Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
| { | |
| "results": { | |
| "harness|arc:challenge|25": { | |
| "acc": 0.6049488054607508, | |
| "acc_stderr": 0.01428589829293817, | |
| "acc_norm": 0.6459044368600683, | |
| "acc_norm_stderr": 0.013975454122756564 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6693885680143398, | |
| "acc_stderr": 0.004694718918225751, | |
| "acc_norm": 0.8587930691097391, | |
| "acc_norm_stderr": 0.003475231889452833 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.3561811505507956, | |
| "mc1_stderr": 0.016763790728446335, | |
| "mc2": 0.5280473232260097, | |
| "mc2_stderr": 0.01553022126123046 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.35, | |
| "acc_stderr": 0.04793724854411021, | |
| "acc_norm": 0.35, | |
| "acc_norm_stderr": 0.04793724854411021 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.5185185185185185, | |
| "acc_stderr": 0.043163785995113245, | |
| "acc_norm": 0.5185185185185185, | |
| "acc_norm_stderr": 0.043163785995113245 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.7302631578947368, | |
| "acc_stderr": 0.03611780560284898, | |
| "acc_norm": 0.7302631578947368, | |
| "acc_norm_stderr": 0.03611780560284898 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.65, | |
| "acc_stderr": 0.0479372485441102, | |
| "acc_norm": 0.65, | |
| "acc_norm_stderr": 0.0479372485441102 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.6377358490566037, | |
| "acc_stderr": 0.029582245128384303, | |
| "acc_norm": 0.6377358490566037, | |
| "acc_norm_stderr": 0.029582245128384303 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.75, | |
| "acc_stderr": 0.03621034121889507, | |
| "acc_norm": 0.75, | |
| "acc_norm_stderr": 0.03621034121889507 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.48, | |
| "acc_stderr": 0.050211673156867795, | |
| "acc_norm": 0.48, | |
| "acc_norm_stderr": 0.050211673156867795 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.59, | |
| "acc_stderr": 0.04943110704237101, | |
| "acc_norm": 0.59, | |
| "acc_norm_stderr": 0.04943110704237101 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "acc": 0.34, | |
| "acc_stderr": 0.04760952285695235, | |
| "acc_norm": 0.34, | |
| "acc_norm_stderr": 0.04760952285695235 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "acc": 0.6011560693641619, | |
| "acc_stderr": 0.0373362665538351, | |
| "acc_norm": 0.6011560693641619, | |
| "acc_norm_stderr": 0.0373362665538351 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.3333333333333333, | |
| "acc_stderr": 0.04690650298201943, | |
| "acc_norm": 0.3333333333333333, | |
| "acc_norm_stderr": 0.04690650298201943 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.71, | |
| "acc_stderr": 0.045604802157206845, | |
| "acc_norm": 0.71, | |
| "acc_norm_stderr": 0.045604802157206845 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.5829787234042553, | |
| "acc_stderr": 0.032232762667117124, | |
| "acc_norm": 0.5829787234042553, | |
| "acc_norm_stderr": 0.032232762667117124 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.41228070175438597, | |
| "acc_stderr": 0.04630653203366595, | |
| "acc_norm": 0.41228070175438597, | |
| "acc_norm_stderr": 0.04630653203366595 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.5793103448275863, | |
| "acc_stderr": 0.0411391498118926, | |
| "acc_norm": 0.5793103448275863, | |
| "acc_norm_stderr": 0.0411391498118926 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.41005291005291006, | |
| "acc_stderr": 0.02533120243894442, | |
| "acc_norm": 0.41005291005291006, | |
| "acc_norm_stderr": 0.02533120243894442 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.4126984126984127, | |
| "acc_stderr": 0.04403438954768176, | |
| "acc_norm": 0.4126984126984127, | |
| "acc_norm_stderr": 0.04403438954768176 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.43, | |
| "acc_stderr": 0.049756985195624284, | |
| "acc_norm": 0.43, | |
| "acc_norm_stderr": 0.049756985195624284 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.7645161290322581, | |
| "acc_stderr": 0.02413763242933771, | |
