Upload output-2 with huggingface_hub
Browse files
output-2
ADDED
|
@@ -0,0 +1,431 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"results": {
|
| 3 |
+
"arc_easy": {
|
| 4 |
+
"acc": 0.8,
|
| 5 |
+
"acc_stderr": 0.040201512610368466,
|
| 6 |
+
"acc_norm": 0.77,
|
| 7 |
+
"acc_norm_stderr": 0.042295258468165065
|
| 8 |
+
},
|
| 9 |
+
"hellaswag": {
|
| 10 |
+
"acc": 0.61,
|
| 11 |
+
"acc_stderr": 0.04902071300001975,
|
| 12 |
+
"acc_norm": 0.81,
|
| 13 |
+
"acc_norm_stderr": 0.03942772444036623
|
| 14 |
+
},
|
| 15 |
+
"hendrycksTest-abstract_algebra": {
|
| 16 |
+
"acc": 0.3,
|
| 17 |
+
"acc_stderr": 0.046056618647183814,
|
| 18 |
+
"acc_norm": 0.3,
|
| 19 |
+
"acc_norm_stderr": 0.046056618647183814
|
| 20 |
+
},
|
| 21 |
+
"hendrycksTest-anatomy": {
|
| 22 |
+
"acc": 0.5,
|
| 23 |
+
"acc_stderr": 0.050251890762960605,
|
| 24 |
+
"acc_norm": 0.5,
|
| 25 |
+
"acc_norm_stderr": 0.050251890762960605
|
| 26 |
+
},
|
| 27 |
+
"hendrycksTest-astronomy": {
|
| 28 |
+
"acc": 0.52,
|
| 29 |
+
"acc_stderr": 0.050211673156867795,
|
| 30 |
+
"acc_norm": 0.52,
|
| 31 |
+
"acc_norm_stderr": 0.050211673156867795
|
| 32 |
+
},
|
| 33 |
+
"hendrycksTest-business_ethics": {
|
| 34 |
+
"acc": 0.45,
|
| 35 |
+
"acc_stderr": 0.05,
|
| 36 |
+
"acc_norm": 0.45,
|
| 37 |
+
"acc_norm_stderr": 0.05
|
| 38 |
+
},
|
| 39 |
+
"hendrycksTest-clinical_knowledge": {
|
| 40 |
+
"acc": 0.57,
|
| 41 |
+
"acc_stderr": 0.04975698519562428,
|
| 42 |
+
"acc_norm": 0.57,
|
| 43 |
+
"acc_norm_stderr": 0.04975698519562428
|
| 44 |
+
},
|
| 45 |
+
"hendrycksTest-college_biology": {
|
| 46 |
+
"acc": 0.5,
|
| 47 |
+
"acc_stderr": 0.050251890762960605,
|
| 48 |
+
"acc_norm": 0.5,
|
| 49 |
+
"acc_norm_stderr": 0.050251890762960605
|
| 50 |
+
},
|
| 51 |
+
"hendrycksTest-college_chemistry": {
|
| 52 |
+
"acc": 0.37,
|
| 53 |
+
"acc_stderr": 0.04852365870939099,
|
| 54 |
+
"acc_norm": 0.37,
|
| 55 |
+
"acc_norm_stderr": 0.04852365870939099
|
| 56 |
+
},
|
| 57 |
+
"hendrycksTest-college_computer_science": {
|
| 58 |
+
"acc": 0.41,
|
| 59 |
+
"acc_stderr": 0.049431107042371025,
|
| 60 |
+
"acc_norm": 0.41,
|
| 61 |
+
"acc_norm_stderr": 0.049431107042371025
|
| 62 |
+
},
|
| 63 |
+
"hendrycksTest-college_mathematics": {
|
| 64 |
+
"acc": 0.31,
|
| 65 |
+
"acc_stderr": 0.04648231987117316,
|
| 66 |
+
"acc_norm": 0.31,
|
| 67 |
+
"acc_norm_stderr": 0.04648231987117316
|
| 68 |
+
},
|
| 69 |
+
"hendrycksTest-college_medicine": {
|
| 70 |
+
"acc": 0.42,
|
| 71 |
+
"acc_stderr": 0.049604496374885836,
|
| 72 |
+
"acc_norm": 0.42,
|
| 73 |
+
"acc_norm_stderr": 0.049604496374885836
|
| 74 |
+
},
|
| 75 |
+
"hendrycksTest-college_physics": {
|
| 76 |
+
"acc": 0.27,
|
| 77 |
+
"acc_stderr": 0.044619604333847394,
