Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
| { | |
| "results": { | |
| "harness|arc:challenge|25": { | |
| "acc": 0.5563139931740614, | |
| "acc_stderr": 0.014518421825670449, | |
| "acc_norm": 0.590443686006826, | |
| "acc_norm_stderr": 0.01437035863247244 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6293567018522207, | |
| "acc_stderr": 0.004819899945342489, | |
| "acc_norm": 0.8193586934873531, | |
| "acc_norm_stderr": 0.0038393444971919545 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.31, | |
| "acc_stderr": 0.046482319871173156, | |
| "acc_norm": 0.31, | |
| "acc_norm_stderr": 0.046482319871173156 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.4740740740740741, | |
| "acc_stderr": 0.04313531696750574, | |
| "acc_norm": 0.4740740740740741, | |
| "acc_norm_stderr": 0.04313531696750574 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.5460526315789473, | |
| "acc_stderr": 0.04051646342874142, | |
| "acc_norm": 0.5460526315789473, | |
| "acc_norm_stderr": 0.04051646342874142 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.53, | |
| "acc_stderr": 0.05016135580465919, | |
| "acc_norm": 0.53, | |
| "acc_norm_stderr": 0.05016135580465919 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.5849056603773585, | |
| "acc_stderr": 0.03032594578928611, | |
| "acc_norm": 0.5849056603773585, | |
| "acc_norm_stderr": 0.03032594578928611 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.5833333333333334, | |
| "acc_stderr": 0.04122728707651282, | |
| "acc_norm": 0.5833333333333334, | |
| "acc_norm_stderr": 0.04122728707651282 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.38, | |
| "acc_stderr": 0.048783173121456316, | |
| "acc_norm": 0.38, | |
| "acc_norm_stderr": 0.048783173121456316 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.47, | |
| "acc_stderr": 0.05016135580465919, | |
| "acc_norm": 0.47, | |
| "acc_norm_stderr": 0.05016135580465919 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "acc": 0.32, | |
| "acc_stderr": 0.046882617226215034, | |
| "acc_norm": 0.32, | |
| "acc_norm_stderr": 0.046882617226215034 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "acc": 0.4624277456647399, | |
| "acc_stderr": 0.0380168510452446, | |
| "acc_norm": 0.4624277456647399, | |
| "acc_norm_stderr": 0.0380168510452446 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.3137254901960784, | |
| "acc_stderr": 0.04617034827006717, | |
| "acc_norm": 0.3137254901960784, | |
| "acc_norm_stderr": 0.04617034827006717 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.68, | |
| "acc_stderr": 0.04688261722621505, | |
| "acc_norm": 0.68, | |
| "acc_norm_stderr": 0.04688261722621505 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.4, | |
| "acc_stderr": 0.03202563076101735, | |
| "acc_norm": 0.4, | |
| "acc_norm_stderr": 0.03202563076101735 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.3157894736842105, | |
| "acc_stderr": 0.043727482902780064, | |
| "acc_norm": 0.3157894736842105, | |
| "acc_norm_stderr": 0.043727482902780064 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.503448275862069, | |
| "acc_stderr": 0.0416656757710158, | |
| "acc_norm": 0.503448275862069, | |
| "acc_norm_stderr": 0.0416656757710158 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.3412698412698413, | |
| "acc_stderr": 0.024419234966819064, | |
| "acc_norm": 0.3412698412698413, | |
| "acc_norm_stderr": 0.024419234966819064 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.30952380952380953, | |
| "acc_stderr": 0.04134913018303316, | |
| "acc_norm": 0.30952380952380953, | |
| "acc_norm_stderr": 0.04134913018303316 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.3, | |
| "acc_stderr": 0.046056618647183814, | |
| "acc_norm": 0.3, | |
| "acc_norm_stderr": 0.046056618647183814 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.6419354838709678, | |
| "acc_stderr": 0.02727389059430064, | |
