|
| 1217 |
-
"group_alias": "humanities",
|
| 1218 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1219 |
-
"dataset_name": "high_school_european_history",
|
| 1220 |
-
"test_split": "test",
|
| 1221 |
-
"fewshot_split": "dev",
|
| 1222 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1223 |
-
"doc_to_target": "answer",
|
| 1224 |
-
"doc_to_choice": [
|
| 1225 |
-
"A",
|
| 1226 |
-
"B",
|
| 1227 |
-
"C",
|
| 1228 |
-
"D"
|
| 1229 |
-
],
|
| 1230 |
-
"description": "The following are multiple choice questions (with answers) about high school european history.\n\n",
|
| 1231 |
-
"target_delimiter": " ",
|
| 1232 |
-
"fewshot_delimiter": "\n\n",
|
| 1233 |
-
"fewshot_config": {
|
| 1234 |
-
"sampler": "first_n"
|
| 1235 |
-
},
|
| 1236 |
-
"num_fewshot": 5,
|
| 1237 |
-
"metric_list": [
|
| 1238 |
-
{
|
| 1239 |
-
"metric": "acc",
|
| 1240 |
-
"aggregation": "mean",
|
| 1241 |
-
"higher_is_better": true
|
| 1242 |
-
}
|
| 1243 |
-
],
|
| 1244 |
-
"output_type": "multiple_choice",
|
| 1245 |
-
"repeats": 1,
|
| 1246 |
-
"should_decontaminate": false,
|
| 1247 |
-
"metadata": {
|
| 1248 |
-
"version": 0.0
|
| 1249 |
-
}
|
| 1250 |
-
},
|
| 1251 |
-
"mmlu_high_school_geography": {
|
| 1252 |
-
"task": "mmlu_high_school_geography",
|
| 1253 |
-
"task_alias": "high_school_geography",
|
| 1254 |
-
"group": "mmlu_social_sciences",
|
| 1255 |
-
"group_alias": "social_sciences",
|
| 1256 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1257 |
-
"dataset_name": "high_school_geography",
|
| 1258 |
-
"test_split": "test",
|
| 1259 |
-
"fewshot_split": "dev",
|
| 1260 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1261 |
-
"doc_to_target": "answer",
|
| 1262 |
-
"doc_to_choice": [
|
| 1263 |
-
"A",
|
| 1264 |
-
"B",
|
| 1265 |
-
"C",
|
| 1266 |
-
"D"
|
| 1267 |
-
],
|
| 1268 |
-
"description": "The following are multiple choice questions (with answers) about high school geography.\n\n",
|
| 1269 |
-
"target_delimiter": " ",
|
| 1270 |
-
"fewshot_delimiter": "\n\n",
|
| 1271 |
-
"fewshot_config": {
|
| 1272 |
-
"sampler": "first_n"
|
| 1273 |
-
},
|
| 1274 |
-
"num_fewshot": 5,
|
| 1275 |
-
"metric_list": [
|
| 1276 |
-
{
|
| 1277 |
-
"metric": "acc",
|
| 1278 |
-
"aggregation": "mean",
|
| 1279 |
-
"higher_is_better": true
|
| 1280 |
-
}
|
| 1281 |
-
],
|
| 1282 |
-
"output_type": "multiple_choice",
|
| 1283 |
-
"repeats": 1,
|
| 1284 |
-
"should_decontaminate": false,
|
| 1285 |
-
"metadata": {
|
| 1286 |
-
"version": 0.0
|
| 1287 |
-
}
|
| 1288 |
-
},
|
| 1289 |
-
"mmlu_high_school_government_and_politics": {
|
| 1290 |
-
"task": "mmlu_high_school_government_and_politics",
|
| 1291 |
-
"task_alias": "high_school_government_and_politics",
|
| 1292 |
-
"group": "mmlu_social_sciences",
|
| 1293 |
-
"group_alias": "social_sciences",
|
| 1294 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1295 |
-
"dataset_name": "high_school_government_and_politics",
|
| 1296 |
-
"test_split": "test",
|
| 1297 |
-
"fewshot_split": "dev",
|
| 1298 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1299 |
-
"doc_to_target": "answer",
|
| 1300 |
-
"doc_to_choice": [
|
| 1301 |
-
"A",
|
| 1302 |
-
"B",
|
| 1303 |
-
"C",
|
| 1304 |
-
"D"
|
| 1305 |
-
],
|
| 1306 |
-
"description": "The following are multiple choice questions (with answers) about high school government and politics.\n\n",
|
| 1307 |
-
"target_delimiter": " ",
|
| 1308 |
-
"fewshot_delimiter": "\n\n",
|
| 1309 |
-
"fewshot_config": {
|
| 1310 |
-
"sampler": "first_n"
|
| 1311 |
-
},
|
| 1312 |
-
"num_fewshot": 5,
|
| 1313 |
-
"metric_list": [
|
| 1314 |
-
{
|
| 1315 |
-
"metric": "acc",
|
| 1316 |
-
"aggregation": "mean",
|
| 1317 |
-
"higher_is_better": true
|
| 1318 |
-
}
|
| 1319 |
-
],
|
| 1320 |
-
"output_type": "multiple_choice",
|
| 1321 |
-
"repeats": 1,
|
| 1322 |
-
"should_decontaminate": false,
|
| 1323 |
-
"metadata": {
|
| 1324 |
-
"version": 0.0
|
| 1325 |
-
}
|
| 1326 |
-
},
|
| 1327 |
-
"mmlu_high_school_macroeconomics": {
|
| 1328 |
-
"task": "mmlu_high_school_macroeconomics",
|
| 1329 |
-
"task_alias": "high_school_macroeconomics",
|
| 1330 |
-
"group": "mmlu_social_sciences",
|
| 1331 |
-
"group_alias": "social_sciences",
|
| 1332 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1333 |
-
"dataset_name": "high_school_macroeconomics",
|
| 1334 |
-
"test_split": "test",
|
| 1335 |
-
"fewshot_split": "dev",
|
| 1336 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1337 |
-
"doc_to_target": "answer",
|
| 1338 |
-
"doc_to_choice": [
|
| 1339 |
-
"A",
|
| 1340 |
-
"B",
|
| 1341 |
-
"C",
|
| 1342 |
-
"D"
|
| 1343 |
-
],
|
| 1344 |
-
"description": "The following are multiple choice questions (with answers) about high school macroeconomics.\n\n",
|
| 1345 |
-
"target_delimiter": " ",
|
| 1346 |
-
"fewshot_delimiter": "\n\n",
|
| 1347 |
-
"fewshot_config": {
|
| 1348 |
-
"sampler": "first_n"
|
| 1349 |
-
},
|
| 1350 |
-
"num_fewshot": 5,
|
| 1351 |
-
"metric_list": [
|
| 1352 |
-
{
|
| 1353 |
-
"metric": "acc",
|
| 1354 |
-
"aggregation": "mean",
|
| 1355 |
-
"higher_is_better": true
|
| 1356 |
-
}
|
| 1357 |
-
],
|
| 1358 |
-
"output_type": "multiple_choice",
|
| 1359 |
-
"repeats": 1,
|
| 1360 |
-
"should_decontaminate": false,
|
| 1361 |
-
"metadata": {
|
| 1362 |
-
"version": 0.0
|
| 1363 |
-
}
|
| 1364 |
-
},
|
| 1365 |
-
"mmlu_high_school_mathematics": {
|
| 1366 |
-
"task": "mmlu_high_school_mathematics",
|
| 1367 |
-
"task_alias": "high_school_mathematics",
|
| 1368 |
-
"group": "mmlu_stem",
|
| 1369 |
-
"group_alias": "stem",
|
| 1370 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1371 |
-
"dataset_name": "high_school_mathematics",
|
| 1372 |
-
"test_split": "test",
|
| 1373 |
-
"fewshot_split": "dev",
|
| 1374 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1375 |
-
"doc_to_target": "answer",
|
| 1376 |
-
"doc_to_choice": [
|
| 1377 |
-
"A",
|
| 1378 |
-
"B",
|
| 1379 |
-
"C",
|
| 1380 |
-
"D"
|
| 1381 |
-
],
|
| 1382 |
-
"description": "The following are multiple choice questions (with answers) about high school mathematics.\n\n",
|
| 1383 |
-
"target_delimiter": " ",
|
| 1384 |
-
"fewshot_delimiter": "\n\n",
|
| 1385 |
-
"fewshot_config": {
|
| 1386 |
-
"sampler": "first_n"
|
| 1387 |
-
},
|
| 1388 |
-
"num_fewshot": 5,
|
| 1389 |
-
"metric_list": [
|
| 1390 |
-
{
|
| 1391 |
-
"metric": "acc",
|
| 1392 |
-
"aggregation": "mean",
|
| 1393 |
-
"higher_is_better": true
|
| 1394 |
-
}
|
| 1395 |
-
],
|
| 1396 |
-
"output_type": "multiple_choice",
|
| 1397 |
-
"repeats": 1,
|
| 1398 |
-
"should_decontaminate": false,
|
| 1399 |
-
"metadata": {
|
| 1400 |
-
"version": 0.0
|
| 1401 |
-
}
|
| 1402 |
-
},
|
| 1403 |
-
"mmlu_high_school_microeconomics": {
|
| 1404 |
-
"task": "mmlu_high_school_microeconomics",
|
| 1405 |
-
"task_alias": "high_school_microeconomics",
|
| 1406 |
-
"group": "mmlu_social_sciences",
|
| 1407 |
-
"group_alias": "social_sciences",
|
| 1408 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1409 |
-
"dataset_name": "high_school_microeconomics",
|
| 1410 |
-
"test_split": "test",
|
| 1411 |
-
"fewshot_split": "dev",
|
| 1412 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1413 |
-
"doc_to_target": "answer",
|
| 1414 |
-
"doc_to_choice": [
|
| 1415 |
-
"A",
|
| 1416 |
-
"B",
|
| 1417 |
-
"C",
|
| 1418 |
-
"D"
|
| 1419 |
-
],
|
| 1420 |
-
"description": "The following are multiple choice questions (with answers) about high school microeconomics.\n\n",
|
| 1421 |
-
"target_delimiter": " ",
|
| 1422 |
-
"fewshot_delimiter": "\n\n",
|
| 1423 |
-
"fewshot_config": {
|
| 1424 |
-
"sampler": "first_n"
|
| 1425 |
-
},
|
| 1426 |
-
"num_fewshot": 5,
|
| 1427 |
-
"metric_list": [
|
| 1428 |
-
{
|
| 1429 |
-
"metric": "acc",
|
| 1430 |
-
"aggregation": "mean",
|
| 1431 |
-
"higher_is_better": true
|
| 1432 |
-
}
|
| 1433 |
-
],
|
| 1434 |
-
"output_type": "multiple_choice",
|
| 1435 |
-
"repeats": 1,
|
| 1436 |
-
"should_decontaminate": false,
|
| 1437 |
-
"metadata": {
|
| 1438 |
-
"version": 0.0
|
| 1439 |
-
}
|
| 1440 |
-
},
|
| 1441 |
-
"mmlu_high_school_physics": {
|
| 1442 |
-
"task": "mmlu_high_school_physics",
|
| 1443 |
-
"task_alias": "high_school_physics",
|
| 1444 |
-
"group": "mmlu_stem",
|
| 1445 |
-
"group_alias": "stem",
|
| 1446 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1447 |
-
"dataset_name": "high_school_physics",
|
| 1448 |
-
"test_split": "test",
|
| 1449 |
-
"fewshot_split": "dev",
|
| 1450 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1451 |
-
"doc_to_target": "answer",
|
| 1452 |
-
"doc_to_choice": [
|
| 1453 |
-
"A",
|
| 1454 |
-
"B",
|
| 1455 |
-
"C",
|
| 1456 |
-
"D"
|
| 1457 |
-
],
|
| 1458 |
-