|
| 78 |
+
"acc_norm": 0.27,
|
| 79 |
+
"acc_norm_stderr": 0.044619604333847394
|
| 80 |
+
},
|
| 81 |
+
"hendrycksTest-computer_security": {
|
| 82 |
+
"acc": 0.59,
|
| 83 |
+
"acc_stderr": 0.049431107042371025,
|
| 84 |
+
"acc_norm": 0.59,
|
| 85 |
+
"acc_norm_stderr": 0.049431107042371025
|
| 86 |
+
},
|
| 87 |
+
"hendrycksTest-conceptual_physics": {
|
| 88 |
+
"acc": 0.4,
|
| 89 |
+
"acc_stderr": 0.049236596391733084,
|
| 90 |
+
"acc_norm": 0.4,
|
| 91 |
+
"acc_norm_stderr": 0.049236596391733084
|
| 92 |
+
},
|
| 93 |
+
"hendrycksTest-econometrics": {
|
| 94 |
+
"acc": 0.28,
|
| 95 |
+
"acc_stderr": 0.04512608598542128,
|
| 96 |
+
"acc_norm": 0.28,
|
| 97 |
+
"acc_norm_stderr": 0.04512608598542128
|
| 98 |
+
},
|
| 99 |
+
"hendrycksTest-electrical_engineering": {
|
| 100 |
+
"acc": 0.44,
|
| 101 |
+
"acc_stderr": 0.04988876515698589,
|
| 102 |
+
"acc_norm": 0.44,
|
| 103 |
+
"acc_norm_stderr": 0.04988876515698589
|
| 104 |
+
},
|
| 105 |
+
"hendrycksTest-elementary_mathematics": {
|
| 106 |
+
"acc": 0.36,
|
| 107 |
+
"acc_stderr": 0.048241815132442176,
|
| 108 |
+
"acc_norm": 0.36,
|
| 109 |
+
"acc_norm_stderr": 0.048241815132442176
|
| 110 |
+
},
|
| 111 |
+
"hendrycksTest-formal_logic": {
|
| 112 |
+
"acc": 0.26,
|
| 113 |
+
"acc_stderr": 0.0440844002276808,
|
| 114 |
+
"acc_norm": 0.26,
|
| 115 |
+
"acc_norm_stderr": 0.0440844002276808
|
| 116 |
+
},
|
| 117 |
+
"hendrycksTest-global_facts": {
|
| 118 |
+
"acc": 0.33,
|
| 119 |
+
"acc_stderr": 0.047258156262526045,
|
| 120 |
+
"acc_norm": 0.33,
|
| 121 |
+
"acc_norm_stderr": 0.047258156262526045
|
| 122 |
+
},
|
| 123 |
+
"hendrycksTest-high_school_biology": {
|
| 124 |
+
"acc": 0.56,
|
| 125 |
+
"acc_stderr": 0.04988876515698589,
|
| 126 |
+
"acc_norm": 0.56,
|
| 127 |
+
"acc_norm_stderr": 0.04988876515698589
|
| 128 |
+
},
|
| 129 |
+
"hendrycksTest-high_school_chemistry": {
|
| 130 |
+
"acc": 0.34,
|
| 131 |
+
"acc_stderr": 0.04760952285695235,
|
| 132 |
+
"acc_norm": 0.34,
|
| 133 |
+
"acc_norm_stderr": 0.04760952285695235
|
| 134 |
+
},
|
| 135 |
+
"hendrycksTest-high_school_computer_science": {
|
| 136 |
+
"acc": 0.42,
|
| 137 |
+
"acc_stderr": 0.04960449637488584,
|
| 138 |
+
"acc_norm": 0.42,
|
| 139 |
+
"acc_norm_stderr": 0.04960449637488584
|
| 140 |
+
},
|
| 141 |
+
"hendrycksTest-high_school_european_history": {
|
| 142 |
+
"acc": 0.66,
|
| 143 |
+
"acc_stderr": 0.04760952285695237,
|
| 144 |
+
"acc_norm": 0.66,
|
| 145 |
+
"acc_norm_stderr": 0.04760952285695237
|
| 146 |
+
},
|
| 147 |
+
"hendrycksTest-high_school_geography": {
|
| 148 |
+
"acc": 0.61,
|
| 149 |
+
"acc_stderr": 0.04902071300001975,
|
| 150 |
+
"acc_norm": 0.61,
|
| 151 |
+
"acc_norm_stderr": 0.04902071300001975
|
| 152 |
