417 lines
12 KiB
JSON
417 lines
12 KiB
JSON
{
|
|
"results": {
|
|
"hendrycksTest-abstract_algebra": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-anatomy": {
|
|
"acc": 0.4,
|
|
"acc_stderr": 0.1,
|
|
"acc_norm": 0.4,
|
|
"acc_norm_stderr": 0.1
|
|
},
|
|
"hendrycksTest-astronomy": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390466,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390466
|
|
},
|
|
"hendrycksTest-business_ethics": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-clinical_knowledge": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081345,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081345
|
|
},
|
|
"hendrycksTest-college_biology": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277262,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277262
|
|
},
|
|
"hendrycksTest-college_chemistry": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-college_computer_science": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390466,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390466
|
|
},
|
|
"hendrycksTest-college_mathematics": {
|
|
"acc": 0.4,
|
|
"acc_stderr": 0.1,
|
|
"acc_norm": 0.4,
|
|
"acc_norm_stderr": 0.1
|
|
},
|
|
"hendrycksTest-college_medicine": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-college_physics": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-computer_security": {
|
|
"acc": 0.52,
|
|
"acc_stderr": 0.10198039027185572,
|
|
"acc_norm": 0.52,
|
|
"acc_norm_stderr": 0.10198039027185572
|
|
},
|
|
"hendrycksTest-conceptual_physics": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081347,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081347
|
|
},
|
|
"hendrycksTest-econometrics": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-electrical_engineering": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.07483314773547882,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.07483314773547882
|
|
},
|
|
"hendrycksTest-elementary_mathematics": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-formal_logic": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081345,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081345
|
|
},
|
|
"hendrycksTest-global_facts": {
|
|
"acc": 0.4,
|
|
"acc_stderr": 0.1,
|
|
"acc_norm": 0.4,
|
|
"acc_norm_stderr": 0.1
|
|
},
|
|
"hendrycksTest-high_school_biology": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390465,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390465
|
|
},
|
|
"hendrycksTest-high_school_chemistry": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-high_school_computer_science": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-high_school_european_history": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-high_school_geography": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390466,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390466
|
|
},
|
|
"hendrycksTest-high_school_government_and_politics": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911677,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911677
|
|
},
|
|
"hendrycksTest-high_school_macroeconomics": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911681,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911681
|
|
},
|
|
"hendrycksTest-high_school_mathematics": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390467,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390467
|
|
},
|
|
"hendrycksTest-high_school_microeconomics": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911678,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911678
|
|
},
|
|
"hendrycksTest-high_school_physics": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-high_school_psychology": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911678,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911678
|
|
},
|
|
"hendrycksTest-high_school_statistics": {
|
|
"acc": 0.4,
|
|
"acc_stderr": 0.10000000000000002,
|
|
"acc_norm": 0.4,
|
|
"acc_norm_stderr": 0.10000000000000002
|
|
},
|
|
"hendrycksTest-high_school_us_history": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.0916515138991168,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.0916515138991168
|
|
},
|
|
"hendrycksTest-high_school_world_history": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.0916515138991168,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.0916515138991168
|
|
},
|
|
"hendrycksTest-human_aging": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-human_sexuality": {
|
|
"acc": 0.44,
|
|
"acc_stderr": 0.10132456102380442,
|
|
"acc_norm": 0.44,
|
|
"acc_norm_stderr": 0.10132456102380442
|
|
},
|
|
"hendrycksTest-international_law": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390466,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390466
|
|
},
|
|
"hendrycksTest-jurisprudence": {
|
|
"acc": 0.36,
|
|
"acc_stderr": 0.09797958971132713,
|
|
"acc_norm": 0.36,
|
|
"acc_norm_stderr": 0.09797958971132713
|
|
},
|
|
"hendrycksTest-logical_fallacies": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081345,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081345
|
|
},
|
|
"hendrycksTest-machine_learning": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.0748331477354788,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.0748331477354788
|
|
},
|
|
"hendrycksTest-management": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390465,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390465
|
|
},
|
|
"hendrycksTest-marketing": {
|
|
"acc": 0.08,
|
|
"acc_stderr": 0.05537749241945382,