"description": "The following are multiple choice questions (with answers) about high school physics.\n\n",
|
| 1459 |
-
"target_delimiter": " ",
|
| 1460 |
-
"fewshot_delimiter": "\n\n",
|
| 1461 |
-
"fewshot_config": {
|
| 1462 |
-
"sampler": "first_n"
|
| 1463 |
-
},
|
| 1464 |
-
"num_fewshot": 5,
|
| 1465 |
-
"metric_list": [
|
| 1466 |
-
{
|
| 1467 |
-
"metric": "acc",
|
| 1468 |
-
"aggregation": "mean",
|
| 1469 |
-
"higher_is_better": true
|
| 1470 |
-
}
|
| 1471 |
-
],
|
| 1472 |
-
"output_type": "multiple_choice",
|
| 1473 |
-
"repeats": 1,
|
| 1474 |
-
"should_decontaminate": false,
|
| 1475 |
-
"metadata": {
|
| 1476 |
-
"version": 0.0
|
| 1477 |
-
}
|
| 1478 |
-
},
|
| 1479 |
-
"mmlu_high_school_psychology": {
|
| 1480 |
-
"task": "mmlu_high_school_psychology",
|
| 1481 |
-
"task_alias": "high_school_psychology",
|
| 1482 |
-
"group": "mmlu_social_sciences",
|
| 1483 |
-
"group_alias": "social_sciences",
|
| 1484 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1485 |
-
"dataset_name": "high_school_psychology",
|
| 1486 |
-
"test_split": "test",
|
| 1487 |
-
"fewshot_split": "dev",
|
| 1488 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1489 |
-
"doc_to_target": "answer",
|
| 1490 |
-
"doc_to_choice": [
|
| 1491 |
-
"A",
|
| 1492 |
-
"B",
|
| 1493 |
-
"C",
|
| 1494 |
-
"D"
|
| 1495 |
-
],
|
| 1496 |
-
"description": "The following are multiple choice questions (with answers) about high school psychology.\n\n",
|
| 1497 |
-
"target_delimiter": " ",
|
| 1498 |
-
"fewshot_delimiter": "\n\n",
|
| 1499 |
-
"fewshot_config": {
|
| 1500 |
-
"sampler": "first_n"
|
| 1501 |
-
},
|
| 1502 |
-
"num_fewshot": 5,
|
| 1503 |
-
"metric_list": [
|
| 1504 |
-
{
|
| 1505 |
-
"metric": "acc",
|
| 1506 |
-
"aggregation": "mean",
|
| 1507 |
-
"higher_is_better": true
|
| 1508 |
-
}
|
| 1509 |
-
],
|
| 1510 |
-
"output_type": "multiple_choice",
|
| 1511 |
-
"repeats": 1,
|
| 1512 |
-
"should_decontaminate": false,
|
| 1513 |
-
"metadata": {
|
| 1514 |
-
"version": 0.0
|
| 1515 |
-
}
|
| 1516 |
-
},
|
| 1517 |
-
"mmlu_high_school_statistics": {
|
| 1518 |
-
"task": "mmlu_high_school_statistics",
|
| 1519 |
-
"task_alias": "high_school_statistics",
|
| 1520 |
-
"group": "mmlu_stem",
|
| 1521 |
-
"group_alias": "stem",
|
| 1522 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1523 |
-
"dataset_name": "high_school_statistics",
|
| 1524 |
-
"test_split": "test",
|
| 1525 |
-
"fewshot_split": "dev",
|
| 1526 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1527 |
-
"doc_to_target": "answer",
|
| 1528 |
-
"doc_to_choice": [
|
| 1529 |
-
"A",
|
| 1530 |
-
"B",
|
| 1531 |
-
"C",
|
| 1532 |
-
"D"
|
| 1533 |
-
],
|
| 1534 |
-
"description": "The following are multiple choice questions (with answers) about high school statistics.\n\n",
|
| 1535 |
-
"target_delimiter": " ",
|
| 1536 |
-
"fewshot_delimiter": "\n\n",
|
| 1537 |
-
"fewshot_config": {
|
| 1538 |
-
"sampler": "first_n"
|
| 1539 |
-
},
|
| 1540 |
-
"num_fewshot": 5,
|
| 1541 |
-
"metric_list": [
|
| 1542 |
-
{
|
| 1543 |
-
"metric": "acc",
|
| 1544 |
-
"aggregation": "mean",
|
| 1545 |
-
"higher_is_better": true
|
| 1546 |
-
}
|
| 1547 |
-
],
|
| 1548 |
-
"output_type": "multiple_choice",
|
| 1549 |
-
"repeats": 1,
|
| 1550 |
-
"should_decontaminate": false,
|
| 1551 |
-
"metadata": {
|
| 1552 |
-
"version": 0.0
|
| 1553 |
-
}
|
| 1554 |
-
},
|
| 1555 |
-
"mmlu_high_school_us_history": {
|
| 1556 |
-
"task": "mmlu_high_school_us_history",
|
| 1557 |
-
"task_alias": "high_school_us_history",
|
| 1558 |
-
"group": "mmlu_humanities",
|
| 1559 |
-
"group_alias": "humanities",
|
| 1560 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1561 |
-
"dataset_name": "high_school_us_history",
|
| 1562 |
-
"test_split": "test",
|
| 1563 |
-
"fewshot_split": "dev",
|
| 1564 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1565 |
-
"doc_to_target": "answer",
|
| 1566 |
-
"doc_to_choice": [
|
| 1567 |
-
"A",
|
| 1568 |
-
"B",
|
| 1569 |
-
"C",
|
| 1570 |
-
"D"
|
| 1571 |
-
],
|
| 1572 |
-
"description": "The following are multiple choice questions (with answers) about high school us history.\n\n",
|
| 1573 |
-
"target_delimiter": " ",
|
| 1574 |
-
"fewshot_delimiter": "\n\n",
|
| 1575 |
-
"fewshot_config": {
|
| 1576 |
-
"sampler": "first_n"
|
| 1577 |
-
},
|
| 1578 |
-
"num_fewshot": 5,
|
| 1579 |
-
"metric_list": [
|
| 1580 |
-
{
|
| 1581 |
-
"metric": "acc",
|
| 1582 |
-
"aggregation": "mean",
|
| 1583 |
-
"higher_is_better": true
|
| 1584 |
-
}
|
| 1585 |
-
],
|
| 1586 |
-
"output_type": "multiple_choice",
|
| 1587 |
-
"repeats": 1,
|
| 1588 |
-
"should_decontaminate": false,
|
| 1589 |
-
"metadata": {
|
| 1590 |
-
"version": 0.0
|
| 1591 |
-
}
|
| 1592 |
-
},
|
| 1593 |
-
"mmlu_high_school_world_history": {
|
| 1594 |
-
"task": "mmlu_high_school_world_history",
|
| 1595 |
-
"task_alias": "high_school_world_history",
|
| 1596 |
-
"group": "mmlu_humanities",
|
| 1597 |
-
"group_alias": "humanities",
|
| 1598 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1599 |
-
"dataset_name": "high_school_world_history",
|
| 1600 |
-
"test_split": "test",
|
| 1601 |
-
"fewshot_split": "dev",
|
| 1602 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1603 |
-
"doc_to_target": "answer",
|
| 1604 |
-
"doc_to_choice": [
|
| 1605 |
-
"A",
|
| 1606 |
-
"B",
|
| 1607 |
-
"C",
|
| 1608 |
-
"D"
|
| 1609 |
-
],
|
| 1610 |
-
"description": "The following are multiple choice questions (with answers) about high school world history.\n\n",
|
| 1611 |
-
"target_delimiter": " ",
|
| 1612 |
-
"fewshot_delimiter": "\n\n",
|
| 1613 |
-
"fewshot_config": {
|
| 1614 |
-
"sampler": "first_n"
|
| 1615 |
-
},
|
| 1616 |
-
"num_fewshot": 5,
|
| 1617 |
-
"metric_list": [
|
| 1618 |
-
{
|
| 1619 |
-
"metric": "acc",
|
| 1620 |
-
"aggregation": "mean",
|
| 1621 |
-
"higher_is_better": true
|
| 1622 |
-
}
|
| 1623 |
-
],
|
| 1624 |
-
"output_type": "multiple_choice",
|
| 1625 |
-
"repeats": 1,
|
| 1626 |
-
"should_decontaminate": false,
|
| 1627 |
-
"metadata": {
|
| 1628 |
-
"version": 0.0
|
| 1629 |
-
}
|
| 1630 |
-
},
|
| 1631 |
-
"mmlu_human_aging": {
|
| 1632 |
-
"task": "mmlu_human_aging",
|
| 1633 |
-
"task_alias": "human_aging",
|
| 1634 |
-
"group": "mmlu_other",
|
| 1635 |
-
"group_alias": "other",
|
| 1636 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1637 |
-
"dataset_name": "human_aging",
|
| 1638 |
-
"test_split": "test",
|
| 1639 |
-
"fewshot_split": "dev",
|
| 1640 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1641 |
-
"doc_to_target": "answer",
|
| 1642 |
-
"doc_to_choice": [
|
| 1643 |
-
"A",
|
| 1644 |
-
"B",
|
| 1645 |
-
"C",
|
| 1646 |
-
"D"
|
| 1647 |
-
],
|
| 1648 |
-
"description": "The following are multiple choice questions (with answers) about human aging.\n\n",
|
| 1649 |
-
"target_delimiter": " ",
|
| 1650 |
-
"fewshot_delimiter": "\n\n",
|
| 1651 |
-
"fewshot_config": {
|
| 1652 |
-
"sampler": "first_n"
|
| 1653 |
-
},
|
| 1654 |
-
"num_fewshot": 5,
|
| 1655 |
-
"metric_list": [
|
| 1656 |
-
{
|
| 1657 |
-
"metric": "acc",
|
| 1658 |
-
"aggregation": "mean",
|
| 1659 |
-
"higher_is_better": true
|
| 1660 |
-
}
|
| 1661 |
-
],
|
| 1662 |
-
"output_type": "multiple_choice",
|
| 1663 |
-
"repeats": 1,
|
| 1664 |
-
"should_decontaminate": false,
|
| 1665 |
-
"metadata": {
|
| 1666 |
-
"version": 0.0
|
| 1667 |
-
}
|
| 1668 |
-
},
|
| 1669 |
-
"mmlu_human_sexuality": {
|
| 1670 |
-
"task": "mmlu_human_sexuality",
|
| 1671 |
-
"task_alias": "human_sexuality",
|
| 1672 |
-
"group": "mmlu_social_sciences",
|
| 1673 |
-
"group_alias": "social_sciences",
|
| 1674 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1675 |
-
"dataset_name": "human_sexuality",
|
| 1676 |
-
"test_split": "test",
|
| 1677 |
-
"fewshot_split": "dev",
|
| 1678 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1679 |
-
"doc_to_target": "answer",
|
| 1680 |
-
"doc_to_choice": [
|
| 1681 |
-
"A",
|
| 1682 |
-
"B",
|
| 1683 |
-
"C",
|
| 1684 |
-
"D"
|
| 1685 |
-
],
|
| 1686 |
-
"description": "The following are multiple choice questions (with answers) about human sexuality.\n\n",
|
| 1687 |
-
"target_delimiter": " ",
|
| 1688 |
-
"fewshot_delimiter": "\n\n",
|
| 1689 |
-
"fewshot_config": {
|
| 1690 |
-
"sampler": "first_n"
|
| 1691 |
-
},
|
| 1692 |
-
"num_fewshot": 5,
|
| 1693 |
-
"metric_list": [
|
| 1694 |
-
{
|
| 1695 |
-
"metric": "acc",
|
| 1696 |
-
"aggregation": "mean",
|
| 1697 |
-
"higher_is_better": true
|
| 1698 |
-
}
|
| 1699 |
-
],
|
| 1700 |
-
"output_type": "multiple_choice",
|
| 1701 |
-
"repeats": 1,
|
| 1702 |
-
"should_decontaminate": false,
|
| 1703 |
-
"metadata": {
|
| 1704 |
-
"version": 0.0