+
},
|
| 153 |
+
"hendrycksTest-high_school_government_and_politics": {
|
| 154 |
+
"acc": 0.63,
|
| 155 |
+
"acc_stderr": 0.04852365870939099,
|
| 156 |
+
"acc_norm": 0.63,
|
| 157 |
+
"acc_norm_stderr": 0.04852365870939099
|
| 158 |
+
},
|
| 159 |
+
"hendrycksTest-high_school_macroeconomics": {
|
| 160 |
+
"acc": 0.47,
|
| 161 |
+
"acc_stderr": 0.050161355804659205,
|
| 162 |
+
"acc_norm": 0.47,
|
| 163 |
+
"acc_norm_stderr": 0.050161355804659205
|
| 164 |
+
},
|
| 165 |
+
"hendrycksTest-high_school_mathematics": {
|
| 166 |
+
"acc": 0.29,
|
| 167 |
+
"acc_stderr": 0.04560480215720684,
|
| 168 |
+
"acc_norm": 0.29,
|
| 169 |
+
"acc_norm_stderr": 0.04560480215720684
|
| 170 |
+
},
|
| 171 |
+
"hendrycksTest-high_school_microeconomics": {
|
| 172 |
+
"acc": 0.52,
|
| 173 |
+
"acc_stderr": 0.050211673156867795,
|
| 174 |
+
"acc_norm": 0.52,
|
| 175 |
+
"acc_norm_stderr": 0.050211673156867795
|
| 176 |
+
},
|
| 177 |
+
"hendrycksTest-high_school_physics": {
|
| 178 |
+
"acc": 0.25,
|
| 179 |
+
"acc_stderr": 0.04351941398892446,
|
| 180 |
+
"acc_norm": 0.25,
|
| 181 |
+
"acc_norm_stderr": 0.04351941398892446
|
| 182 |
+
},
|
| 183 |
+
"hendrycksTest-high_school_psychology": {
|
| 184 |
+
"acc": 0.66,
|
| 185 |
+
"acc_stderr": 0.04760952285695237,
|
| 186 |
+
"acc_norm": 0.66,
|
| 187 |
+
"acc_norm_stderr": 0.04760952285695237
|
| 188 |
+
},
|
| 189 |
+
"hendrycksTest-high_school_statistics": {
|
| 190 |
+
"acc": 0.37,
|
| 191 |
+
"acc_stderr": 0.048523658709391,
|
| 192 |
+
"acc_norm": 0.37,
|
| 193 |
+
"acc_norm_stderr": 0.048523658709391
|
| 194 |
+
},
|
| 195 |
+
"hendrycksTest-high_school_us_history": {
|
| 196 |
+
"acc": 0.61,
|
| 197 |
+
"acc_stderr": 0.04902071300001974,
|
| 198 |
+
"acc_norm": 0.61,
|
| 199 |
+
"acc_norm_stderr": 0.04902071300001974
|
| 200 |
+
},
|
| 201 |
+
"hendrycksTest-high_school_world_history": {
|
| 202 |
+
"acc": 0.73,
|
| 203 |
+
"acc_stderr": 0.0446196043338474,
|
| 204 |
+
"acc_norm": 0.73,
|
| 205 |
+
"acc_norm_stderr": 0.0446196043338474
|
| 206 |
+
},
|
| 207 |
+
"hendrycksTest-human_aging": {
|
| 208 |
+
"acc": 0.66,
|
| 209 |
+
"acc_stderr": 0.04760952285695238,
|
| 210 |
+
"acc_norm": 0.66,
|
| 211 |
+
"acc_norm_stderr": 0.04760952285695238
|
| 212 |
+
},
|
| 213 |
+
"hendrycksTest-human_sexuality": {
|
| 214 |
+
"acc": 0.6,
|
| 215 |
+
"acc_stderr": 0.049236596391733084,
|
| 216 |
+
"acc_norm": 0.6,
|
| 217 |
+
"acc_norm_stderr": 0.049236596391733084
|
| 218 |
+
},
|
| 219 |
+
"hendrycksTest-international_law": {
|
| 220 |
+
"acc": 0.62,
|
| 221 |
+
"acc_stderr": 0.04878317312145632,
|
| 222 |
+
"acc_norm": 0.62,
|
| 223 |
+
"acc_norm_stderr": 0.04878317312145632
|
| 224 |
+
},
|
| 225 |
+
"hendrycksTest-jurisprudence": {