| "acc_norm": 0.7645161290322581, | |
| "acc_norm_stderr": 0.02413763242933771 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "acc": 0.4630541871921182, | |
| "acc_stderr": 0.035083705204426656, | |
| "acc_norm": 0.4630541871921182, | |
| "acc_norm_stderr": 0.035083705204426656 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "acc": 0.65, | |
| "acc_stderr": 0.047937248544110196, | |
| "acc_norm": 0.65, | |
| "acc_norm_stderr": 0.047937248544110196 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.8181818181818182, | |
| "acc_stderr": 0.03011768892950359, | |
| "acc_norm": 0.8181818181818182, | |
| "acc_norm_stderr": 0.03011768892950359 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.8080808080808081, | |
| "acc_stderr": 0.02805779167298902, | |
| "acc_norm": 0.8080808080808081, | |
| "acc_norm_stderr": 0.02805779167298902 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.8911917098445595, | |
| "acc_stderr": 0.022473253332768783, | |
| "acc_norm": 0.8911917098445595, | |
| "acc_norm_stderr": 0.022473253332768783 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.6410256410256411, | |
| "acc_stderr": 0.02432173848460235, | |
| "acc_norm": 0.6410256410256411, | |
| "acc_norm_stderr": 0.02432173848460235 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.3, | |
| "acc_stderr": 0.027940457136228416, | |
| "acc_norm": 0.3, | |
| "acc_norm_stderr": 0.027940457136228416 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.6596638655462185, | |
| "acc_stderr": 0.030778057422931673, | |
| "acc_norm": 0.6596638655462185, | |
| "acc_norm_stderr": 0.030778057422931673 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.423841059602649, | |
| "acc_stderr": 0.04034846678603397, | |
| "acc_norm": 0.423841059602649, | |
| "acc_norm_stderr": 0.04034846678603397 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.8385321100917431, | |
| "acc_stderr": 0.015776239256163255, | |
| "acc_norm": 0.8385321100917431, | |
| "acc_norm_stderr": 0.015776239256163255 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.48148148148148145, | |
| "acc_stderr": 0.03407632093854052, | |
| "acc_norm": 0.48148148148148145, | |
| "acc_norm_stderr": 0.03407632093854052 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.8578431372549019, | |
| "acc_stderr": 0.024509803921568606, | |
| "acc_norm": 0.8578431372549019, | |
| "acc_norm_stderr": 0.024509803921568606 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.8438818565400844, | |
| "acc_stderr": 0.02362715946031867, | |
| "acc_norm": 0.8438818565400844, | |
| "acc_norm_stderr": 0.02362715946031867 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.726457399103139, | |
| "acc_stderr": 0.02991858670779883, | |
| "acc_norm": 0.726457399103139, | |
| "acc_norm_stderr": 0.02991858670779883 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.7099236641221374, | |
| "acc_stderr": 0.039800662464677665, | |
| "acc_norm": 0.7099236641221374, | |
| "acc_norm_stderr": 0.039800662464677665 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.8016528925619835, | |
| "acc_stderr": 0.03640118271990946, | |
| "acc_norm": 0.8016528925619835, | |
| "acc_norm_stderr": 0.03640118271990946 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.8240740740740741, | |
| "acc_stderr": 0.036809181416738807, | |
| "acc_norm": 0.8240740740740741, | |
| "acc_norm_stderr": 0.036809181416738807 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.7607361963190185, | |
| "acc_stderr": 0.033519538795212696, | |
| "acc_norm": 0.7607361963190185, | |
| "acc_norm_stderr": 0.033519538795212696 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.48214285714285715, | |
| "acc_stderr": 0.047427623612430116, | |
| "acc_norm": 0.48214285714285715, | |
| "acc_norm_stderr": 0.047427623612430116 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.8058252427184466, | |
| "acc_stderr": 0.03916667762822584, | |
| "acc_norm": 0.8058252427184466, | |
| "acc_norm_stderr": 0.03916667762822584 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.8717948717948718, | |
| "acc_stderr": 0.02190190511507332, | |