|
| 1705 |
-
}
|
| 1706 |
-
},
|
| 1707 |
-
"mmlu_international_law": {
|
| 1708 |
-
"task": "mmlu_international_law",
|
| 1709 |
-
"task_alias": "international_law",
|
| 1710 |
-
"group": "mmlu_humanities",
|
| 1711 |
-
"group_alias": "humanities",
|
| 1712 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1713 |
-
"dataset_name": "international_law",
|
| 1714 |
-
"test_split": "test",
|
| 1715 |
-
"fewshot_split": "dev",
|
| 1716 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1717 |
-
"doc_to_target": "answer",
|
| 1718 |
-
"doc_to_choice": [
|
| 1719 |
-
"A",
|
| 1720 |
-
"B",
|
| 1721 |
-
"C",
|
| 1722 |
-
"D"
|
| 1723 |
-
],
|
| 1724 |
-
"description": "The following are multiple choice questions (with answers) about international law.\n\n",
|
| 1725 |
-
"target_delimiter": " ",
|
| 1726 |
-
"fewshot_delimiter": "\n\n",
|
| 1727 |
-
"fewshot_config": {
|
| 1728 |
-
"sampler": "first_n"
|
| 1729 |
-
},
|
| 1730 |
-
"num_fewshot": 5,
|
| 1731 |
-
"metric_list": [
|
| 1732 |
-
{
|
| 1733 |
-
"metric": "acc",
|
| 1734 |
-
"aggregation": "mean",
|
| 1735 |
-
"higher_is_better": true
|
| 1736 |
-
}
|
| 1737 |
-
],
|
| 1738 |
-
"output_type": "multiple_choice",
|
| 1739 |
-
"repeats": 1,
|
| 1740 |
-
"should_decontaminate": false,
|
| 1741 |
-
"metadata": {
|
| 1742 |
-
"version": 0.0
|
| 1743 |
-
}
|
| 1744 |
-
},
|
| 1745 |
-
"mmlu_jurisprudence": {
|
| 1746 |
-
"task": "mmlu_jurisprudence",
|
| 1747 |
-
"task_alias": "jurisprudence",
|
| 1748 |
-
"group": "mmlu_humanities",
|
| 1749 |
-
"group_alias": "humanities",
|
| 1750 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1751 |
-
"dataset_name": "jurisprudence",
|
| 1752 |
-
"test_split": "test",
|
| 1753 |
-
"fewshot_split": "dev",
|
| 1754 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1755 |
-
"doc_to_target": "answer",
|
| 1756 |
-
"doc_to_choice": [
|
| 1757 |
-
"A",
|
| 1758 |
-
"B",
|
| 1759 |
-
"C",
|
| 1760 |
-
"D"
|
| 1761 |
-
],
|
| 1762 |
-
"description": "The following are multiple choice questions (with answers) about jurisprudence.\n\n",
|
| 1763 |
-
"target_delimiter": " ",
|
| 1764 |
-
"fewshot_delimiter": "\n\n",
|
| 1765 |
-
"fewshot_config": {
|
| 1766 |
-
"sampler": "first_n"
|
| 1767 |
-
},
|
| 1768 |
-
"num_fewshot": 5,
|
| 1769 |
-
"metric_list": [
|
| 1770 |
-
{
|
| 1771 |
-
"metric": "acc",
|
| 1772 |
-
"aggregation": "mean",
|
| 1773 |
-
"higher_is_better": true
|
| 1774 |
-
}
|
| 1775 |
-
],
|
| 1776 |
-
"output_type": "multiple_choice",
|
| 1777 |
-
"repeats": 1,
|
| 1778 |
-
"should_decontaminate": false,
|
| 1779 |
-
"metadata": {
|
| 1780 |
-
"version": 0.0
|
| 1781 |
-
}
|
| 1782 |
-
},
|
| 1783 |
-
"mmlu_logical_fallacies": {
|
| 1784 |
-
"task": "mmlu_logical_fallacies",
|
| 1785 |
-
"task_alias": "logical_fallacies",
|
| 1786 |
-
"group": "mmlu_humanities",
|
| 1787 |
-
"group_alias": "humanities",
|
| 1788 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1789 |
-
"dataset_name": "logical_fallacies",
|
| 1790 |
-
"test_split": "test",
|
| 1791 |
-
"fewshot_split": "dev",
|
| 1792 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1793 |
-
"doc_to_target": "answer",
|
| 1794 |
-
"doc_to_choice": [
|
| 1795 |
-
"A",
|
| 1796 |
-
"B",
|
| 1797 |
-
"C",
|
| 1798 |
-
"D"
|
| 1799 |
-
],
|
| 1800 |
-
"description": "The following are multiple choice questions (with answers) about logical fallacies.\n\n",
|
| 1801 |
-
"target_delimiter": " ",
|
| 1802 |
-
"fewshot_delimiter": "\n\n",
|
| 1803 |
-
"fewshot_config": {
|
| 1804 |
-
"sampler": "first_n"
|
| 1805 |
-
},
|
| 1806 |
-
"num_fewshot": 5,
|
| 1807 |
-
"metric_list": [
|
| 1808 |
-
{
|
| 1809 |
-
"metric": "acc",
|
| 1810 |
-
"aggregation": "mean",
|
| 1811 |
-
"higher_is_better": true
|
| 1812 |
-
}
|
| 1813 |
-
],
|
| 1814 |
-
"output_type": "multiple_choice",
|
| 1815 |
-
"repeats": 1,
|
| 1816 |
-
"should_decontaminate": false,
|
| 1817 |
-
"metadata": {
|
| 1818 |
-
"version": 0.0
|
| 1819 |
-
}
|
| 1820 |
-
},
|
| 1821 |
-
"mmlu_machine_learning": {
|
| 1822 |
-
"task": "mmlu_machine_learning",
|
| 1823 |
-
"task_alias": "machine_learning",
|
| 1824 |
-
"group": "mmlu_stem",
|
| 1825 |
-
"group_alias": "stem",
|
| 1826 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1827 |
-
"dataset_name": "machine_learning",
|
| 1828 |
-
"test_split": "test",
|
| 1829 |
-
"fewshot_split": "dev",
|
| 1830 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1831 |
-
"doc_to_target": "answer",
|
| 1832 |
-
"doc_to_choice": [
|
| 1833 |
-
"A",
|
| 1834 |
-
"B",
|
| 1835 |
-
"C",
|
| 1836 |
-
"D"
|
| 1837 |
-
],
|
| 1838 |
-
"description": "The following are multiple choice questions (with answers) about machine learning.\n\n",
|
| 1839 |
-
"target_delimiter": " ",
|
| 1840 |
-
"fewshot_delimiter": "\n\n",
|
| 1841 |
-
"fewshot_config": {
|
| 1842 |
-
"sampler": "first_n"
|
| 1843 |
-
},
|
| 1844 |
-
"num_fewshot": 5,
|
| 1845 |
-
"metric_list": [
|
| 1846 |
-
{
|
| 1847 |
-
"metric": "acc",
|
| 1848 |
-
"aggregation": "mean",
|
| 1849 |
-
"higher_is_better": true
|
| 1850 |
-
}
|
| 1851 |
-
],
|
| 1852 |
-
"output_type": "multiple_choice",
|
| 1853 |
-
"repeats": 1,
|
| 1854 |
-
"should_decontaminate": false,
|
| 1855 |
-
"metadata": {
|
| 1856 |
-
"version": 0.0
|
| 1857 |
-
}
|
| 1858 |
-
},
|
| 1859 |
-
"mmlu_management": {
|
| 1860 |
-
"task": "mmlu_management",
|
| 1861 |
-
"task_alias": "management",
|
| 1862 |
-
"group": "mmlu_other",
|
| 1863 |
-
"group_alias": "other",
|
| 1864 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1865 |
-
"dataset_name": "management",
|
| 1866 |
-
"test_split": "test",
|
| 1867 |
-
"fewshot_split": "dev",
|
| 1868 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1869 |
-
"doc_to_target": "answer",
|
| 1870 |
-
"doc_to_choice": [
|
| 1871 |
-
"A",
|
| 1872 |
-
"B",
|
| 1873 |
-
"C",
|
| 1874 |
-
"D"
|
| 1875 |
-
],
|
| 1876 |
-
"description": "The following are multiple choice questions (with answers) about management.\n\n",
|
| 1877 |
-
"target_delimiter": " ",
|
| 1878 |
-
"fewshot_delimiter": "\n\n",
|
| 1879 |
-
"fewshot_config": {
|
| 1880 |
-
"sampler": "first_n"
|
| 1881 |
-
},
|
| 1882 |
-
"num_fewshot": 5,
|
| 1883 |
-
"metric_list": [
|
| 1884 |
-
{
|
| 1885 |
-
"metric": "acc",
|
| 1886 |
-
"aggregation": "mean",
|
| 1887 |
-
"higher_is_better": true
|
| 1888 |
-
}
|
| 1889 |
-
],
|
| 1890 |
-
"output_type": "multiple_choice",
|
| 1891 |
-
"repeats": 1,
|
| 1892 |
-
"should_decontaminate": false,
|
| 1893 |
-
"metadata": {
|
| 1894 |
-
"version": 0.0
|
| 1895 |
-
}
|
| 1896 |
-
},
|
| 1897 |
-
"mmlu_marketing": {
|
| 1898 |
-
"task": "mmlu_marketing",
|
| 1899 |
-
"task_alias": "marketing",
|
| 1900 |
-
"group": "mmlu_other",
|
| 1901 |
-
"group_alias": "other",
|
| 1902 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1903 |
-
"dataset_name": "marketing",
|
| 1904 |
-
"test_split": "test",
|
| 1905 |
-
"fewshot_split": "dev",
|
| 1906 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1907 |
-
"doc_to_target": "answer",
|
| 1908 |
-
"doc_to_choice": [
|
| 1909 |
-
"A",
|
| 1910 |
-
"B",
|
| 1911 |
-
"C",
|
| 1912 |
-
"D"
|
| 1913 |
-
],
|
| 1914 |
-
"description": "The following are multiple choice questions (with answers) about marketing.\n\n",
|
| 1915 |
-
"target_delimiter": " ",
|
| 1916 |
-
"fewshot_delimiter": "\n\n",
|
| 1917 |
-
"fewshot_config": {
|
| 1918 |
-
"sampler": "first_n"
|
| 1919 |
-
},
|
| 1920 |
-
"num_fewshot": 5,
|
| 1921 |
-
"metric_list": [
|
| 1922 |
-
{
|
| 1923 |
-
"metric": "acc",
|
| 1924 |
-
"aggregation": "mean",
|
| 1925 |
-
"higher_is_better": true
|
| 1926 |
-
}
|
| 1927 |
-
],
|
| 1928 |
-
"output_type": "multiple_choice",
|
| 1929 |
-
"repeats": 1,
|
| 1930 |
-
"should_decontaminate": false,
|
| 1931 |
-
"metadata": {
|
| 1932 |
-
"version": 0.0
|
| 1933 |
-
}
|
| 1934 |
-
},
|
| 1935 |
-
"mmlu_medical_genetics": {
|
| 1936 |
-
"task": "mmlu_medical_genetics",
|
| 1937 |
-
"task_alias": "medical_genetics",
|
| 1938 |
-
"group": "mmlu_other",
|
| 1939 |
-
"group_alias": "other",
|
| 1940 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1941 |
-
"dataset_name": "medical_genetics",
|
| 1942 |
-
"test_split": "test",
|
| 1943 |
-
"fewshot_split": "dev",
|
| 1944 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1945 |
-
"doc_to_target": "answer",
|
| 1946 |
-
"doc_to_choice": [
|
| 1947 |
-
"A",
|
| 1948 |
-
"B",
|
| 1949 |
-
"C",
|
| 1950 |
-
"D"
|
| 1951 |
-
],
|
| 1952 |
-
"description": "The following are multiple choice questions (with answers) about medical genetics.\n\n",
|
| 1953 |
-
"target_delimiter": " ",
|
| 1954 |
-
"fewshot_delimiter": "\n\n",
|
| 1955 |
-
"fewshot_config": {