|
|
"acc_norm": 0.08,
|
|
"acc_norm_stderr": 0.05537749241945382
|
|
},
|
|
"hendrycksTest-medical_genetics": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911678,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911678
|
|
},
|
|
"hendrycksTest-miscellaneous": {
|
|
"acc": 0.08,
|
|
"acc_stderr": 0.05537749241945382,
|
|
"acc_norm": 0.08,
|
|
"acc_norm_stderr": 0.05537749241945382
|
|
},
|
|
"hendrycksTest-moral_disputes": {
|
|
"acc": 0.12,
|
|
"acc_stderr": 0.066332495807108,
|
|
"acc_norm": 0.12,
|
|
"acc_norm_stderr": 0.066332495807108
|
|
},
|
|
"hendrycksTest-moral_scenarios": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-nutrition": {
|
|
"acc": 0.52,
|
|
"acc_stderr": 0.10198039027185572,
|
|
"acc_norm": 0.52,
|
|
"acc_norm_stderr": 0.10198039027185572
|
|
},
|
|
"hendrycksTest-philosophy": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-prehistory": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.07483314773547882,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.07483314773547882
|
|
},
|
|
"hendrycksTest-professional_accounting": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390465,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390465
|
|
},
|
|
"hendrycksTest-professional_law": {
|
|
"acc": 0.16,
|
|
"acc_stderr": 0.07483314773547882,
|
|
"acc_norm": 0.16,
|
|
"acc_norm_stderr": 0.07483314773547882
|
|
},
|
|
"hendrycksTest-professional_medicine": {
|
|
"acc": 0.4,
|
|
"acc_stderr": 0.10000000000000002,
|
|
"acc_norm": 0.4,
|
|
"acc_norm_stderr": 0.10000000000000002
|
|
},
|
|
"hendrycksTest-professional_psychology": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081345,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081345
|
|
},
|
|
"hendrycksTest-public_relations": {
|
|
"acc": 0.2,
|
|
"acc_stderr": 0.08164965809277261,
|
|
"acc_norm": 0.2,
|
|
"acc_norm_stderr": 0.08164965809277261
|
|
},
|
|
"hendrycksTest-security_studies": {
|
|
"acc": 0.32,
|
|
"acc_stderr": 0.09521904571390466,
|
|
"acc_norm": 0.32,
|
|
"acc_norm_stderr": 0.09521904571390466
|
|
},
|
|
"hendrycksTest-sociology": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911678,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911678
|
|
},
|
|
"hendrycksTest-us_foreign_policy": {
|
|
"acc": 0.24,
|
|
"acc_stderr": 0.08717797887081345,
|
|
"acc_norm": 0.24,
|
|
"acc_norm_stderr": 0.08717797887081345
|
|
},
|
|
"hendrycksTest-virology": {
|
|
"acc": 0.12,
|
|
"acc_stderr": 0.06633249580710801,
|
|
"acc_norm": 0.12,
|
|
"acc_norm_stderr": 0.06633249580710801
|
|
},
|
|
"hendrycksTest-world_religions": {
|
|
"acc": 0.28,
|
|
"acc_stderr": 0.09165151389911678,
|
|
"acc_norm": 0.28,
|
|
"acc_norm_stderr": 0.09165151389911678
|
|
}
|
|
},
|
|
"versions": {
|
|
"hendrycksTest-abstract_algebra": 1,
|
|
"hendrycksTest-anatomy": 1,
|
|
"hendrycksTest-astronomy": 1,
|
|
"hendrycksTest-business_ethics": 1,
|
|
"hendrycksTest-clinical_knowledge": 1,
|
|
"hendrycksTest-college_biology": 1,
|
|
"hendrycksTest-college_chemistry": 1,
|
|
"hendrycksTest-college_computer_science": 1,
|
|
"hendrycksTest-college_mathematics": 1,
|
|
"hendrycksTest-college_medicine": 1,
|
|
"hendrycksTest-college_physics": 1,
|
|
"hendrycksTest-computer_security": 1,
|
|
"hendrycksTest-conceptual_physics": 1,
|
|
"hendrycksTest-econometrics": 1,
|
|
"hendrycksTest-electrical_engineering": 1,
|
|
"hendrycksTest-elementary_mathematics": 1,
|
|
"hendrycksTest-formal_logic": 1,
|
|
"hendrycksTest-global_facts": 1,
|
|
"hendrycksTest-high_school_biology": 1,
|
|
"hendrycksTest-high_school_chemistry": 1,
|
|
"hendrycksTest-high_school_computer_science": 1,
|
|
"hendrycksTest-high_school_european_history": 1,
|
|
"hendrycksTest-high_school_geography": 1,
|
|
"hendrycksTest-high_school_government_and_politics": 1,
|
|
"hendrycksTest-high_school_macroeconomics": 1,
|
|
"hendrycksTest-high_school_mathematics": 1,
|
|
"hendrycksTest-high_school_microeconomics": 1,
|
|
"hendrycksTest-high_school_physics": 1,
|
|
"hendrycksTest-high_school_psychology": 1,
|
|
"hendrycksTest-high_school_statistics": 1,
|
|
"hendrycksTest-high_school_us_history": 1,
|
|
"hendrycksTest-high_school_world_history": 1,
|
|
"hendrycksTest-human_aging": 1,
|
|
"hendrycksTest-human_sexuality": 1,
|
|
"hendrycksTest-international_law": 1,
|
|
"hendrycksTest-jurisprudence": 1,
|
|
"hendrycksTest-logical_fallacies": 1,
|
|
"hendrycksTest-machine_learning": 1,
|
|
"hendrycksTest-management": 1,
|
|
"hendrycksTest-marketing": 1,
|
|
"hendrycksTest-medical_genetics": 1,
|
|
"hendrycksTest-miscellaneous": 1,
|
|
"hendrycksTest-moral_disputes": 1,
|
|
"hendrycksTest-moral_scenarios": 1,
|
|
"hendrycksTest-nutrition": 1,
|
|
"hendrycksTest-philosophy": 1,
|
|
"hendrycksTest-prehistory": 1,
|
|
"hendrycksTest-professional_accounting": 1,
|
|
"hendrycksTest-professional_law": 1,
|
|
"hendrycksTest-professional_medicine": 1,
|
|
"hendrycksTest-professional_psychology": 1,
|
|
"hendrycksTest-public_relations": 1,
|
|
"hendrycksTest-security_studies": 1,
|
|
"hendrycksTest-sociology": 1,
|
|
"hendrycksTest-us_foreign_policy": 1,
|
|
"hendrycksTest-virology": 1,
|
|
"hendrycksTest-world_religions": 1
|
|
},
|
|
"config": {
|
|
"model": "hf-causal-experimental",
|
|
"model_args": "pretrained=BEE-spoke-data/smol_llama-220M-GQA,revision=main,trust_remote_code=True,dtype='bfloat16'",
|
|
"num_fewshot": 5,
|
|
"batch_size": "8",
|
|
"batch_sizes": [],
|
|
"device": "cuda",
|
|
"no_cache": false,
|
|
"limit": 0.25,
|
|
"bootstrap_iters": 100000,
|
|
"description_dict": {}
|
|
}
|
|
} |