|
| 226 |
+
"acc": 0.57,
|
| 227 |
+
"acc_stderr": 0.049756985195624284,
|
| 228 |
+
"acc_norm": 0.57,
|
| 229 |
+
"acc_norm_stderr": 0.049756985195624284
|
| 230 |
+
},
|
| 231 |
+
"hendrycksTest-logical_fallacies": {
|
| 232 |
+
"acc": 0.6,
|
| 233 |
+
"acc_stderr": 0.04923659639173309,
|
| 234 |
+
"acc_norm": 0.6,
|
| 235 |
+
"acc_norm_stderr": 0.04923659639173309
|
| 236 |
+
},
|
| 237 |
+
"hendrycksTest-machine_learning": {
|
| 238 |
+
"acc": 0.31,
|
| 239 |
+
"acc_stderr": 0.04648231987117316,
|
| 240 |
+
"acc_norm": 0.31,
|
| 241 |
+
"acc_norm_stderr": 0.04648231987117316
|
| 242 |
+
},
|
| 243 |
+
"hendrycksTest-management": {
|
| 244 |
+
"acc": 0.71,
|
| 245 |
+
"acc_stderr": 0.04560480215720684,
|
| 246 |
+
"acc_norm": 0.71,
|
| 247 |
+
"acc_norm_stderr": 0.04560480215720684
|
| 248 |
+
},
|
| 249 |
+
"hendrycksTest-marketing": {
|
| 250 |
+
"acc": 0.68,
|
| 251 |
+
"acc_stderr": 0.04688261722621505,
|
| 252 |
+
"acc_norm": 0.68,
|
| 253 |
+
"acc_norm_stderr": 0.04688261722621505
|
| 254 |
+
},
|
| 255 |
+
"hendrycksTest-medical_genetics": {
|
| 256 |
+
"acc": 0.52,
|
| 257 |
+
"acc_stderr": 0.05021167315686779,
|
| 258 |
+
"acc_norm": 0.52,
|
| 259 |
+
"acc_norm_stderr": 0.05021167315686779
|
| 260 |
+
},
|
| 261 |
+
"hendrycksTest-miscellaneous": {
|
| 262 |
+
"acc": 0.65,
|
| 263 |
+
"acc_stderr": 0.0479372485441102,
|
| 264 |
+
"acc_norm": 0.65,
|
| 265 |
+
"acc_norm_stderr": 0.0479372485441102
|
| 266 |
+
},
|
| 267 |
+
"hendrycksTest-moral_disputes": {
|
| 268 |
+
"acc": 0.61,
|
| 269 |
+
"acc_stderr": 0.04902071300001975,
|
| 270 |
+
"acc_norm": 0.61,
|
| 271 |
+
"acc_norm_stderr": 0.04902071300001975
|
| 272 |
+
},
|
| 273 |
+
"hendrycksTest-moral_scenarios": {
|
| 274 |
+
"acc": 0.19,
|
| 275 |
+
"acc_stderr": 0.03942772444036623,
|
| 276 |
+
"acc_norm": 0.19,
|
| 277 |
+
"acc_norm_stderr": 0.03942772444036623
|
| 278 |
+
},
|
| 279 |
+
"hendrycksTest-nutrition": {
|
| 280 |
+
"acc": 0.57,
|
| 281 |
+
"acc_stderr": 0.049756985195624284,
|
| 282 |
+
"acc_norm": 0.57,
|
| 283 |
+
"acc_norm_stderr": 0.049756985195624284
|
| 284 |
+
},
|
| 285 |
+
"hendrycksTest-philosophy": {
|
| 286 |
+
"acc": 0.57,
|
| 287 |
+
"acc_stderr": 0.04975698519562428,
|
| 288 |
+
"acc_norm": 0.57,
|
| 289 |
+
"acc_norm_stderr": 0.04975698519562428
|
| 290 |
+
},
|
| 291 |
+
"hendrycksTest-prehistory": {
|
| 292 |
+
"acc": 0.53,
|
| 293 |
+
"acc_stderr": 0.050161355804659205,
|
| 294 |
+
"acc_norm": 0.53,
|
| 295 |
+
"acc_norm_stderr": 0.050161355804659205
|
| 296 |
+
},
|
| 297 |
+
"hendrycksTest-professional_accounting": {
|
| 298 |
+
"acc": 0.42,
|
| 299 |
+
"acc_stderr": 0.049604496374885836,
|
| 300 |
+
"acc_norm": 0.42,
|
| 301 |
+