|
| 1956 |
-
"sampler": "first_n"
|
| 1957 |
-
},
|
| 1958 |
-
"num_fewshot": 5,
|
| 1959 |
-
"metric_list": [
|
| 1960 |
-
{
|
| 1961 |
-
"metric": "acc",
|
| 1962 |
-
"aggregation": "mean",
|
| 1963 |
-
"higher_is_better": true
|
| 1964 |
-
}
|
| 1965 |
-
],
|
| 1966 |
-
"output_type": "multiple_choice",
|
| 1967 |
-
"repeats": 1,
|
| 1968 |
-
"should_decontaminate": false,
|
| 1969 |
-
"metadata": {
|
| 1970 |
-
"version": 0.0
|
| 1971 |
-
}
|
| 1972 |
-
},
|
| 1973 |
-
"mmlu_miscellaneous": {
|
| 1974 |
-
"task": "mmlu_miscellaneous",
|
| 1975 |
-
"task_alias": "miscellaneous",
|
| 1976 |
-
"group": "mmlu_other",
|
| 1977 |
-
"group_alias": "other",
|
| 1978 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 1979 |
-
"dataset_name": "miscellaneous",
|
| 1980 |
-
"test_split": "test",
|
| 1981 |
-
"fewshot_split": "dev",
|
| 1982 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 1983 |
-
"doc_to_target": "answer",
|
| 1984 |
-
"doc_to_choice": [
|
| 1985 |
-
"A",
|
| 1986 |
-
"B",
|
| 1987 |
-
"C",
|
| 1988 |
-
"D"
|
| 1989 |
-
],
|
| 1990 |
-
"description": "The following are multiple choice questions (with answers) about miscellaneous.\n\n",
|
| 1991 |
-
"target_delimiter": " ",
|
| 1992 |
-
"fewshot_delimiter": "\n\n",
|
| 1993 |
-
"fewshot_config": {
|
| 1994 |
-
"sampler": "first_n"
|
| 1995 |
-
},
|
| 1996 |
-
"num_fewshot": 5,
|
| 1997 |
-
"metric_list": [
|
| 1998 |
-
{
|
| 1999 |
-
"metric": "acc",
|
| 2000 |
-
"aggregation": "mean",
|
| 2001 |
-
"higher_is_better": true
|
| 2002 |
-
}
|
| 2003 |
-
],
|
| 2004 |
-
"output_type": "multiple_choice",
|
| 2005 |
-
"repeats": 1,
|
| 2006 |
-
"should_decontaminate": false,
|
| 2007 |
-
"metadata": {
|
| 2008 |
-
"version": 0.0
|
| 2009 |
-
}
|
| 2010 |
-
},
|
| 2011 |
-
"mmlu_moral_disputes": {
|
| 2012 |
-
"task": "mmlu_moral_disputes",
|
| 2013 |
-
"task_alias": "moral_disputes",
|
| 2014 |
-
"group": "mmlu_humanities",
|
| 2015 |
-
"group_alias": "humanities",
|
| 2016 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2017 |
-
"dataset_name": "moral_disputes",
|
| 2018 |
-
"test_split": "test",
|
| 2019 |
-
"fewshot_split": "dev",
|
| 2020 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2021 |
-
"doc_to_target": "answer",
|
| 2022 |
-
"doc_to_choice": [
|
| 2023 |
-
"A",
|
| 2024 |
-
"B",
|
| 2025 |
-
"C",
|
| 2026 |
-
"D"
|
| 2027 |
-
],
|
| 2028 |
-
"description": "The following are multiple choice questions (with answers) about moral disputes.\n\n",
|
| 2029 |
-
"target_delimiter": " ",
|
| 2030 |
-
"fewshot_delimiter": "\n\n",
|
| 2031 |
-
"fewshot_config": {
|
| 2032 |
-
"sampler": "first_n"
|
| 2033 |
-
},
|
| 2034 |
-
"num_fewshot": 5,
|
| 2035 |
-
"metric_list": [
|
| 2036 |
-
{
|
| 2037 |
-
"metric": "acc",
|
| 2038 |
-
"aggregation": "mean",
|
| 2039 |
-
"higher_is_better": true
|
| 2040 |
-
}
|
| 2041 |
-
],
|
| 2042 |
-
"output_type": "multiple_choice",
|
| 2043 |
-
"repeats": 1,
|
| 2044 |
-
"should_decontaminate": false,
|
| 2045 |
-
"metadata": {
|
| 2046 |
-
"version": 0.0
|
| 2047 |
-
}
|
| 2048 |
-
},
|
| 2049 |
-
"mmlu_moral_scenarios": {
|
| 2050 |
-
"task": "mmlu_moral_scenarios",
|
| 2051 |
-
"task_alias": "moral_scenarios",
|
| 2052 |
-
"group": "mmlu_humanities",
|
| 2053 |
-
"group_alias": "humanities",
|
| 2054 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2055 |
-
"dataset_name": "moral_scenarios",
|
| 2056 |
-
"test_split": "test",
|
| 2057 |
-
"fewshot_split": "dev",
|
| 2058 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2059 |
-
"doc_to_target": "answer",
|
| 2060 |
-
"doc_to_choice": [
|
| 2061 |
-
"A",
|
| 2062 |
-
"B",
|
| 2063 |
-
"C",
|
| 2064 |
-
"D"
|
| 2065 |
-
],
|
| 2066 |
-
"description": "The following are multiple choice questions (with answers) about moral scenarios.\n\n",
|
| 2067 |
-
"target_delimiter": " ",
|
| 2068 |
-
"fewshot_delimiter": "\n\n",
|
| 2069 |
-
"fewshot_config": {
|
| 2070 |
-
"sampler": "first_n"
|
| 2071 |
-
},
|
| 2072 |
-
"num_fewshot": 5,
|
| 2073 |
-
"metric_list": [
|
| 2074 |
-
{
|
| 2075 |
-
"metric": "acc",
|
| 2076 |
-
"aggregation": "mean",
|
| 2077 |
-
"higher_is_better": true
|
| 2078 |
-
}
|
| 2079 |
-
],
|
| 2080 |
-
"output_type": "multiple_choice",
|
| 2081 |
-
"repeats": 1,
|
| 2082 |
-
"should_decontaminate": false,
|
| 2083 |
-
"metadata": {
|
| 2084 |
-
"version": 0.0
|
| 2085 |
-
}
|
| 2086 |
-
},
|
| 2087 |
-
"mmlu_nutrition": {
|
| 2088 |
-
"task": "mmlu_nutrition",
|
| 2089 |
-
"task_alias": "nutrition",
|
| 2090 |
-
"group": "mmlu_other",
|
| 2091 |
-
"group_alias": "other",
|
| 2092 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2093 |
-
"dataset_name": "nutrition",
|
| 2094 |
-
"test_split": "test",
|
| 2095 |
-
"fewshot_split": "dev",
|
| 2096 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2097 |
-
"doc_to_target": "answer",
|
| 2098 |
-
"doc_to_choice": [
|
| 2099 |
-
"A",
|
| 2100 |
-
"B",
|
| 2101 |
-
"C",
|
| 2102 |
-
"D"
|
| 2103 |
-
],
|
| 2104 |
-
"description": "The following are multiple choice questions (with answers) about nutrition.\n\n",
|
| 2105 |
-
"target_delimiter": " ",
|
| 2106 |
-
"fewshot_delimiter": "\n\n",
|
| 2107 |
-
"fewshot_config": {
|
| 2108 |
-
"sampler": "first_n"
|
| 2109 |
-
},
|
| 2110 |
-
"num_fewshot": 5,
|
| 2111 |
-
"metric_list": [
|
| 2112 |
-
{
|
| 2113 |
-
"metric": "acc",
|
| 2114 |
-
"aggregation": "mean",
|
| 2115 |
-
"higher_is_better": true
|
| 2116 |
-
}
|
| 2117 |
-
],
|
| 2118 |
-
"output_type": "multiple_choice",
|
| 2119 |
-
"repeats": 1,
|
| 2120 |
-
"should_decontaminate": false,
|
| 2121 |
-
"metadata": {
|
| 2122 |
-
"version": 0.0
|
| 2123 |
-
}
|
| 2124 |
-
},
|
| 2125 |
-
"mmlu_philosophy": {
|
| 2126 |
-
"task": "mmlu_philosophy",
|
| 2127 |
-
"task_alias": "philosophy",
|
| 2128 |
-
"group": "mmlu_humanities",
|
| 2129 |
-
"group_alias": "humanities",
|
| 2130 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2131 |
-
"dataset_name": "philosophy",
|
| 2132 |
-
"test_split": "test",
|
| 2133 |
-
"fewshot_split": "dev",
|
| 2134 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2135 |
-
"doc_to_target": "answer",
|
| 2136 |
-
"doc_to_choice": [
|
| 2137 |
-
"A",
|
| 2138 |
-
"B",
|
| 2139 |
-
"C",
|
| 2140 |
-
"D"
|
| 2141 |
-
],
|
| 2142 |
-
"description": "The following are multiple choice questions (with answers) about philosophy.\n\n",
|
| 2143 |
-
"target_delimiter": " ",
|
| 2144 |
-
"fewshot_delimiter": "\n\n",
|
| 2145 |
-
"fewshot_config": {
|
| 2146 |
-
"sampler": "first_n"
|
| 2147 |
-
},
|
| 2148 |
-
"num_fewshot": 5,
|
| 2149 |
-
"metric_list": [
|
| 2150 |
-
{
|
| 2151 |
-
"metric": "acc",
|
| 2152 |
-
"aggregation": "mean",
|
| 2153 |
-
"higher_is_better": true
|
| 2154 |
-
}
|
| 2155 |
-
],
|
| 2156 |
-
"output_type": "multiple_choice",
|
| 2157 |
-
"repeats": 1,
|
| 2158 |
-
"should_decontaminate": false,
|
| 2159 |
-
"metadata": {
|
| 2160 |
-
"version": 0.0
|
| 2161 |
-
}
|
| 2162 |
-
},
|
| 2163 |
-
"mmlu_prehistory": {
|
| 2164 |
-
"task": "mmlu_prehistory",
|
| 2165 |
-
"task_alias": "prehistory",
|
| 2166 |
-
"group": "mmlu_humanities",
|
| 2167 |
-
"group_alias": "humanities",
|
| 2168 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2169 |
-
"dataset_name": "prehistory",
|
| 2170 |
-
"test_split": "test",
|
| 2171 |
-
"fewshot_split": "dev",
|
| 2172 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2173 |
-
"doc_to_target": "answer",
|
| 2174 |
-
"doc_to_choice": [
|
| 2175 |
-
"A",
|
| 2176 |
-
"B",
|
| 2177 |
-
"C",
|
| 2178 |
-
"D"
|
| 2179 |
-
],
|
| 2180 |
-
"description": "The following are multiple choice questions (with answers) about prehistory.\n\n",
|
| 2181 |
-
"target_delimiter": " ",
|
| 2182 |
-
"fewshot_delimiter": "\n\n",
|
| 2183 |
-
"fewshot_config": {
|
| 2184 |
-
"sampler": "first_n"
|
| 2185 |
-
},
|
| 2186 |
-
"num_fewshot": 5,
|
| 2187 |
-
"metric_list": [
|
| 2188 |
-
{
|
| 2189 |
-
"metric": "acc",
|
| 2190 |
-
"aggregation": "mean",
|
| 2191 |
-
"higher_is_better": true
|
| 2192 |
-
}
|
| 2193 |
-
],
|
| 2194 |
-
"output_type": "multiple_choice",
|
| 2195 |
-
"repeats": 1,
|
| 2196 |
-
"should_decontaminate": false,
|
| 2197 |
-
"metadata": {
|
| 2198 |
-
"version": 0.0
|
| 2199 |
-
}
|
| 2200 |
-
},
|
| 2201 |
-
"mmlu_professional_accounting": {
|
| 2202 |
-
"task": "mmlu_professional_accounting",
|
| 2203 |
-
"task_alias": "professional_accounting",
|
| 2204 |
-
"group": "mmlu_other",
|
| 2205 |
-
"group_alias": "other",
|
| 2206 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2207 |
-
"dataset_name": "professional_accounting",
|
| 2208 |
-
"test_split": "test",
|
| 2209 |
-
"fewshot_split": "dev",
|