| "acc_norm": 0.8717948717948718, | |
| "acc_norm_stderr": 0.02190190511507332 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.65, | |
| "acc_stderr": 0.047937248544110196, | |
| "acc_norm": 0.65, | |
| "acc_norm_stderr": 0.047937248544110196 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.8275862068965517, | |
| "acc_stderr": 0.013507943909371798, | |
| "acc_norm": 0.8275862068965517, | |
| "acc_norm_stderr": 0.013507943909371798 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.7167630057803468, | |
| "acc_stderr": 0.02425790170532338, | |
| "acc_norm": 0.7167630057803468, | |
| "acc_norm_stderr": 0.02425790170532338 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.39553072625698327, | |
| "acc_stderr": 0.01635341541007577, | |
| "acc_norm": 0.39553072625698327, | |
| "acc_norm_stderr": 0.01635341541007577 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.6993464052287581, | |
| "acc_stderr": 0.026256053835718968, | |
| "acc_norm": 0.6993464052287581, | |
| "acc_norm_stderr": 0.026256053835718968 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.7041800643086816, | |
| "acc_stderr": 0.02592237178881877, | |
| "acc_norm": 0.7041800643086816, | |
| "acc_norm_stderr": 0.02592237178881877 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.7098765432098766, | |
| "acc_stderr": 0.025251173936495036, | |
| "acc_norm": 0.7098765432098766, | |
| "acc_norm_stderr": 0.025251173936495036 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.5070921985815603, | |
| "acc_stderr": 0.02982449855912901, | |
| "acc_norm": 0.5070921985815603, | |
| "acc_norm_stderr": 0.02982449855912901 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.4771838331160365, | |
| "acc_stderr": 0.012756933382823694, | |
| "acc_norm": 0.4771838331160365, | |
| "acc_norm_stderr": 0.012756933382823694 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.5772058823529411, | |
| "acc_stderr": 0.030008562845003476, | |
| "acc_norm": 0.5772058823529411, | |
| "acc_norm_stderr": 0.030008562845003476 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.6699346405228758, | |
| "acc_stderr": 0.019023726160724556, | |
| "acc_norm": 0.6699346405228758, | |
| "acc_norm_stderr": 0.019023726160724556 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "acc": 0.6909090909090909, | |
| "acc_stderr": 0.044262946482000985, | |
| "acc_norm": 0.6909090909090909, | |
| "acc_norm_stderr": 0.044262946482000985 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "acc": 0.7877551020408163, | |
| "acc_stderr": 0.026176967197866767, | |
| "acc_norm": 0.7877551020408163, | |
| "acc_norm_stderr": 0.026176967197866767 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.8706467661691543, | |
| "acc_stderr": 0.023729830881018526, | |
| "acc_norm": 0.8706467661691543, | |
| "acc_norm_stderr": 0.023729830881018526 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.87, | |
| "acc_stderr": 0.03379976689896309, | |
| "acc_norm": 0.87, | |
| "acc_norm_stderr": 0.03379976689896309 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.5120481927710844, | |
| "acc_stderr": 0.03891364495835817, | |
| "acc_norm": 0.5120481927710844, | |
| "acc_norm_stderr": 0.03891364495835817 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.8187134502923976, | |
| "acc_stderr": 0.029547741687640038, | |
| "acc_norm": 0.8187134502923976, | |
| "acc_norm_stderr": 0.029547741687640038 | |
| }, | |
| "all": { | |
| "acc": 0.6390701952816291, | |
| "acc_stderr": 0.03365809160773111, | |
| "acc_norm": 0.6390701952816291, | |
| "acc_norm_stderr": 0.03365809160773111 | |
| } | |
| }, | |
| "versions": { | |
| "harness|arc:challenge|25": 0, | |
| "harness|hellaswag|10": 0, | |
| "harness|truthfulqa:mc|0": 1, | |
| "harness|hendrycksTest-abstract_algebra|5": 1, | |
| "harness|hendrycksTest-anatomy|5": 1, | |
| "harness|hendrycksTest-astronomy|5": 1, | |
| "harness|hendrycksTest-business_ethics|5": 1, | |
| "harness|hendrycksTest-clinical_knowledge|5": 1, | |
| "harness|hendrycksTest-college_biology|5": 1, | |
| "harness|hendrycksTest-college_chemistry|5": 1, | |