"acc_norm_stderr": 0.049604496374885836
|
| 302 |
+
},
|
| 303 |
+
"hendrycksTest-professional_law": {
|
| 304 |
+
"acc": 0.41,
|
| 305 |
+
"acc_stderr": 0.049431107042371025,
|
| 306 |
+
"acc_norm": 0.41,
|
| 307 |
+
"acc_norm_stderr": 0.049431107042371025
|
| 308 |
+
},
|
| 309 |
+
"hendrycksTest-professional_medicine": {
|
| 310 |
+
"acc": 0.56,
|
| 311 |
+
"acc_stderr": 0.04988876515698589,
|
| 312 |
+
"acc_norm": 0.56,
|
| 313 |
+
"acc_norm_stderr": 0.04988876515698589
|
| 314 |
+
},
|
| 315 |
+
"hendrycksTest-professional_psychology": {
|
| 316 |
+
"acc": 0.4,
|
| 317 |
+
"acc_stderr": 0.049236596391733084,
|
| 318 |
+
"acc_norm": 0.4,
|
| 319 |
+
"acc_norm_stderr": 0.049236596391733084
|
| 320 |
+
},
|
| 321 |
+
"hendrycksTest-public_relations": {
|
| 322 |
+
"acc": 0.52,
|
| 323 |
+
"acc_stderr": 0.050211673156867795,
|
| 324 |
+
"acc_norm": 0.52,
|
| 325 |
+
"acc_norm_stderr": 0.050211673156867795
|
| 326 |
+
},
|
| 327 |
+
"hendrycksTest-security_studies": {
|
| 328 |
+
"acc": 0.62,
|
| 329 |
+
"acc_stderr": 0.048783173121456316,
|
| 330 |
+
"acc_norm": 0.62,
|
| 331 |
+
"acc_norm_stderr": 0.048783173121456316
|
| 332 |
+
},
|
| 333 |
+
"hendrycksTest-sociology": {
|
| 334 |
+
"acc": 0.65,
|
| 335 |
+
"acc_stderr": 0.0479372485441102,
|
| 336 |
+
"acc_norm": 0.65,
|
| 337 |
+
"acc_norm_stderr": 0.0479372485441102
|
| 338 |
+
},
|
| 339 |
+
"hendrycksTest-us_foreign_policy": {
|
| 340 |
+
"acc": 0.68,
|
| 341 |
+
"acc_stderr": 0.04688261722621505,
|
| 342 |
+
"acc_norm": 0.68,
|
| 343 |
+
"acc_norm_stderr": 0.04688261722621505
|
| 344 |
+
},
|
| 345 |
+
"hendrycksTest-virology": {
|
| 346 |
+
"acc": 0.4,
|
| 347 |
+
"acc_stderr": 0.04923659639173309,
|
| 348 |
+
"acc_norm": 0.4,
|
| 349 |
+
"acc_norm_stderr": 0.04923659639173309
|
| 350 |
+
},
|
| 351 |
+
"hendrycksTest-world_religions": {
|
| 352 |
+
"acc": 0.74,
|
| 353 |
+
"acc_stderr": 0.04408440022768078,
|
| 354 |
+
"acc_norm": 0.74,
|
| 355 |
+
"acc_norm_stderr": 0.04408440022768078
|
| 356 |
+
}
|
| 357 |
+
},
|
| 358 |
+
"versions": {
|
| 359 |
+
"arc_easy": 0,
|
| 360 |
+
"hellaswag": 0,
|
| 361 |
+
"hendrycksTest-abstract_algebra": 1,
|
| 362 |
+
"hendrycksTest-anatomy": 1,
|
| 363 |
+
"hendrycksTest-astronomy": 1,
|
| 364 |
+
"hendrycksTest-business_ethics": 1,
|
| 365 |
+
"hendrycksTest-clinical_knowledge": 1,
|
| 366 |
+
"hendrycksTest-college_biology": 1,
|
| 367 |
+
"hendrycksTest-college_chemistry": 1,
|
| 368 |
+
"hendrycksTest-college_computer_science": 1,
|
| 369 |
+
"hendrycksTest-college_mathematics": 1,
|
| 370 |
+
"hendrycksTest-college_medicine": 1,
|
| 371 |
+
"hendrycksTest-college_physics": 1,
|
| 372 |
+
"hendrycksTest-computer_security": 1,
|
| 373 |
+
"hendrycksTest-conceptual_physics": 1,