| 2210 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2211 |
-
"doc_to_target": "answer",
|
| 2212 |
-
"doc_to_choice": [
|
| 2213 |
-
"A",
|
| 2214 |
-
"B",
|
| 2215 |
-
"C",
|
| 2216 |
-
"D"
|
| 2217 |
-
],
|
| 2218 |
-
"description": "The following are multiple choice questions (with answers) about professional accounting.\n\n",
|
| 2219 |
-
"target_delimiter": " ",
|
| 2220 |
-
"fewshot_delimiter": "\n\n",
|
| 2221 |
-
"fewshot_config": {
|
| 2222 |
-
"sampler": "first_n"
|
| 2223 |
-
},
|
| 2224 |
-
"num_fewshot": 5,
|
| 2225 |
-
"metric_list": [
|
| 2226 |
-
{
|
| 2227 |
-
"metric": "acc",
|
| 2228 |
-
"aggregation": "mean",
|
| 2229 |
-
"higher_is_better": true
|
| 2230 |
-
}
|
| 2231 |
-
],
|
| 2232 |
-
"output_type": "multiple_choice",
|
| 2233 |
-
"repeats": 1,
|
| 2234 |
-
"should_decontaminate": false,
|
| 2235 |
-
"metadata": {
|
| 2236 |
-
"version": 0.0
|
| 2237 |
-
}
|
| 2238 |
-
},
|
| 2239 |
-
"mmlu_professional_law": {
|
| 2240 |
-
"task": "mmlu_professional_law",
|
| 2241 |
-
"task_alias": "professional_law",
|
| 2242 |
-
"group": "mmlu_humanities",
|
| 2243 |
-
"group_alias": "humanities",
|
| 2244 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2245 |
-
"dataset_name": "professional_law",
|
| 2246 |
-
"test_split": "test",
|
| 2247 |
-
"fewshot_split": "dev",
|
| 2248 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2249 |
-
"doc_to_target": "answer",
|
| 2250 |
-
"doc_to_choice": [
|
| 2251 |
-
"A",
|
| 2252 |
-
"B",
|
| 2253 |
-
"C",
|
| 2254 |
-
"D"
|
| 2255 |
-
],
|
| 2256 |
-
"description": "The following are multiple choice questions (with answers) about professional law.\n\n",
|
| 2257 |
-
"target_delimiter": " ",
|
| 2258 |
-
"fewshot_delimiter": "\n\n",
|
| 2259 |
-
"fewshot_config": {
|
| 2260 |
-
"sampler": "first_n"
|
| 2261 |
-
},
|
| 2262 |
-
"num_fewshot": 5,
|
| 2263 |
-
"metric_list": [
|
| 2264 |
-
{
|
| 2265 |
-
"metric": "acc",
|
| 2266 |
-
"aggregation": "mean",
|
| 2267 |
-
"higher_is_better": true
|
| 2268 |
-
}
|
| 2269 |
-
],
|
| 2270 |
-
"output_type": "multiple_choice",
|
| 2271 |
-
"repeats": 1,
|
| 2272 |
-
"should_decontaminate": false,
|
| 2273 |
-
"metadata": {
|
| 2274 |
-
"version": 0.0
|
| 2275 |
-
}
|
| 2276 |
-
},
|
| 2277 |
-
"mmlu_professional_medicine": {
|
| 2278 |
-
"task": "mmlu_professional_medicine",
|
| 2279 |
-
"task_alias": "professional_medicine",
|
| 2280 |
-
"group": "mmlu_other",
|
| 2281 |
-
"group_alias": "other",
|
| 2282 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2283 |
-
"dataset_name": "professional_medicine",
|
| 2284 |
-
"test_split": "test",
|
| 2285 |
-
"fewshot_split": "dev",
|
| 2286 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2287 |
-
"doc_to_target": "answer",
|
| 2288 |
-
"doc_to_choice": [
|
| 2289 |
-
"A",
|
| 2290 |
-
"B",
|
| 2291 |
-
"C",
|
| 2292 |
-
"D"
|
| 2293 |
-
],
|
| 2294 |
-
"description": "The following are multiple choice questions (with answers) about professional medicine.\n\n",
|
| 2295 |
-
"target_delimiter": " ",
|
| 2296 |
-
"fewshot_delimiter": "\n\n",
|
| 2297 |
-
"fewshot_config": {
|
| 2298 |
-
"sampler": "first_n"
|
| 2299 |
-
},
|
| 2300 |
-
"num_fewshot": 5,
|
| 2301 |
-
"metric_list": [
|
| 2302 |
-
{
|
| 2303 |
-
"metric": "acc",
|
| 2304 |
-
"aggregation": "mean",
|
| 2305 |
-
"higher_is_better": true
|
| 2306 |
-
}
|
| 2307 |
-
],
|
| 2308 |
-
"output_type": "multiple_choice",
|
| 2309 |
-
"repeats": 1,
|
| 2310 |
-
"should_decontaminate": false,
|
| 2311 |
-
"metadata": {
|
| 2312 |
-
"version": 0.0
|
| 2313 |
-
}
|
| 2314 |
-
},
|
| 2315 |
-
"mmlu_professional_psychology": {
|
| 2316 |
-
"task": "mmlu_professional_psychology",
|
| 2317 |
-
"task_alias": "professional_psychology",
|
| 2318 |
-
"group": "mmlu_social_sciences",
|
| 2319 |
-
"group_alias": "social_sciences",
|
| 2320 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2321 |
-
"dataset_name": "professional_psychology",
|
| 2322 |
-
"test_split": "test",
|
| 2323 |
-
"fewshot_split": "dev",
|
| 2324 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2325 |
-
"doc_to_target": "answer",
|
| 2326 |
-
"doc_to_choice": [
|
| 2327 |
-
"A",
|
| 2328 |
-
"B",
|
| 2329 |
-
"C",
|
| 2330 |
-
"D"
|
| 2331 |
-
],
|
| 2332 |
-
"description": "The following are multiple choice questions (with answers) about professional psychology.\n\n",
|
| 2333 |
-
"target_delimiter": " ",
|
| 2334 |
-
"fewshot_delimiter": "\n\n",
|
| 2335 |
-
"fewshot_config": {
|
| 2336 |
-
"sampler": "first_n"
|
| 2337 |
-
},
|
| 2338 |
-
"num_fewshot": 5,
|
| 2339 |
-
"metric_list": [
|
| 2340 |
-
{
|
| 2341 |
-
"metric": "acc",
|
| 2342 |
-
"aggregation": "mean",
|
| 2343 |
-
"higher_is_better": true
|
| 2344 |
-
}
|
| 2345 |
-
],
|
| 2346 |
-
"output_type": "multiple_choice",
|
| 2347 |
-
"repeats": 1,
|
| 2348 |
-
"should_decontaminate": false,
|
| 2349 |
-
"metadata": {
|
| 2350 |
-
"version": 0.0
|
| 2351 |
-
}
|
| 2352 |
-
},
|
| 2353 |
-
"mmlu_public_relations": {
|
| 2354 |
-
"task": "mmlu_public_relations",
|
| 2355 |
-
"task_alias": "public_relations",
|
| 2356 |
-
"group": "mmlu_social_sciences",
|
| 2357 |
-
"group_alias": "social_sciences",
|
| 2358 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2359 |
-
"dataset_name": "public_relations",
|
| 2360 |
-
"test_split": "test",
|
| 2361 |
-
"fewshot_split": "dev",
|
| 2362 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2363 |
-
"doc_to_target": "answer",
|
| 2364 |
-
"doc_to_choice": [
|
| 2365 |
-
"A",
|
| 2366 |
-
"B",
|
| 2367 |
-
"C",
|
| 2368 |
-
"D"
|
| 2369 |
-
],
|
| 2370 |
-
"description": "The following are multiple choice questions (with answers) about public relations.\n\n",
|
| 2371 |
-
"target_delimiter": " ",
|
| 2372 |
-
"fewshot_delimiter": "\n\n",
|
| 2373 |
-
"fewshot_config": {
|
| 2374 |
-
"sampler": "first_n"
|
| 2375 |
-
},
|
| 2376 |
-
"num_fewshot": 5,
|
| 2377 |
-
"metric_list": [
|
| 2378 |
-
{
|
| 2379 |
-
"metric": "acc",
|
| 2380 |
-
"aggregation": "mean",
|
| 2381 |
-
"higher_is_better": true
|
| 2382 |
-
}
|
| 2383 |
-
],
|
| 2384 |
-
"output_type": "multiple_choice",
|
| 2385 |
-
"repeats": 1,
|
| 2386 |
-
"should_decontaminate": false,
|
| 2387 |
-
"metadata": {
|
| 2388 |
-
"version": 0.0
|
| 2389 |
-
}
|
| 2390 |
-
},
|
| 2391 |
-
"mmlu_security_studies": {
|
| 2392 |
-
"task": "mmlu_security_studies",
|
| 2393 |
-
"task_alias": "security_studies",
|
| 2394 |
-
"group": "mmlu_social_sciences",
|
| 2395 |
-
"group_alias": "social_sciences",
|
| 2396 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2397 |
-
"dataset_name": "security_studies",
|
| 2398 |
-
"test_split": "test",
|
| 2399 |
-
"fewshot_split": "dev",
|
| 2400 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2401 |
-
"doc_to_target": "answer",
|
| 2402 |
-
"doc_to_choice": [
|
| 2403 |
-
"A",
|
| 2404 |
-
"B",
|
| 2405 |
-
"C",
|
| 2406 |
-
"D"
|
| 2407 |
-
],
|
| 2408 |
-
"description": "The following are multiple choice questions (with answers) about security studies.\n\n",
|
| 2409 |
-
"target_delimiter": " ",
|
| 2410 |
-
"fewshot_delimiter": "\n\n",
|
| 2411 |
-
"fewshot_config": {
|
| 2412 |
-
"sampler": "first_n"
|
| 2413 |
-
},
|
| 2414 |
-
"num_fewshot": 5,
|
| 2415 |
-
"metric_list": [
|
| 2416 |
-
{
|
| 2417 |
-
"metric": "acc",
|
| 2418 |
-
"aggregation": "mean",
|
| 2419 |
-
"higher_is_better": true
|
| 2420 |
-
}
|
| 2421 |
-
],
|
| 2422 |
-
"output_type": "multiple_choice",
|
| 2423 |
-
"repeats": 1,
|
| 2424 |
-
"should_decontaminate": false,
|
| 2425 |
-
"metadata": {
|
| 2426 |
-
"version": 0.0
|
| 2427 |
-
}
|
| 2428 |
-
},
|
| 2429 |
-
"mmlu_sociology": {
|
| 2430 |
-
"task": "mmlu_sociology",
|
| 2431 |
-
"task_alias": "sociology",
|
| 2432 |
-
"group": "mmlu_social_sciences",
|
| 2433 |
-
"group_alias": "social_sciences",
|
| 2434 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2435 |
-
"dataset_name": "sociology",
|
| 2436 |
-
"test_split": "test",
|
| 2437 |
-
"fewshot_split": "dev",
|
| 2438 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2439 |
-
"doc_to_target": "answer",
|
| 2440 |
-
"doc_to_choice": [
|
| 2441 |
-
"A",
|
| 2442 |
-
"B",
|
| 2443 |
-
"C",
|
| 2444 |
-
"D"
|
| 2445 |
-
],
|
| 2446 |
-
"description": "The following are multiple choice questions (with answers) about sociology.\n\n",
|
| 2447 |
-
"target_delimiter": " ",
|
| 2448 |
-
"fewshot_delimiter": "\n\n",
|
| 2449 |
-
"fewshot_config": {
|
| 2450 |
-
"sampler": "first_n"
|
| 2451 |
-
},
|
| 2452 |
-
"num_fewshot": 5,
|
| 2453 |
-
"metric_list": [
|
| 2454 |
-
{
|
| 2455 |
-
"metric": "acc",
|
| 2456 |
-
"aggregation": "mean",
|
| 2457 |
-
"higher_is_better": true
|
| 2458 |
-
}
|
| 2459 |
-
],
|