| "acc_norm": 0.6419354838709678, | |
| "acc_norm_stderr": 0.02727389059430064 | |
| }, | |
| "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": 0.59, | |
| "acc_stderr": 0.04943110704237102, | |
| "acc_norm": 0.59, | |
| "acc_norm_stderr": 0.04943110704237102 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.6727272727272727, | |
| "acc_stderr": 0.03663974994391244, | |
| "acc_norm": 0.6727272727272727, | |
| "acc_norm_stderr": 0.03663974994391244 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.702020202020202, | |
| "acc_stderr": 0.03258630383836556, | |
| "acc_norm": 0.702020202020202, | |
| "acc_norm_stderr": 0.03258630383836556 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.7875647668393783, | |
| "acc_stderr": 0.029519282616817234, | |
| "acc_norm": 0.7875647668393783, | |
| "acc_norm_stderr": 0.029519282616817234 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.49230769230769234, | |
| "acc_stderr": 0.025348006031534788, | |
| "acc_norm": 0.49230769230769234, | |
| "acc_norm_stderr": 0.025348006031534788 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.3111111111111111, | |
| "acc_stderr": 0.028226446749683522, | |
| "acc_norm": 0.3111111111111111, | |
| "acc_norm_stderr": 0.028226446749683522 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.5294117647058824, | |
| "acc_stderr": 0.03242225027115007, | |
| "acc_norm": 0.5294117647058824, | |
| "acc_norm_stderr": 0.03242225027115007 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.33774834437086093, | |
| "acc_stderr": 0.038615575462551684, | |
| "acc_norm": 0.33774834437086093, | |
| "acc_norm_stderr": 0.038615575462551684 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.7321100917431193, | |
| "acc_stderr": 0.018987462257978652, | |
| "acc_norm": 0.7321100917431193, | |
| "acc_norm_stderr": 0.018987462257978652 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.3888888888888889, | |
| "acc_stderr": 0.03324708911809117, | |
| "acc_norm": 0.3888888888888889, | |
| "acc_norm_stderr": 0.03324708911809117 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.75, | |
| "acc_stderr": 0.03039153369274154, | |
| "acc_norm": 0.75, | |
| "acc_norm_stderr": 0.03039153369274154 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.7172995780590717, | |
| "acc_stderr": 0.02931281415395592, | |
| "acc_norm": 0.7172995780590717, | |
| "acc_norm_stderr": 0.02931281415395592 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.6457399103139013, | |
| "acc_stderr": 0.032100621541349864, | |
| "acc_norm": 0.6457399103139013, | |
| "acc_norm_stderr": 0.032100621541349864 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.6335877862595419, | |
| "acc_stderr": 0.04225875451969637, | |
| "acc_norm": 0.6335877862595419, | |
| "acc_norm_stderr": 0.04225875451969637 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.768595041322314, | |
| "acc_stderr": 0.03849856098794089, | |
| "acc_norm": 0.768595041322314, | |
| "acc_norm_stderr": 0.03849856098794089 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.6944444444444444, | |
| "acc_stderr": 0.044531975073749834, | |
| "acc_norm": 0.6944444444444444, | |
| "acc_norm_stderr": 0.044531975073749834 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.6503067484662577, | |
| "acc_stderr": 0.037466683254700206, | |
| "acc_norm": 0.6503067484662577, | |
| "acc_norm_stderr": 0.037466683254700206 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.35714285714285715, | |
| "acc_stderr": 0.04547960999764376, | |
| "acc_norm": 0.35714285714285715, | |
| "acc_norm_stderr": 0.04547960999764376 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.7378640776699029, | |
| "acc_stderr": 0.04354631077260595, | |
| "acc_norm": 0.7378640776699029, | |
| "acc_norm_stderr": 0.04354631077260595 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.7863247863247863, | |
| "acc_stderr": 0.026853450377009175, | |