| "harness|hendrycksTest-college_computer_science|5": 1, | |
| "harness|hendrycksTest-college_mathematics|5": 1, | |
| "harness|hendrycksTest-college_medicine|5": 1, | |
| "harness|hendrycksTest-college_physics|5": 1, | |
| "harness|hendrycksTest-computer_security|5": 1, | |
| "harness|hendrycksTest-conceptual_physics|5": 1, | |
| "harness|hendrycksTest-econometrics|5": 1, | |
| "harness|hendrycksTest-electrical_engineering|5": 1, | |
| "harness|hendrycksTest-elementary_mathematics|5": 1, | |
| "harness|hendrycksTest-formal_logic|5": 1, | |
| "harness|hendrycksTest-global_facts|5": 1, | |
| "harness|hendrycksTest-high_school_biology|5": 1, | |
| "harness|hendrycksTest-high_school_chemistry|5": 1, | |
| "harness|hendrycksTest-high_school_computer_science|5": 1, | |
| "harness|hendrycksTest-high_school_european_history|5": 1, | |
| "harness|hendrycksTest-high_school_geography|5": 1, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": 1, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": 1, | |
| "harness|hendrycksTest-high_school_mathematics|5": 1, | |
| "harness|hendrycksTest-high_school_microeconomics|5": 1, | |
| "harness|hendrycksTest-high_school_physics|5": 1, | |
| "harness|hendrycksTest-high_school_psychology|5": 1, | |
| "harness|hendrycksTest-high_school_statistics|5": 1, | |
| "harness|hendrycksTest-high_school_us_history|5": 1, | |
| "harness|hendrycksTest-high_school_world_history|5": 1, | |
| "harness|hendrycksTest-human_aging|5": 1, | |
| "harness|hendrycksTest-human_sexuality|5": 1, | |
| "harness|hendrycksTest-international_law|5": 1, | |
| "harness|hendrycksTest-jurisprudence|5": 1, | |
| "harness|hendrycksTest-logical_fallacies|5": 1, | |
| "harness|hendrycksTest-machine_learning|5": 1, | |
| "harness|hendrycksTest-management|5": 1, | |
| "harness|hendrycksTest-marketing|5": 1, | |
| "harness|hendrycksTest-medical_genetics|5": 1, | |
| "harness|hendrycksTest-miscellaneous|5": 1, | |
| "harness|hendrycksTest-moral_disputes|5": 1, | |
| "harness|hendrycksTest-moral_scenarios|5": 1, | |
| "harness|hendrycksTest-nutrition|5": 1, | |
| "harness|hendrycksTest-philosophy|5": 1, | |
| "harness|hendrycksTest-prehistory|5": 1, | |
| "harness|hendrycksTest-professional_accounting|5": 1, | |
| "harness|hendrycksTest-professional_law|5": 1, | |
| "harness|hendrycksTest-professional_medicine|5": 1, | |
| "harness|hendrycksTest-professional_psychology|5": 1, | |
| "harness|hendrycksTest-public_relations|5": 1, | |
| "harness|hendrycksTest-security_studies|5": 1, | |
| "harness|hendrycksTest-sociology|5": 1, | |
| "harness|hendrycksTest-us_foreign_policy|5": 1, | |
| "harness|hendrycksTest-virology|5": 1, | |
| "harness|hendrycksTest-world_religions|5": 1, | |
| "all": 0 | |
| }, | |
| "config": { | |
| "model_name": "meta-llama/Llama-2-70b-chat-hf", | |
| "model_sha": "7f54101c0fbb67a8143ca23eb8bd09b71f269c74", | |
| "model_dtype": "torch.float16", | |
| "lighteval_sha": "43cff840721bd0214adb4e29236a5e2ca1813937", | |
| "num_few_shot_default": 0, | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null | |
| }, | |
| "task_config": { | |
| "harness|arc:challenge": "LM Harness task", | |
| "harness|hellaswag": "LM Harness task", | |
| "harness|truthfulqa:mc": "LM Harness task", | |
| "harness|hendrycksTest-abstract_algebra": "LM Harness task", | |
| "harness|hendrycksTest-anatomy": "LM Harness task", | |
| "harness|hendrycksTest-astronomy": "LM Harness task", | |
| "harness|hendrycksTest-business_ethics": "LM Harness task", | |
| "harness|hendrycksTest-clinical_knowledge": "LM Harness task", | |
| "harness|hendrycksTest-college_biology": "LM Harness task", | |
| "harness|hendrycksTest-college_chemistry": "LM Harness task", | |
| "harness|hendrycksTest-college_computer_science": "LM Harness task", | |
| "harness|hendrycksTest-college_mathematics": "LM Harness task", | |
| "harness|hendrycksTest-college_medicine": "LM Harness task", | |
| "harness|hendrycksTest-college_physics": "LM Harness task", | |
| "harness|hendrycksTest-computer_security": "LM Harness task", | |
| "harness|hendrycksTest-conceptual_physics": "LM Harness task", | |