| 2460 |
-
"output_type": "multiple_choice",
|
| 2461 |
-
"repeats": 1,
|
| 2462 |
-
"should_decontaminate": false,
|
| 2463 |
-
"metadata": {
|
| 2464 |
-
"version": 0.0
|
| 2465 |
-
}
|
| 2466 |
-
},
|
| 2467 |
-
"mmlu_us_foreign_policy": {
|
| 2468 |
-
"task": "mmlu_us_foreign_policy",
|
| 2469 |
-
"task_alias": "us_foreign_policy",
|
| 2470 |
-
"group": "mmlu_social_sciences",
|
| 2471 |
-
"group_alias": "social_sciences",
|
| 2472 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2473 |
-
"dataset_name": "us_foreign_policy",
|
| 2474 |
-
"test_split": "test",
|
| 2475 |
-
"fewshot_split": "dev",
|
| 2476 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2477 |
-
"doc_to_target": "answer",
|
| 2478 |
-
"doc_to_choice": [
|
| 2479 |
-
"A",
|
| 2480 |
-
"B",
|
| 2481 |
-
"C",
|
| 2482 |
-
"D"
|
| 2483 |
-
],
|
| 2484 |
-
"description": "The following are multiple choice questions (with answers) about us foreign policy.\n\n",
|
| 2485 |
-
"target_delimiter": " ",
|
| 2486 |
-
"fewshot_delimiter": "\n\n",
|
| 2487 |
-
"fewshot_config": {
|
| 2488 |
-
"sampler": "first_n"
|
| 2489 |
-
},
|
| 2490 |
-
"num_fewshot": 5,
|
| 2491 |
-
"metric_list": [
|
| 2492 |
-
{
|
| 2493 |
-
"metric": "acc",
|
| 2494 |
-
"aggregation": "mean",
|
| 2495 |
-
"higher_is_better": true
|
| 2496 |
-
}
|
| 2497 |
-
],
|
| 2498 |
-
"output_type": "multiple_choice",
|
| 2499 |
-
"repeats": 1,
|
| 2500 |
-
"should_decontaminate": false,
|
| 2501 |
-
"metadata": {
|
| 2502 |
-
"version": 0.0
|
| 2503 |
-
}
|
| 2504 |
-
},
|
| 2505 |
-
"mmlu_virology": {
|
| 2506 |
-
"task": "mmlu_virology",
|
| 2507 |
-
"task_alias": "virology",
|
| 2508 |
-
"group": "mmlu_other",
|
| 2509 |
-
"group_alias": "other",
|
| 2510 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2511 |
-
"dataset_name": "virology",
|
| 2512 |
-
"test_split": "test",
|
| 2513 |
-
"fewshot_split": "dev",
|
| 2514 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2515 |
-
"doc_to_target": "answer",
|
| 2516 |
-
"doc_to_choice": [
|
| 2517 |
-
"A",
|
| 2518 |
-
"B",
|
| 2519 |
-
"C",
|
| 2520 |
-
"D"
|
| 2521 |
-
],
|
| 2522 |
-
"description": "The following are multiple choice questions (with answers) about virology.\n\n",
|
| 2523 |
-
"target_delimiter": " ",
|
| 2524 |
-
"fewshot_delimiter": "\n\n",
|
| 2525 |
-
"fewshot_config": {
|
| 2526 |
-
"sampler": "first_n"
|
| 2527 |
-
},
|
| 2528 |
-
"num_fewshot": 5,
|
| 2529 |
-
"metric_list": [
|
| 2530 |
-
{
|
| 2531 |
-
"metric": "acc",
|
| 2532 |
-
"aggregation": "mean",
|
| 2533 |
-
"higher_is_better": true
|
| 2534 |
-
}
|
| 2535 |
-
],
|
| 2536 |
-
"output_type": "multiple_choice",
|
| 2537 |
-
"repeats": 1,
|
| 2538 |
-
"should_decontaminate": false,
|
| 2539 |
-
"metadata": {
|
| 2540 |
-
"version": 0.0
|
| 2541 |
-
}
|
| 2542 |
-
},
|
| 2543 |
-
"mmlu_world_religions": {
|
| 2544 |
-
"task": "mmlu_world_religions",
|
| 2545 |
-
"task_alias": "world_religions",
|
| 2546 |
-
"group": "mmlu_humanities",
|
| 2547 |
-
"group_alias": "humanities",
|
| 2548 |
-
"dataset_path": "hails/mmlu_no_train",
|
| 2549 |
-
"dataset_name": "world_religions",
|
| 2550 |
-
"test_split": "test",
|
| 2551 |
-
"fewshot_split": "dev",
|
| 2552 |
-
"doc_to_text": "{{question.strip()}}\nA. {{choices[0]}}\nB. {{choices[1]}}\nC. {{choices[2]}}\nD. {{choices[3]}}\nAnswer:",
|
| 2553 |
-
"doc_to_target": "answer",
|
| 2554 |
-
"doc_to_choice": [
|
| 2555 |
-
"A",
|
| 2556 |
-
"B",
|
| 2557 |
-
"C",
|
| 2558 |
-
"D"
|
| 2559 |
-
],
|
| 2560 |
-
"description": "The following are multiple choice questions (with answers) about world religions.\n\n",
|
| 2561 |
-
"target_delimiter": " ",
|
| 2562 |
-
"fewshot_delimiter": "\n\n",
|
| 2563 |
-
"fewshot_config": {
|
| 2564 |
-
"sampler": "first_n"
|
| 2565 |
-
},
|
| 2566 |
-
"num_fewshot": 5,
|
| 2567 |
-
"metric_list": [
|
| 2568 |
-
{
|
| 2569 |
-
"metric": "acc",
|
| 2570 |
-
"aggregation": "mean",
|
| 2571 |
-
"higher_is_better": true
|
| 2572 |
-
}
|
| 2573 |
-
],
|
| 2574 |
-
"output_type": "multiple_choice",
|
| 2575 |
-
"repeats": 1,
|
| 2576 |
-
"should_decontaminate": false,
|
| 2577 |
-
"metadata": {
|
| 2578 |
-
"version": 0.0
|
| 2579 |
-
}
|
| 2580 |
-
}
|
| 2581 |
-
},
|
| 2582 |
-
"versions": {
|
| 2583 |
-
"mmlu_abstract_algebra": 0.0,
|
| 2584 |
-
"mmlu_anatomy": 0.0,
|
| 2585 |
-
"mmlu_astronomy": 0.0,
|
| 2586 |
-
"mmlu_business_ethics": 0.0,
|
| 2587 |
-
"mmlu_clinical_knowledge": 0.0,
|
| 2588 |
-
"mmlu_college_biology": 0.0,
|
| 2589 |
-
"mmlu_college_chemistry": 0.0,
|
| 2590 |
-
"mmlu_college_computer_science": 0.0,
|
| 2591 |
-
"mmlu_college_mathematics": 0.0,
|
| 2592 |
-
"mmlu_college_medicine": 0.0,
|
| 2593 |
-
"mmlu_college_physics": 0.0,
|
| 2594 |
-
"mmlu_computer_security": 0.0,
|
| 2595 |
-
"mmlu_conceptual_physics": 0.0,
|
| 2596 |
-
"mmlu_econometrics": 0.0,
|
| 2597 |
-
"mmlu_electrical_engineering": 0.0,
|
| 2598 |
-
"mmlu_elementary_mathematics": 0.0,
|
| 2599 |
-
"mmlu_formal_logic": 0.0,
|
| 2600 |
-
"mmlu_global_facts": 0.0,
|
| 2601 |
-
"mmlu_high_school_biology": 0.0,
|
| 2602 |
-
"mmlu_high_school_chemistry": 0.0,
|
| 2603 |
-
"mmlu_high_school_computer_science": 0.0,
|
| 2604 |
-
"mmlu_high_school_european_history": 0.0,
|
| 2605 |
-
"mmlu_high_school_geography": 0.0,
|
| 2606 |
-
"mmlu_high_school_government_and_politics": 0.0,
|
| 2607 |
-
"mmlu_high_school_macroeconomics": 0.0,
|
| 2608 |
-
"mmlu_high_school_mathematics": 0.0,
|
| 2609 |
-
"mmlu_high_school_microeconomics": 0.0,
|
| 2610 |
-
"mmlu_high_school_physics": 0.0,
|
| 2611 |
-
"mmlu_high_school_psychology": 0.0,
|
| 2612 |
-
"mmlu_high_school_statistics": 0.0,
|
| 2613 |
-
"mmlu_high_school_us_history": 0.0,
|
| 2614 |
-
"mmlu_high_school_world_history": 0.0,
|
| 2615 |
-
"mmlu_human_aging": 0.0,
|
| 2616 |
-
"mmlu_human_sexuality": 0.0,
|
| 2617 |
-
"mmlu_international_law": 0.0,
|
| 2618 |
-
"mmlu_jurisprudence": 0.0,
|
| 2619 |
-
"mmlu_logical_fallacies": 0.0,
|
| 2620 |
-
"mmlu_machine_learning": 0.0,
|
| 2621 |
-
"mmlu_management": 0.0,
|
| 2622 |
-
"mmlu_marketing": 0.0,
|
| 2623 |
-
"mmlu_medical_genetics": 0.0,
|
| 2624 |
-
"mmlu_miscellaneous": 0.0,
|
| 2625 |
-
"mmlu_moral_disputes": 0.0,
|
| 2626 |
-
"mmlu_moral_scenarios": 0.0,
|
| 2627 |
-
"mmlu_nutrition": 0.0,
|
| 2628 |
-
"mmlu_philosophy": 0.0,
|
| 2629 |
-
"mmlu_prehistory": 0.0,
|
| 2630 |
-
"mmlu_professional_accounting": 0.0,
|
| 2631 |
-
"mmlu_professional_law": 0.0,
|
| 2632 |
-
"mmlu_professional_medicine": 0.0,
|
| 2633 |
-
"mmlu_professional_psychology": 0.0,
|
| 2634 |
-
"mmlu_public_relations": 0.0,
|
| 2635 |
-
"mmlu_security_studies": 0.0,
|
| 2636 |
-
"mmlu_sociology": 0.0,
|
| 2637 |
-
"mmlu_us_foreign_policy": 0.0,
|
| 2638 |
-
"mmlu_virology": 0.0,
|
| 2639 |
-
"mmlu_world_religions": 0.0
|
| 2640 |
-
},
|
| 2641 |
-
"n-shot": {
|
| 2642 |
-
"mmlu": 0,
|
| 2643 |
-
"mmlu_abstract_algebra": 5,
|
| 2644 |
-
"mmlu_anatomy": 5,
|
| 2645 |
-
"mmlu_astronomy": 5,
|
| 2646 |
-
"mmlu_business_ethics": 5,
|
| 2647 |
-
"mmlu_clinical_knowledge": 5,
|
| 2648 |
-
"mmlu_college_biology": 5,
|
| 2649 |
-
"mmlu_college_chemistry": 5,
|
| 2650 |
-
"mmlu_college_computer_science": 5,
|
| 2651 |
-
"mmlu_college_mathematics": 5,
|
| 2652 |
-
"mmlu_college_medicine": 5,
|
| 2653 |
-
"mmlu_college_physics": 5,
|
| 2654 |
-
"mmlu_computer_security": 5,
|
| 2655 |
-
"mmlu_conceptual_physics": 5,
|
| 2656 |
-
"mmlu_econometrics": 5,
|
| 2657 |
-
"mmlu_electrical_engineering": 5,
|
| 2658 |
-
"mmlu_elementary_mathematics": 5,
|
| 2659 |
-
"mmlu_formal_logic": 5,
|
| 2660 |
-
"mmlu_global_facts": 5,
|
| 2661 |
-
"mmlu_high_school_biology": 5,
|
| 2662 |
-
"mmlu_high_school_chemistry": 5,
|
| 2663 |
-
"mmlu_high_school_computer_science": 5,
|
| 2664 |
-
"mmlu_high_school_european_history": 5,
|
| 2665 |
-
"mmlu_high_school_geography": 5,
|
| 2666 |
-
"mmlu_high_school_government_and_politics": 5,
|
| 2667 |
-
"mmlu_high_school_macroeconomics": 5,
|
| 2668 |
-
"mmlu_high_school_mathematics": 5,
|
| 2669 |
-
"mmlu_high_school_microeconomics": 5,
|
| 2670 |
-
"mmlu_high_school_physics": 5,
|
| 2671 |
-
"mmlu_high_school_psychology": 5,
|
| 2672 |
-
"mmlu_high_school_statistics": 5,
|
| 2673 |
-
"mmlu_high_school_us_history": 5,
|
| 2674 |
-
"mmlu_high_school_world_history": 5,
|
| 2675 |
-
"mmlu_human_aging": 5,
|
| 2676 |
-
"mmlu_human_sexuality": 5,
|
| 2677 |
-
"mmlu_humanities": 5,
|
| 2678 |
-
"mmlu_international_law": 5,
|
| 2679 |
-
"mmlu_jurisprudence": 5,
|
| 2680 |
-
"mmlu_logical_fallacies": 5,
|
| 2681 |
-
"mmlu_machine_learning": 5,
|
| 2682 |
-
"mmlu_management": 5,
|
| 2683 |
-
"mmlu_marketing": 5,
|
| 2684 |
-
"mmlu_medical_genetics": 5,
|
| 2685 |