| "acc_norm": 0.7863247863247863, | |
| "acc_norm_stderr": 0.026853450377009175 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.57, | |
| "acc_stderr": 0.049756985195624284, | |
| "acc_norm": 0.57, | |
| "acc_norm_stderr": 0.049756985195624284 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.7471264367816092, | |
| "acc_stderr": 0.015543377313719681, | |
| "acc_norm": 0.7471264367816092, | |
| "acc_norm_stderr": 0.015543377313719681 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.6127167630057804, | |
| "acc_stderr": 0.026226158605124655, | |
| "acc_norm": 0.6127167630057804, | |
| "acc_norm_stderr": 0.026226158605124655 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.30502793296089387, | |
| "acc_stderr": 0.015398723510916716, | |
| "acc_norm": 0.30502793296089387, | |
| "acc_norm_stderr": 0.015398723510916716 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.5947712418300654, | |
| "acc_stderr": 0.028110928492809068, | |
| "acc_norm": 0.5947712418300654, | |
| "acc_norm_stderr": 0.028110928492809068 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.5884244372990354, | |
| "acc_stderr": 0.02795048149440127, | |
| "acc_norm": 0.5884244372990354, | |
| "acc_norm_stderr": 0.02795048149440127 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.6111111111111112, | |
| "acc_stderr": 0.02712511551316687, | |
| "acc_norm": 0.6111111111111112, | |
| "acc_norm_stderr": 0.02712511551316687 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.38652482269503546, | |
| "acc_stderr": 0.029049190342543454, | |
| "acc_norm": 0.38652482269503546, | |
| "acc_norm_stderr": 0.029049190342543454 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.39113428943937417, | |
| "acc_stderr": 0.012463861839982058, | |
| "acc_norm": 0.39113428943937417, | |
| "acc_norm_stderr": 0.012463861839982058 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.5, | |
| "acc_stderr": 0.030372836961539352, | |
| "acc_norm": 0.5, | |
| "acc_norm_stderr": 0.030372836961539352 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.5424836601307189, | |
| "acc_stderr": 0.020154685712590888, | |
| "acc_norm": 0.5424836601307189, | |
| "acc_norm_stderr": 0.020154685712590888 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "acc": 0.6636363636363637, | |
| "acc_stderr": 0.04525393596302505, | |
| "acc_norm": 0.6636363636363637, | |
| "acc_norm_stderr": 0.04525393596302505 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "acc": 0.6408163265306123, | |
| "acc_stderr": 0.030713560455108493, | |
| "acc_norm": 0.6408163265306123, | |
| "acc_norm_stderr": 0.030713560455108493 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.7512437810945274, | |
| "acc_stderr": 0.030567675938916714, | |
| "acc_norm": 0.7512437810945274, | |
| "acc_norm_stderr": 0.030567675938916714 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.81, | |
| "acc_stderr": 0.03942772444036625, | |
| "acc_norm": 0.81, | |
| "acc_norm_stderr": 0.03942772444036625 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.4819277108433735, | |
| "acc_stderr": 0.038899512528272166, | |
| "acc_norm": 0.4819277108433735, | |
| "acc_norm_stderr": 0.038899512528272166 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.7309941520467836, | |
| "acc_stderr": 0.03401052620104089, | |
| "acc_norm": 0.7309941520467836, | |
| "acc_norm_stderr": 0.03401052620104089 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.28518971848225216, | |
| "mc1_stderr": 0.015805827874454895, | |
| "mc2": 0.4411794590119937, | |
| "mc2_stderr": 0.015755921757439843 | |
| }, | |
| "all": { | |
| "acc": 0.5479380524707899, | |
| "acc_stderr": 0.03451142729909022, | |
| "acc_norm": 0.5517368945804153, | |
| "acc_norm_stderr": 0.03449229816957583, | |
| "mc1": 0.28518971848225216, | |
| "mc1_stderr": 0.015805827874454895, | |
| "mc2": 0.4411794590119937, | |
| "mc2_stderr": 0.015755921757439843 | |
| } | |
| }, | |