| "harness|hendrycksTest-econometrics": "LM Harness task", | |
| "harness|hendrycksTest-electrical_engineering": "LM Harness task", | |
| "harness|hendrycksTest-elementary_mathematics": "LM Harness task", | |
| "harness|hendrycksTest-formal_logic": "LM Harness task", | |
| "harness|hendrycksTest-global_facts": "LM Harness task", | |
| "harness|hendrycksTest-high_school_biology": "LM Harness task", | |
| "harness|hendrycksTest-high_school_chemistry": "LM Harness task", | |
| "harness|hendrycksTest-high_school_computer_science": "LM Harness task", | |
| "harness|hendrycksTest-high_school_european_history": "LM Harness task", | |
| "harness|hendrycksTest-high_school_geography": "LM Harness task", | |
| "harness|hendrycksTest-high_school_government_and_politics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_macroeconomics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_mathematics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_microeconomics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_physics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_psychology": "LM Harness task", | |
| "harness|hendrycksTest-high_school_statistics": "LM Harness task", | |
| "harness|hendrycksTest-high_school_us_history": "LM Harness task", | |
| "harness|hendrycksTest-high_school_world_history": "LM Harness task", | |
| "harness|hendrycksTest-human_aging": "LM Harness task", | |
| "harness|hendrycksTest-human_sexuality": "LM Harness task", | |
| "harness|hendrycksTest-international_law": "LM Harness task", | |
| "harness|hendrycksTest-jurisprudence": "LM Harness task", | |
| "harness|hendrycksTest-logical_fallacies": "LM Harness task", | |
| "harness|hendrycksTest-machine_learning": "LM Harness task", | |
| "harness|hendrycksTest-management": "LM Harness task", | |
| "harness|hendrycksTest-marketing": "LM Harness task", | |
| "harness|hendrycksTest-medical_genetics": "LM Harness task", | |
| "harness|hendrycksTest-miscellaneous": "LM Harness task", | |
| "harness|hendrycksTest-moral_disputes": "LM Harness task", | |
| "harness|hendrycksTest-moral_scenarios": "LM Harness task", | |
| "harness|hendrycksTest-nutrition": "LM Harness task", | |
| "harness|hendrycksTest-philosophy": "LM Harness task", | |
| "harness|hendrycksTest-prehistory": "LM Harness task", | |
| "harness|hendrycksTest-professional_accounting": "LM Harness task", | |
| "harness|hendrycksTest-professional_law": "LM Harness task", | |
| "harness|hendrycksTest-professional_medicine": "LM Harness task", | |
| "harness|hendrycksTest-professional_psychology": "LM Harness task", | |
| "harness|hendrycksTest-public_relations": "LM Harness task", | |
| "harness|hendrycksTest-security_studies": "LM Harness task", | |
| "harness|hendrycksTest-sociology": "LM Harness task", | |
| "harness|hendrycksTest-us_foreign_policy": "LM Harness task", | |
| "harness|hendrycksTest-virology": "LM Harness task", | |
| "harness|hendrycksTest-world_religions": "LM Harness task" | |
| }, | |
| "hashes": { | |
| "harness|arc:challenge|25": { | |
| "hash_examples": "fb8c51b1872daeda", | |
| "hash_full_prompts": "045cbb916e5145c6", | |
| "hash_input_tokens": "fab18a8dbccd885e", | |
| "hash_cont_tokens": "e8abf848493b50f7" | |
| }, | |
| "harness|hellaswag|10": { | |
| "hash_examples": "e1768ecb99d7ecf0", | |
| "hash_full_prompts": "0b4c16983130f84f", | |
| "hash_input_tokens": "fd3d11be48664a7e", | |
| "hash_cont_tokens": "9fe0a5c42e1532db" | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "hash_examples": "23176c0531c7b867", | |
| "hash_full_prompts": "36a6d90e75d92d4a", | |
| "hash_input_tokens": "e3c2231820d87234", | |
| "hash_cont_tokens": "f5da56a132aab151" | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "hash_examples": "280f9f325b40559a", | |
| "hash_full_prompts": "2f776a367d23aea2", | |
| "hash_input_tokens": "c3792fce2534965f", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "hash_examples": "2f83a4f1cab4ba18", | |
| "hash_full_prompts": "516f74bef25df620", | |
| "hash_input_tokens": "1bfeea5736b995ee", | |