-
"mmlu_miscellaneous": 5,
|
| 2686 |
-
"mmlu_moral_disputes": 5,
|
| 2687 |
-
"mmlu_moral_scenarios": 5,
|
| 2688 |
-
"mmlu_nutrition": 5,
|
| 2689 |
-
"mmlu_other": 5,
|
| 2690 |
-
"mmlu_philosophy": 5,
|
| 2691 |
-
"mmlu_prehistory": 5,
|
| 2692 |
-
"mmlu_professional_accounting": 5,
|
| 2693 |
-
"mmlu_professional_law": 5,
|
| 2694 |
-
"mmlu_professional_medicine": 5,
|
| 2695 |
-
"mmlu_professional_psychology": 5,
|
| 2696 |
-
"mmlu_public_relations": 5,
|
| 2697 |
-
"mmlu_security_studies": 5,
|
| 2698 |
-
"mmlu_social_sciences": 5,
|
| 2699 |
-
"mmlu_sociology": 5,
|
| 2700 |
-
"mmlu_stem": 5,
|
| 2701 |
-
"mmlu_us_foreign_policy": 5,
|
| 2702 |
-
"mmlu_virology": 5,
|
| 2703 |
-
"mmlu_world_religions": 5
|
| 2704 |
-
},
|
| 2705 |
-
"higher_is_better": {
|
| 2706 |
-
"mmlu": {
|
| 2707 |
-
"acc": true
|
| 2708 |
-
},
|
| 2709 |
-
"mmlu_abstract_algebra": {
|
| 2710 |
-
"acc": true
|
| 2711 |
-
},
|
| 2712 |
-
"mmlu_anatomy": {
|
| 2713 |
-
"acc": true
|
| 2714 |
-
},
|
| 2715 |
-
"mmlu_astronomy": {
|
| 2716 |
-
"acc": true
|
| 2717 |
-
},
|
| 2718 |
-
"mmlu_business_ethics": {
|
| 2719 |
-
"acc": true
|
| 2720 |
-
},
|
| 2721 |
-
"mmlu_clinical_knowledge": {
|
| 2722 |
-
"acc": true
|
| 2723 |
-
},
|
| 2724 |
-
"mmlu_college_biology": {
|
| 2725 |
-
"acc": true
|
| 2726 |
-
},
|
| 2727 |
-
"mmlu_college_chemistry": {
|
| 2728 |
-
"acc": true
|
| 2729 |
-
},
|
| 2730 |
-
"mmlu_college_computer_science": {
|
| 2731 |
-
"acc": true
|
| 2732 |
-
},
|
| 2733 |
-
"mmlu_college_mathematics": {
|
| 2734 |
-
"acc": true
|
| 2735 |
-
},
|
| 2736 |
-
"mmlu_college_medicine": {
|
| 2737 |
-
"acc": true
|
| 2738 |
-
},
|
| 2739 |
-
"mmlu_college_physics": {
|
| 2740 |
-
"acc": true
|
| 2741 |
-
},
|
| 2742 |
-
"mmlu_computer_security": {
|
| 2743 |
-
"acc": true
|
| 2744 |
-
},
|
| 2745 |
-
"mmlu_conceptual_physics": {
|
| 2746 |
-
"acc": true
|
| 2747 |
-
},
|
| 2748 |
-
"mmlu_econometrics": {
|
| 2749 |
-
"acc": true
|
| 2750 |
-
},
|
| 2751 |
-
"mmlu_electrical_engineering": {
|
| 2752 |
-
"acc": true
|
| 2753 |
-
},
|
| 2754 |
-
"mmlu_elementary_mathematics": {
|
| 2755 |
-
"acc": true
|
| 2756 |
-
},
|
| 2757 |
-
"mmlu_formal_logic": {
|
| 2758 |
-
"acc": true
|
| 2759 |
-
},
|
| 2760 |
-
"mmlu_global_facts": {
|
| 2761 |
-
"acc": true
|
| 2762 |
-
},
|
| 2763 |
-
"mmlu_high_school_biology": {
|
| 2764 |
-
"acc": true
|
| 2765 |
-
},
|
| 2766 |
-
"mmlu_high_school_chemistry": {
|
| 2767 |
-
"acc": true
|
| 2768 |
-
},
|
| 2769 |
-
"mmlu_high_school_computer_science": {
|
| 2770 |
-
"acc": true
|
| 2771 |
-
},
|
| 2772 |
-
"mmlu_high_school_european_history": {
|
| 2773 |
-
"acc": true
|
| 2774 |
-
},
|
| 2775 |
-
"mmlu_high_school_geography": {
|
| 2776 |
-
"acc": true
|
| 2777 |
-
},
|
| 2778 |
-
"mmlu_high_school_government_and_politics": {
|
| 2779 |
-
"acc": true
|
| 2780 |
-
},
|
| 2781 |
-
"mmlu_high_school_macroeconomics": {
|
| 2782 |
-
"acc": true
|
| 2783 |
-
},
|
| 2784 |
-
"mmlu_high_school_mathematics": {
|
| 2785 |
-
"acc": true
|
| 2786 |
-
},
|
| 2787 |
-
"mmlu_high_school_microeconomics": {
|
| 2788 |
-
"acc": true
|
| 2789 |
-
},
|
| 2790 |
-
"mmlu_high_school_physics": {
|
| 2791 |
-
"acc": true
|
| 2792 |
-
},
|
| 2793 |
-
"mmlu_high_school_psychology": {
|
| 2794 |
-
"acc": true
|
| 2795 |
-
},
|
| 2796 |
-
"mmlu_high_school_statistics": {
|
| 2797 |
-
"acc": true
|
| 2798 |
-
},
|
| 2799 |
-
"mmlu_high_school_us_history": {
|
| 2800 |
-
"acc": true
|
| 2801 |
-
},
|
| 2802 |
-
"mmlu_high_school_world_history": {
|
| 2803 |
-
"acc": true
|
| 2804 |
-
},
|
| 2805 |
-
"mmlu_human_aging": {
|
| 2806 |
-
"acc": true
|
| 2807 |
-
},
|
| 2808 |
-
"mmlu_human_sexuality": {
|
| 2809 |
-
"acc": true
|
| 2810 |
-
},
|
| 2811 |
-
"mmlu_humanities": {
|
| 2812 |
-
"acc": true
|
| 2813 |
-
},
|
| 2814 |
-
"mmlu_international_law": {
|
| 2815 |
-
"acc": true
|
| 2816 |
-
},
|
| 2817 |
-
"mmlu_jurisprudence": {
|
| 2818 |
-
"acc": true
|
| 2819 |
-
},
|
| 2820 |
-
"mmlu_logical_fallacies": {
|
| 2821 |
-
"acc": true
|
| 2822 |
-
},
|
| 2823 |
-
"mmlu_machine_learning": {
|
| 2824 |
-
"acc": true
|
| 2825 |
-
},
|
| 2826 |
-
"mmlu_management": {
|
| 2827 |
-
"acc": true
|
| 2828 |
-
},
|
| 2829 |
-
"mmlu_marketing": {
|
| 2830 |
-
"acc": true
|
| 2831 |
-
},
|
| 2832 |
-
"mmlu_medical_genetics": {
|
| 2833 |
-
"acc": true
|
| 2834 |
-
},
|
| 2835 |
-
"mmlu_miscellaneous": {
|
| 2836 |
-
"acc": true
|
| 2837 |
-
},
|
| 2838 |
-
"mmlu_moral_disputes": {
|
| 2839 |
-
"acc": true
|
| 2840 |
-
},
|
| 2841 |
-
"mmlu_moral_scenarios": {
|
| 2842 |
-
"acc": true
|
| 2843 |
-
},
|
| 2844 |
-
"mmlu_nutrition": {
|
| 2845 |
-
"acc": true
|
| 2846 |
-
},
|
| 2847 |
-
"mmlu_other": {
|
| 2848 |
-
"acc": true
|
| 2849 |
-
},
|
| 2850 |
-
"mmlu_philosophy": {
|
| 2851 |
-
"acc": true
|
| 2852 |
-
},
|
| 2853 |
-
"mmlu_prehistory": {
|
| 2854 |
-
"acc": true
|
| 2855 |
-
},
|
| 2856 |
-
"mmlu_professional_accounting": {
|
| 2857 |
-
"acc": true
|
| 2858 |
-
},
|
| 2859 |
-
"mmlu_professional_law": {
|
| 2860 |
-
"acc": true
|
| 2861 |
-
},
|
| 2862 |
-
"mmlu_professional_medicine": {
|
| 2863 |
-
"acc": true
|
| 2864 |
-
},
|
| 2865 |
-
"mmlu_professional_psychology": {
|
| 2866 |
-
"acc": true
|
| 2867 |
-
},
|
| 2868 |
-
"mmlu_public_relations": {
|
| 2869 |
-
"acc": true
|
| 2870 |
-
},
|
| 2871 |
-
"mmlu_security_studies": {
|
| 2872 |
-
"acc": true
|
| 2873 |
-
},
|
| 2874 |
-
"mmlu_social_sciences": {
|
| 2875 |
-
"acc": true
|
| 2876 |
-
},
|
| 2877 |
-
"mmlu_sociology": {
|
| 2878 |
-
"acc": true
|
| 2879 |
-
},
|
| 2880 |
-
"mmlu_stem": {
|
| 2881 |
-
"acc": true
|
| 2882 |
-
},
|
| 2883 |
-
"mmlu_us_foreign_policy": {
|
| 2884 |
-
"acc": true
|
| 2885 |
-
},
|
| 2886 |
-
"mmlu_virology": {
|
| 2887 |
-
"acc": true
|
| 2888 |
-
},
|
| 2889 |
-
"mmlu_world_religions": {
|
| 2890 |
-
"acc": true
|
| 2891 |
-
}
|
| 2892 |
-
},
|
| 2893 |
-
"n-samples": {
|
| 2894 |
-
"mmlu_moral_scenarios": {
|
| 2895 |
-
"original": 895,
|
| 2896 |
-
"effective": 895
|
| 2897 |
-
},
|
| 2898 |
-
"mmlu_high_school_us_history": {
|
| 2899 |
-
"original": 204,
|
| 2900 |
-
"effective": 204
|
| 2901 |
-
},
|
| 2902 |
-
"mmlu_high_school_world_history": {
|
| 2903 |
-
"original": 237,
|
| 2904 |
-
"effective": 237
|
| 2905 |
-
},
|
| 2906 |
-
"mmlu_world_religions": {
|
| 2907 |
-
"original": 171,
|
| 2908 |
-
"effective": 171
|
| 2909 |
-
},
|
| 2910 |
-
"mmlu_formal_logic": {
|
| 2911 |
-
"original": 126,
|
| 2912 |
-
"effective": 126
|
| 2913 |
-
},
|
| 2914 |
-
"mmlu_moral_disputes": {
|
| 2915 |
-
"original": 346,
|
| 2916 |
-
"effective": 346
|
| 2917 |
-
},
|
| 2918 |
-
"mmlu_prehistory": {
|
| 2919 |
-
"original": 324,
|
| 2920 |
-
"effective": 324
|
| 2921 |
-
},
|
| 2922 |
-
"mmlu_international_law": {
|
| 2923 |
-
"original": 121,
|
| 2924 |
-
"effective": 121
|
| 2925 |
-
},
|
| 2926 |
-
"mmlu_logical_fallacies": {
|
| 2927 |
-
"original": 163,
|
| 2928 |
-
"effective": 163
|
| 2929 |
-
},
|
| 2930 |
-
"mmlu_professional_law": {
|
| 2931 |
-
"original": 1534,
|
| 2932 |
-
"effective": 1534
|
| 2933 |
-
},
|
| 2934 |
-
"mmlu_philosophy": {
|
| 2935 |
-
"original": 311,
|
| 2936 |
-
"effective": 311
|
| 2937 |
-
},
|
| 2938 |
-
"mmlu_high_school_european_history": {
|
| 2939 |
-
"original": 165,
|
| 2940 |
-
"effective": 165
|
| 2941 |
-
},
|
| 2942 |
-
"mmlu_jurisprudence": {
|
| 2943 |
-
"original": 108,
|
| 2944 |
-
"effective": 108
|
| 2945 |
-
},
|
| 2946 |
-
"mmlu_high_school_psychology": {
|
| 2947 |
-
"original": 545,
|
| 2948 |
-
"effective": 545
|
| 2949 |
-
},
|
| 2950 |
-
"mmlu_high_school_geography": {
|
| 2951 |
-
"original": 198,
|
| 2952 |
-
"effective": 198
|
| 2953 |
-
},
|
| 2954 |
-
"mmlu_high_school_macroeconomics": {
|
| 2955 |
-
"original": 390,
|
| 2956 |
-
"effective": 390
|
| 2957 |
-
},
|
| 2958 |
-
"mmlu_public_relations": {
|
| 2959 |
-
"original": 110,
|
| 2960 |
-
"effective": 110
|
| 2961 |
-
},
|
| 2962 |
-
"mmlu_security_studies": {
|
| 2963 |
-
"original": 245,
|
| 2964 |
-
"effective": 245
|
| 2965 |
-
},
|
| 2966 |
-
"mmlu_high_school_microeconomics": {
|
| 2967 |
-
"original": 238,
|
| 2968 |
-
"effective": 238
|
| 2969 |
-
},
|
| 2970 |
-
"mmlu_human_sexuality": {
|
| 2971 |
-
"original": 131,
|
| 2972 |
-
"effective": 131
|
| 2973 |
-
},
|
| 2974 |
-
"mmlu_sociology": {
|
| 2975 |
-
"original": 201,
|
| 2976 |
-
"effective": 201
|
| 2977 |
-
},
|
| 2978 |
-
"mmlu_professional_psychology": {
|
| 2979 |
-
"original": 612,
|
| 2980 |
-
"effective": 612
|
| 2981 |
-
},
|
| 2982 |
-
"mmlu_econometrics": {
|
| 2983 |
-
"original": 114,
|
| 2984 |
-
"effective": 114
|
| 2985 |
-
},
|
| 2986 |
-
"mmlu_us_foreign_policy": {
|