|
| 374 |
+
"hendrycksTest-econometrics": 1,
|
| 375 |
+
"hendrycksTest-electrical_engineering": 1,
|
| 376 |
+
"hendrycksTest-elementary_mathematics": 1,
|
| 377 |
+
"hendrycksTest-formal_logic": 1,
|
| 378 |
+
"hendrycksTest-global_facts": 1,
|
| 379 |
+
"hendrycksTest-high_school_biology": 1,
|
| 380 |
+
"hendrycksTest-high_school_chemistry": 1,
|
| 381 |
+
"hendrycksTest-high_school_computer_science": 1,
|
| 382 |
+
"hendrycksTest-high_school_european_history": 1,
|
| 383 |
+
"hendrycksTest-high_school_geography": 1,
|
| 384 |
+
"hendrycksTest-high_school_government_and_politics": 1,
|
| 385 |
+
"hendrycksTest-high_school_macroeconomics": 1,
|
| 386 |
+
"hendrycksTest-high_school_mathematics": 1,
|
| 387 |
+
"hendrycksTest-high_school_microeconomics": 1,
|
| 388 |
+
"hendrycksTest-high_school_physics": 1,
|
| 389 |
+
"hendrycksTest-high_school_psychology": 1,
|
| 390 |
+
"hendrycksTest-high_school_statistics": 1,
|
| 391 |
+
"hendrycksTest-high_school_us_history": 1,
|
| 392 |
+
"hendrycksTest-high_school_world_history": 1,
|
| 393 |
+
"hendrycksTest-human_aging": 1,
|
| 394 |
+
"hendrycksTest-human_sexuality": 1,
|
| 395 |
+
"hendrycksTest-international_law": 1,
|
| 396 |
+
"hendrycksTest-jurisprudence": 1,
|
| 397 |
+
"hendrycksTest-logical_fallacies": 1,
|
| 398 |
+
"hendrycksTest-machine_learning": 1,
|
| 399 |
+
"hendrycksTest-management": 1,
|
| 400 |
+
"hendrycksTest-marketing": 1,
|
| 401 |
+
"hendrycksTest-medical_genetics": 1,
|
| 402 |
+
"hendrycksTest-miscellaneous": 1,
|
| 403 |
+
"hendrycksTest-moral_disputes": 1,
|
| 404 |
+
"hendrycksTest-moral_scenarios": 1,
|
| 405 |
+
"hendrycksTest-nutrition": 1,
|
| 406 |
+
"hendrycksTest-philosophy": 1,
|
| 407 |
+
"hendrycksTest-prehistory": 1,
|
| 408 |
+
"hendrycksTest-professional_accounting": 1,
|
| 409 |
+
"hendrycksTest-professional_law": 1,
|
| 410 |
+
"hendrycksTest-professional_medicine": 1,
|
| 411 |
+
"hendrycksTest-professional_psychology": 1,
|
| 412 |
+
"hendrycksTest-public_relations": 1,
|
| 413 |
+
"hendrycksTest-security_studies": 1,
|
| 414 |
+
"hendrycksTest-sociology": 1,
|
| 415 |
+
"hendrycksTest-us_foreign_policy": 1,
|
| 416 |
+
"hendrycksTest-virology": 1,
|
| 417 |
+
"hendrycksTest-world_religions": 1
|
| 418 |
+
},
|
| 419 |
+
"config": {
|
| 420 |
+
"model": "hf-causal-experimental",
|
| 421 |
+
"model_args": "pretrained=/workspace/airoboros-l2-7b-2.2.1-bs-test-bs-2,use_accelerate=True",
|
| 422 |
+
"num_fewshot": 5,
|
| 423 |
+
"batch_size": "8",
|
| 424 |
+
"batch_sizes": [],
|
| 425 |
+
"device": null,
|
| 426 |
+
"no_cache": false,
|
| 427 |
+
"limit": 100.0,
|
| 428 |
+
"bootstrap_iters": 100000,
|
| 429 |
+
"description_dict": {}
|
| 430 |
+
}
|
| 431 |
+
}
|