| "hash_cont_tokens": "f11971a765cb609f" | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "hash_examples": "7d587b908da4d762", | |
| "hash_full_prompts": "faf4e80f65de93ca", | |
| "hash_input_tokens": "c4b2f1160f746871", | |
| "hash_cont_tokens": "440a970fadecdc7b" | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "hash_examples": "33e51740670de686", | |
| "hash_full_prompts": "db01c3ef8e1479d4", | |
| "hash_input_tokens": "b98d6ef1d1e2e17b", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "hash_examples": "f3366dbe7eefffa4", | |
| "hash_full_prompts": "49654f71d94b65c3", | |
| "hash_input_tokens": "9851119dacda883c", | |
| "hash_cont_tokens": "7ecd60c25b9bfe5b" | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "hash_examples": "ca2b6753a0193e7f", | |
| "hash_full_prompts": "2b460b75f1fdfefd", | |
| "hash_input_tokens": "81a92a54cddefc2f", | |
| "hash_cont_tokens": "875cde3af7a0ee14" | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "hash_examples": "22ff85f1d34f42d1", | |
| "hash_full_prompts": "242c9be6da583e95", | |
| "hash_input_tokens": "fd4c0cebdc2c1c3d", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "hash_examples": "30318289d717a5cf", | |
| "hash_full_prompts": "ed2bdb4e87c4b371", | |
| "hash_input_tokens": "49f6021f4c075e0d", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "hash_examples": "4944d1f0b6b5d911", | |
| "hash_full_prompts": "770bc4281c973190", | |
| "hash_input_tokens": "db61bad69399bfe8", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "hash_examples": "dd69cc33381275af", | |
| "hash_full_prompts": "ad2a53e5250ab46e", | |
| "hash_input_tokens": "c458392f38424d77", | |
| "hash_cont_tokens": "702fb6d82ff0d6ac" | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "hash_examples": "875dd26d22655b0d", | |
| "hash_full_prompts": "833a0d7b55aed500", | |
| "hash_input_tokens": "49cf4d8d8696b588", | |
| "hash_cont_tokens": "f7b8097afc16a47c" | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "hash_examples": "006451eedc0ededb", | |
| "hash_full_prompts": "94034c97e85d8f46", | |
| "hash_input_tokens": "e81d46ca85fa2b7c", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "hash_examples": "8874ece872d2ca4c", | |
| "hash_full_prompts": "e40d15a34640d6fa", | |
| "hash_input_tokens": "d5e231a26622e7d5", | |
| "hash_cont_tokens": "aa0e8bc655f2f641" | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "hash_examples": "64d3623b0bfaa43f", | |
| "hash_full_prompts": "612f340fae41338d", | |
| "hash_input_tokens": "afa3603fd1622706", | |
| "hash_cont_tokens": "b1cc6e7e9fcd3827" | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "hash_examples": "e98f51780c674d7e", | |
| "hash_full_prompts": "10275b312d812ae6", | |
| "hash_input_tokens": "e0c62cf84ed22e7e", | |
| "hash_cont_tokens": "2425a3f084a591ef" | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "hash_examples": "fc48208a5ac1c0ce", | |
| "hash_full_prompts": "5ec274c6c82aca23", | |
| "hash_input_tokens": "303123d2b857f30b", | |
| "hash_cont_tokens": "bd87bf0c060fd925" | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "hash_examples": "5a6525665f63ea72", | |
| "hash_full_prompts": "07b92638c4a6b500", | |
| "hash_input_tokens": "3fd8073b90b9736d", | |
| "hash_cont_tokens": "eb8932890e0605db" | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "hash_examples": "371d70d743b2b89b", | |
| "hash_full_prompts": "332fdee50a1921b4", | |
| "hash_input_tokens": "f65051acd3210902", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "hash_examples": "a79e1018b1674052", | |
| "hash_full_prompts": "e624e26ede922561", | |
| "hash_input_tokens": "264263fc8c2123bc", | |
| "hash_cont_tokens": "1ddcb86d28cde266" | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "hash_examples": "44bfc25c389f0e03", | |
| "hash_full_prompts": "0e3e5f5d9246482a", | |
| "hash_input_tokens": "42e1a18523b075e7", | |
| "hash_cont_tokens": "176c8dcff38c5f8f" | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "hash_examples": "8b8cdb1084f24169", | |
| "hash_full_prompts": "c00487e67c1813cc", | |
| "hash_input_tokens": "6f109fbd505d364b", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "hash_examples": "11cd32d0ef440171", | |