| 2987 |
-
"original": 100,
|
| 2988 |
-
"effective": 100
|
| 2989 |
-
},
|
| 2990 |
-
"mmlu_high_school_government_and_politics": {
|
| 2991 |
-
"original": 193,
|
| 2992 |
-
"effective": 193
|
| 2993 |
-
},
|
| 2994 |
-
"mmlu_marketing": {
|
| 2995 |
-
"original": 234,
|
| 2996 |
-
"effective": 234
|
| 2997 |
-
},
|
| 2998 |
-
"mmlu_professional_accounting": {
|
| 2999 |
-
"original": 282,
|
| 3000 |
-
"effective": 282
|
| 3001 |
-
},
|
| 3002 |
-
"mmlu_clinical_knowledge": {
|
| 3003 |
-
"original": 265,
|
| 3004 |
-
"effective": 265
|
| 3005 |
-
},
|
| 3006 |
-
"mmlu_college_medicine": {
|
| 3007 |
-
"original": 173,
|
| 3008 |
-
"effective": 173
|
| 3009 |
-
},
|
| 3010 |
-
"mmlu_miscellaneous": {
|
| 3011 |
-
"original": 783,
|
| 3012 |
-
"effective": 783
|
| 3013 |
-
},
|
| 3014 |
-
"mmlu_virology": {
|
| 3015 |
-
"original": 166,
|
| 3016 |
-
"effective": 166
|
| 3017 |
-
},
|
| 3018 |
-
"mmlu_business_ethics": {
|
| 3019 |
-
"original": 100,
|
| 3020 |
-
"effective": 100
|
| 3021 |
-
},
|
| 3022 |
-
"mmlu_professional_medicine": {
|
| 3023 |
-
"original": 272,
|
| 3024 |
-
"effective": 272
|
| 3025 |
-
},
|
| 3026 |
-
"mmlu_global_facts": {
|
| 3027 |
-
"original": 100,
|
| 3028 |
-
"effective": 100
|
| 3029 |
-
},
|
| 3030 |
-
"mmlu_nutrition": {
|
| 3031 |
-
"original": 306,
|
| 3032 |
-
"effective": 306
|
| 3033 |
-
},
|
| 3034 |
-
"mmlu_human_aging": {
|
| 3035 |
-
"original": 223,
|
| 3036 |
-
"effective": 223
|
| 3037 |
-
},
|
| 3038 |
-
"mmlu_management": {
|
| 3039 |
-
"original": 103,
|
| 3040 |
-
"effective": 103
|
| 3041 |
-
},
|
| 3042 |
-
"mmlu_medical_genetics": {
|
| 3043 |
-
"original": 100,
|
| 3044 |
-
"effective": 100
|
| 3045 |
-
},
|
| 3046 |
-
"mmlu_college_biology": {
|
| 3047 |
-
"original": 144,
|
| 3048 |
-
"effective": 144
|
| 3049 |
-
},
|
| 3050 |
-
"mmlu_high_school_computer_science": {
|
| 3051 |
-
"original": 100,
|
| 3052 |
-
"effective": 100
|
| 3053 |
-
},
|
| 3054 |
-
"mmlu_elementary_mathematics": {
|
| 3055 |
-
"original": 378,
|
| 3056 |
-
"effective": 378
|
| 3057 |
-
},
|
| 3058 |
-
"mmlu_astronomy": {
|
| 3059 |
-
"original": 152,
|
| 3060 |
-
"effective": 152
|
| 3061 |
-
},
|
| 3062 |
-
"mmlu_machine_learning": {
|
| 3063 |
-
"original": 112,
|
| 3064 |
-
"effective": 112
|
| 3065 |
-
},
|
| 3066 |
-
"mmlu_high_school_mathematics": {
|
| 3067 |
-
"original": 270,
|
| 3068 |
-
"effective": 270
|
| 3069 |
-
},
|
| 3070 |
-
"mmlu_electrical_engineering": {
|
| 3071 |
-
"original": 145,
|
| 3072 |
-
"effective": 145
|
| 3073 |
-
},
|
| 3074 |
-
"mmlu_college_chemistry": {
|
| 3075 |
-
"original": 100,
|
| 3076 |
-
"effective": 100
|
| 3077 |
-
},
|
| 3078 |
-
"mmlu_college_mathematics": {
|
| 3079 |
-
"original": 100,
|
| 3080 |
-
"effective": 100
|
| 3081 |
-
},
|
| 3082 |
-
"mmlu_high_school_statistics": {
|
| 3083 |
-
"original": 216,
|
| 3084 |
-
"effective": 216
|
| 3085 |
-
},
|
| 3086 |
-
"mmlu_high_school_biology": {
|
| 3087 |
-
"original": 310,
|
| 3088 |
-
"effective": 310
|
| 3089 |
-
},
|
| 3090 |
-
"mmlu_abstract_algebra": {
|
| 3091 |
-
"original": 100,
|
| 3092 |
-
"effective": 100
|
| 3093 |
-
},
|
| 3094 |
-
"mmlu_college_physics": {
|
| 3095 |
-
"original": 102,
|
| 3096 |
-
"effective": 102
|
| 3097 |
-
},
|
| 3098 |
-
"mmlu_conceptual_physics": {
|
| 3099 |
-
"original": 235,
|
| 3100 |
-
"effective": 235
|
| 3101 |
-
},
|
| 3102 |
-
"mmlu_computer_security": {
|
| 3103 |
-
"original": 100,
|
| 3104 |
-
"effective": 100
|
| 3105 |
-
},
|
| 3106 |
-
"mmlu_anatomy": {
|
| 3107 |
-
"original": 135,
|
| 3108 |
-
"effective": 135
|
| 3109 |
-
},
|
| 3110 |
-
"mmlu_college_computer_science": {
|
| 3111 |
-
"original": 100,
|
| 3112 |
-
"effective": 100
|
| 3113 |
-
},
|
| 3114 |
-
"mmlu_high_school_physics": {
|
| 3115 |
-
"original": 151,
|
| 3116 |
-
"effective": 151
|
| 3117 |
-
},
|
| 3118 |
-
"mmlu_high_school_chemistry": {
|
| 3119 |
-
"original": 203,
|
| 3120 |
-
"effective": 203
|
| 3121 |
-
}
|
| 3122 |
-
},
|
| 3123 |
-
"config": {
|
| 3124 |
-
"model": "hf",
|
| 3125 |
-
"model_args": "pretrained=/home/migel/Tess-v2.5-qwen2-72B-safetensors,parallelize=True",
|
| 3126 |
-
"model_num_parameters": 72706203648,
|
| 3127 |
-
"model_dtype": "torch.float16",
|
| 3128 |
-
"model_revision": "main",
|
| 3129 |
-
"model_sha": "",
|
| 3130 |
-
"batch_size": "8",
|
| 3131 |
-
"batch_sizes": [],
|
| 3132 |
-
"device": null,
|
| 3133 |
-
"use_cache": null,
|
| 3134 |
-
"limit": null,
|
| 3135 |
-
"bootstrap_iters": 100000,
|
| 3136 |
-
"gen_kwargs": null,
|
| 3137 |
-
"random_seed": 0,
|
| 3138 |
-
"numpy_seed": 1234,
|
| 3139 |
-
"torch_seed": 1234,
|
| 3140 |
-
"fewshot_seed": 1234
|
| 3141 |
-
},
|
| 3142 |
-
"git_hash": "b3e4c49a",
|
| 3143 |
-
"date": 1718167288.656124,
|
| 3144 |
-
"pretty_env_info": "PyTorch version: 2.3.0+cu121\nIs debug build: False\nCUDA used to build PyTorch: 12.1\nROCM used to build PyTorch: N/A\n\nOS: Ubuntu 20.04.6 LTS (x86_64)\nGCC version: (Ubuntu 9.4.0-1ubuntu1~20.04.2) 9.4.0\nClang version: Could not collect\nCMake version: version 3.29.3\nLibc version: glibc-2.31\n\nPython version: 3.10.14 (main, Apr 6 2024, 18:45:05) [GCC 9.4.0] (64-bit runtime)\nPython platform: Linux-5.15.0-1050-azure-x86_64-with-glibc2.31\nIs CUDA available: True\nCUDA runtime version: 12.1.105\nCUDA_MODULE_LOADING set to: LAZY\nGPU models and configuration: \nGPU 0: NVIDIA A100 80GB PCIe\nGPU 1: NVIDIA A100 80GB PCIe\nGPU 2: NVIDIA A100 80GB PCIe\nGPU 3: NVIDIA A100 80GB PCIe\n\nNvidia driver version: 530.30.02\ncuDNN version: Could not collect\nHIP runtime version: N/A\nMIOpen runtime version: N/A\nIs XNNPACK available: True\n\nCPU:\nArchitecture: x86_64\nCPU op-mode(s): 32-bit, 64-bit\nByte Order: Little Endian\nAddress sizes: 48 bits physical, 48 bits virtual\nCPU(s): 96\nOn-line CPU(s) list: 0-95\nThread(s) per core: 1\nCore(s) per socket: 48\nSocket(s): 2\nNUMA node(s): 4\nVendor ID: AuthenticAMD\nCPU family: 25\nModel: 1\nModel name: AMD EPYC 7V13 64-Core Processor\nStepping: 1\nCPU MHz: 2445.435\nBogoMIPS: 4890.87\nHypervisor vendor: Microsoft\nVirtualization type: full\nL1d cache: 3 MiB\nL1i cache: 3 MiB\nL2 cache: 48 MiB\nL3 cache: 384 MiB\nNUMA node0 CPU(s): 0-23\nNUMA node1 CPU(s): 24-47\nNUMA node2 CPU(s): 48-71\nNUMA node3 CPU(s): 72-95\nVulnerability Gather data sampling: Not affected\nVulnerability Itlb multihit: Not affected\nVulnerability L1tf: Not affected\nVulnerability Mds: Not affected\nVulnerability Meltdown: Not affected\nVulnerability Mmio stale data: Not affected\nVulnerability Retbleed: Not affected\nVulnerability Spec rstack overflow: Mitigation; safe RET, no microcode\nVulnerability Spec store bypass: Vulnerable\nVulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization\nVulnerability Spectre v2: Mitigation; Retpolines, STIBP disabled, RSB filling, PBRSB-eIBRS Not affected\nVulnerability Srbds: Not affected\nVulnerability Tsx async abort: Not affected\nFlags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl tsc_reliable nonstop_tsc cpuid extd_apicid aperfmperf pni pclmulqdq ssse3 fma cx16 pcid sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw topoext perfctr_core invpcid_single vmmcall fsgsbase bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves clzero xsaveerptr rdpru arat umip vaes vpclmulqdq rdpid fsrm\n\nVersions of relevant libraries:\n[pip3] numpy==1.26.4\n[pip3] torch==2.3.0\n[pip3] triton==2.3.0\n[conda] magma-cuda117 2.6.1 1 pytorch\n[conda] mkl 2022.2.1 pypi_0 pypi\n[conda] mkl-include 2022.2.1 pypi_0 pypi\n[conda] numpy 1.24.4 pypi_0 pypi\n[conda] pytorch-lightning 1.9.5 pypi_0 pypi\n[conda] torch 2.0.1 pypi_0 pypi\n[conda] torch-nebula 0.16.10 pypi_0 pypi\n[conda] torch-ort 1.17.0 pypi_0 pypi\n[conda] torchaudio 2.0.2+cu117 pypi_0 pypi\n[conda] torchdata 0.6.1 pypi_0 pypi\n[conda] torchmetrics 1.2.0 pypi_0 pypi\n[conda] torchsnapshot 0.1.0 pypi_0 pypi\n[conda] torchvision 0.15.2+cu117 pypi_0 pypi\n[conda] triton 2.0.0 pypi_0 pypi",
|
| 3145 |
-
"transformers_version": "4.41.1",
|
| 3146 |
-
"upper_git_hash": null,
|
| 3147 |
-
"task_hashes": {},
|
| 3148 |
-
"model_source": "hf",
|
| 3149 |
-
"model_name": "/home/migel/Tess-v2.5-qwen2-72B-safetensors",
|
| 3150 |
-
"model_name_sanitized": "__home__migel__Tess-v2.5-qwen2-72B-safetensors",
|
| 3151 |
-
"system_instruction": null,
|
| 3152 |
-
"system_instruction_sha": null,
|
| 3153 |
-
"chat_template": null,
|
| 3154 |
-
"chat_template_sha": null,
|
| 3155 |
-
"start_time": 380863.826540975,
|
| 3156 |
-
"end_time": 388726.503174757,
|
| 3157 |
-
"total_evaluation_time_seconds": "7862.676633781986"
|
| 3158 |
-
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|