| "versions": { | |
| "harness|arc:challenge|25": 0, | |
| "harness|hellaswag|10": 0, | |
| "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, | |
| "harness|truthfulqa:mc|0": 1, | |
| "all": 0 | |
| }, | |
| "config": { | |
| "model_name": "meta-llama/Llama-2-13b-chat-hf", | |
| "model_sha": "f848cf15ab9a51ae5735ab28120a9a0773eeb541", | |
| "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|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", | |
| "harness|truthfulqa:mc": "LM Harness task" | |
| }, | |
| "hashes": { | |
| "harness|arc:challenge|25": { | |
| "hash_examples": "fb8c51b1872daeda", | |
| "hash_full_prompts": "045cbb916e5145c6", | |
| "hash_input_tokens": "61571bf68d6d89aa", | |
| "hash_cont_tokens": "8210decc6ff6f7df" | |
| }, | |
| "harness|hellaswag|10": { | |
| "hash_examples": "e1768ecb99d7ecf0", | |
| "hash_full_prompts": "0b4c16983130f84f", | |
| "hash_input_tokens": "29906669b1c7054a", | |
| "hash_cont_tokens": "b3b9e9017afa63af" | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "hash_examples": "280f9f325b40559a", | |
| "hash_full_prompts": "2f776a367d23aea2", | |
| "hash_input_tokens": "c54ff61ad0273dd7", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "hash_examples": "2f83a4f1cab4ba18", | |
| "hash_full_prompts": "516f74bef25df620", | |
| "hash_input_tokens": "be31a1e22aef5f90", | |
| "hash_cont_tokens": "f11971a765cb609f" | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "hash_examples": "7d587b908da4d762", | |
| "hash_full_prompts": "faf4e80f65de93ca", | |
| "hash_input_tokens": "277a7b1fad566940", | |
| "hash_cont_tokens": "bf30e5d3f48250cb" | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "hash_examples": "33e51740670de686", | |
| "hash_full_prompts": "db01c3ef8e1479d4", | |
| "hash_input_tokens": "ba552605bc116de5", | |
| "hash_cont_tokens": "bc1dd9b2d995eb61" | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "hash_examples": "f3366dbe7eefffa4", | |
| "hash_full_prompts": "49654f71d94b65c3", | |
| "hash_input_tokens": "428c7563d0b98ab9", | |
| "hash_cont_tokens": "890a119624b3b935" | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "hash_examples": "ca2b6753a0193e7f", | |
| "hash_full_prompts": "2b460b75f1fdfefd", | |
| "hash_input_tokens": "da036601573942e2", | |
| "hash_cont_tokens": "875cde3af7a0ee14" | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "hash_examples": "22ff85f1d34f42d1", | |
| "hash_full_prompts": "242c9be6da583e95", | |
| "hash_input_tokens": "94e0196d6aded13d", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "hash_examples": "30318289d717a5cf", | |
| "hash_full_prompts": "ed2bdb4e87c4b371", | |
| "hash_input_tokens": "6e4d0f4a8d36690b", | |
| "hash_cont_tokens": "ffc0fe414cdc4a83" | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "hash_examples": "4944d1f0b6b5d911", | |
| "hash_full_prompts": "770bc4281c973190", | |
| "hash_input_tokens": "614054d17109a25d", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "hash_examples": "dd69cc33381275af", | |
| "hash_full_prompts": "ad2a53e5250ab46e", | |
| "hash_input_tokens": "1d633b3cc0524ba8", | |
| "hash_cont_tokens": "1f88b00d41957d82" | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "hash_examples": "875dd26d22655b0d", | |
| "hash_full_prompts": "833a0d7b55aed500", | |
| "hash_input_tokens": "5421d9a1af86cbd4", | |
| "hash_cont_tokens": "f7b8097afc16a47c" | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "hash_examples": "006451eedc0ededb", | |
| "hash_full_prompts": "94034c97e85d8f46", | |
| "hash_input_tokens": "5e6b70ecb333cf18", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "hash_examples": "8874ece872d2ca4c", | |
| "hash_full_prompts": "e40d15a34640d6fa", | |
| "hash_input_tokens": "c2ef11a87264ceed", | |
| "hash_cont_tokens": "aa0e8bc655f2f641" | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "hash_examples": "64d3623b0bfaa43f", | |
| "hash_full_prompts": "612f340fae41338d", | |