| "hash_full_prompts": "318f4513c537c6bf", | |
| "hash_input_tokens": "f1f73dd687da18d7", | |
| "hash_cont_tokens": "674fc454bdc5ac93" | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "hash_examples": "b60019b9e80b642f", | |
| "hash_full_prompts": "ee5789fcc1a81b1e", | |
| "hash_input_tokens": "575ea4d290807e79", | |
| "hash_cont_tokens": "03a5012b916274ea" | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "hash_examples": "d221ec983d143dc3", | |
| "hash_full_prompts": "ac42d888e1ce1155", | |
| "hash_input_tokens": "5954aff17f30959c", | |
| "hash_cont_tokens": "873d2aab226ba1d8" | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "hash_examples": "59c2915cacfd3fbb", | |
| "hash_full_prompts": "c6bd9d25158abd0e", | |
| "hash_input_tokens": "cc4bb974def176ee", | |
| "hash_cont_tokens": "c583432ad27fcfe0" | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "hash_examples": "1f8ac897608de342", | |
| "hash_full_prompts": "5d88f41fc2d643a8", | |
| "hash_input_tokens": "94100bcb23e1a13e", | |
| "hash_cont_tokens": "d7907b61bcb8c123" | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "hash_examples": "ead6a0f2f6c83370", | |
| "hash_full_prompts": "bfc393381298609e", | |
| "hash_input_tokens": "129c79724487131d", | |
| "hash_cont_tokens": "f47f041de50333b9" | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "hash_examples": "c3f2025990afec64", | |
| "hash_full_prompts": "fc78b4997e436734", | |
| "hash_input_tokens": "82c2ac81ad5b141c", | |
| "hash_cont_tokens": "0d56317b3e5eedb5" | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "hash_examples": "21f8aab618f6d636", | |
| "hash_full_prompts": "d5c76aa40b9dbc43", | |
| "hash_input_tokens": "422b8bb7add88cc5", | |
| "hash_cont_tokens": "09ba1243e7390c0f" | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "hash_examples": "2386a60a11fc5de3", | |
| "hash_full_prompts": "4c5c8be5aafac432", | |
| "hash_input_tokens": "d3e6f7198120fbdc", | |
| "hash_cont_tokens": "9cc29889c3d3f77d" | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "hash_examples": "74961543be40f04f", | |
| "hash_full_prompts": "5d5ca4840131ba21", | |
| "hash_input_tokens": "50c9ff438c85a69e", | |
| "hash_cont_tokens": "cdd0b3dc06d933e5" | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "hash_examples": "2ad2f6b7198b2234", | |
| "hash_full_prompts": "11845057459afd72", | |
| "hash_input_tokens": "054824cc474caef5", | |
| "hash_cont_tokens": "e02816433ff28daf" | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "hash_examples": "1a7199dc733e779b", | |
| "hash_full_prompts": "756b9096b8eaf892", | |
| "hash_input_tokens": "151f31a573d81257", | |
| "hash_cont_tokens": "142a4a8a1138a214" | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "hash_examples": "7acb8fdad97f88a6", | |
| "hash_full_prompts": "731a52ff15b8cfdb", | |
| "hash_input_tokens": "b77763767fb18cc4", | |
| "hash_cont_tokens": "bc54813e809b796d" | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "hash_examples": "1300bfd0dfc59114", | |
| "hash_full_prompts": "db2aefbff5eec996", | |
| "hash_input_tokens": "a4e52c47400b8bca", | |
| "hash_cont_tokens": "8ea8c5ff76a15bca" | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "hash_examples": "083b1e4904c48dc2", | |
| "hash_full_prompts": "0f89ee3fe03d6a21", | |
| "hash_input_tokens": "69644001a800b0f7", | |
| "hash_cont_tokens": "e3a8cd951b6e3469" | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "hash_examples": "709128f9926a634c", | |
| "hash_full_prompts": "98a04b1f8f841069", | |
| "hash_input_tokens": "332ca144a888ad7f", | |
| "hash_cont_tokens": "3e9e0bdc248fd88a" | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "hash_examples": "88f22a636029ae47", | |
| "hash_full_prompts": "2e1c8d4b1e0cc921", | |
| "hash_input_tokens": "a27f6dd3c2837ded", | |
| "hash_cont_tokens": "55b12fb138c6a064" | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "hash_examples": "8c8a1e07a2151dca", | |
| "hash_full_prompts": "f51611f514b265b0", | |
| "hash_input_tokens": "9f72696f5f9c4c80", | |
| "hash_cont_tokens": "a01d6d39a83c4597" | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "hash_examples": "2668953431f91e96", | |
| "hash_full_prompts": "77562bef997c7650", | |
| "hash_input_tokens": "0d9707022133f086", | |
| "hash_cont_tokens": "6aeaed4d823c98aa" | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "hash_examples": "9c2dda34a2ea4fd2", | |
| "hash_full_prompts": "202139046daa118f", | |
| "hash_input_tokens": "e957962a583e58a2", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "hash_examples": "41adb694024809c2", | |
| "hash_full_prompts": "bffec9fc237bcf93", | |
| "hash_input_tokens": "46fe4585062aa36a", | |
| "hash_cont_tokens": "9b0ab02a64603081" | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "hash_examples": "3171c13ba3c594c4", | |
| "hash_full_prompts": "170831fc36f1d59e", | |
| "hash_input_tokens": "cf9834b2c07721dc", | |
| "hash_cont_tokens": "3b8bbe9108e55ce9" | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "hash_examples": "9873e077e83e0546", | |
| "hash_full_prompts": "08f4ceba3131a068", | |
| "hash_input_tokens": "f257b7cce9ddb541", | |
| "hash_cont_tokens": "3e9bfc0362e97330" | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "hash_examples": "7db1d8142ec14323", | |
| "hash_full_prompts": "4c0e68e3586cb453", | |
| "hash_input_tokens": "8650a7e901b42458", | |
| "hash_cont_tokens": "23b2dc6ee2da4cfc" | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "hash_examples": "9b455b7d72811cc8", | |
| "hash_full_prompts": "e467f822d8a0d3ff", | |
| "hash_input_tokens": "4ba4c1d13e1040ec", | |
| "hash_cont_tokens": "9f6ff69d23a48783" | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "hash_examples": "8be90d0f538f1560", | |
| "hash_full_prompts": "152187949bcd0921", | |
| "hash_input_tokens": "7431d7b2d5c13409", | |
| "hash_cont_tokens": "d6458d743d875837" | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "hash_examples": "8d377597916cd07e", | |
| "hash_full_prompts": "0eb7345d6144ee0d", | |
| "hash_input_tokens": "e7bbb4a15e991424", | |
| "hash_cont_tokens": "922a195f53a35662" | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "hash_examples": "cd9dbc52b3c932d6", | |
| "hash_full_prompts": "36ac764272bfb182", | |
| "hash_input_tokens": "9178e10bd0763ec4", | |
| "hash_cont_tokens": "2e590029ef41fbcd" | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "hash_examples": "b20e4e816c1e383e", | |
| "hash_full_prompts": "7b8d69ea2acaf2f7", | |
| "hash_input_tokens": "f5a22012a54f70ea", | |
| "hash_cont_tokens": "7cfee54dbddd5a98" | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "hash_examples": "d45b73b22f9cc039", | |
| "hash_full_prompts": "fe8937e9ffc99771", | |
| "hash_input_tokens": "8eeb91b3a7cbea0a", | |
| "hash_cont_tokens": "a86677b2a45c20e1" | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "hash_examples": "0d25072e1761652a", | |
| "hash_full_prompts": "f9adc39cfa9f42ba", | |
| "hash_input_tokens": "bdfc559a40a1e8ec", | |
| "hash_cont_tokens": "0d756ccaae031757" | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "hash_examples": "62bb8197e63d60d4", | |
| "hash_full_prompts": "869c9c3ae196b7c3", | |
| "hash_input_tokens": "d49711415961ced7", | |
| "hash_cont_tokens": "b2229bc2cfbf594b" | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "hash_examples": "e7959df87dea8672", | |
| "hash_full_prompts": "1a1fc00e17b3a52a", | |
| "hash_input_tokens": "f9a00c6fc5e9cea7", | |
| "hash_cont_tokens": "c3a3bdfd177eed5b" | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "hash_examples": "4a56a01ddca44dca", | |
| "hash_full_prompts": "0c7a7081c71c07b6", | |
| "hash_input_tokens": "647f2d7d9075afaa", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "hash_examples": "451cc86a8c4f4fe9", | |
| "hash_full_prompts": "01e95325d8b738e4", | |
| "hash_input_tokens": "784f75f0ad6e0698", | |
| "hash_cont_tokens": "af8b3658088cb37f" | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "hash_examples": "3b29cfaf1a81c379", | |
| "hash_full_prompts": "e0d79a15083dfdff", | |
| "hash_input_tokens": "17766ebe38853371", | |
| "hash_cont_tokens": "060118bef6de4e0a" | |
| } | |
| } | |
| } |