| "hash_input_tokens": "ecaccd912a4c3978", | |
| "hash_cont_tokens": "bfb7e3c3c88313f1" | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "hash_examples": "e98f51780c674d7e", | |
| "hash_full_prompts": "10275b312d812ae6", | |
| "hash_input_tokens": "1590c84291399be8", | |
| "hash_cont_tokens": "2425a3f084a591ef" | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "hash_examples": "fc48208a5ac1c0ce", | |
| "hash_full_prompts": "5ec274c6c82aca23", | |
| "hash_input_tokens": "3269597f715b0da1", | |
| "hash_cont_tokens": "f52691aef15a407b" | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "hash_examples": "5a6525665f63ea72", | |
| "hash_full_prompts": "07b92638c4a6b500", | |
| "hash_input_tokens": "a2800d20f3ab8d7c", | |
| "hash_cont_tokens": "f515d598d9c21263" | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "hash_examples": "371d70d743b2b89b", | |
| "hash_full_prompts": "332fdee50a1921b4", | |
| "hash_input_tokens": "94ed44b3772505ad", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "hash_examples": "a79e1018b1674052", | |
| "hash_full_prompts": "e624e26ede922561", | |
| "hash_input_tokens": "24423acb928db768", | |
| "hash_cont_tokens": "bd85a4156a3613ee" | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "hash_examples": "44bfc25c389f0e03", | |
| "hash_full_prompts": "0e3e5f5d9246482a", | |
| "hash_input_tokens": "831ff35c474e5cef", | |
| "hash_cont_tokens": "a95c97af1c14e068" | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "hash_examples": "8b8cdb1084f24169", | |
| "hash_full_prompts": "c00487e67c1813cc", | |
| "hash_input_tokens": "8c34e0f2bda77358", | |
| "hash_cont_tokens": "8abfedef914e33c9" | |
| }, | |
| "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": "7c5547c7da5bc793", | |
| "hash_cont_tokens": "03a5012b916274ea" | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "hash_examples": "d221ec983d143dc3", | |
| "hash_full_prompts": "ac42d888e1ce1155", | |
| "hash_input_tokens": "f62991cb6a496b05", | |
| "hash_cont_tokens": "a83effb8f76b7d7c" | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "hash_examples": "59c2915cacfd3fbb", | |
| "hash_full_prompts": "c6bd9d25158abd0e", | |
| "hash_input_tokens": "4cef2aff6e3d59ed", | |
| "hash_cont_tokens": "c583432ad27fcfe0" | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "hash_examples": "1f8ac897608de342", | |
| "hash_full_prompts": "5d88f41fc2d643a8", | |
| "hash_input_tokens": "6e2577ea4082ed2b", | |
| "hash_cont_tokens": "24f5dc613660300b" | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "hash_examples": "ead6a0f2f6c83370", | |
| "hash_full_prompts": "bfc393381298609e", | |
| "hash_input_tokens": "c5fc9aeb1079c8e4", | |
| "hash_cont_tokens": "f47f041de50333b9" | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "hash_examples": "c3f2025990afec64", | |
| "hash_full_prompts": "fc78b4997e436734", | |
| "hash_input_tokens": "555fc385cffa84ca", | |
| "hash_cont_tokens": "ba2efcd283e938cc" | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "hash_examples": "21f8aab618f6d636", | |
| "hash_full_prompts": "d5c76aa40b9dbc43", | |
| "hash_input_tokens": "febd23cbf9973b7f", | |
| "hash_cont_tokens": "942069cd363844d9" | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "hash_examples": "2386a60a11fc5de3", | |
| "hash_full_prompts": "4c5c8be5aafac432", | |
| "hash_input_tokens": "424b02981230ee83", | |
| "hash_cont_tokens": "955ed42b6f7fa019" | |
| }, | |
| "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": "9a864184946033ac" | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "hash_examples": "1a7199dc733e779b", | |
| "hash_full_prompts": "756b9096b8eaf892", | |
| "hash_input_tokens": "541a75f071dcf579", | |
| "hash_cont_tokens": "142a4a8a1138a214" | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "hash_examples": "7acb8fdad97f88a6", | |
| "hash_full_prompts": "731a52ff15b8cfdb", | |
| "hash_input_tokens": "04269e5c5a257dd9", | |
| "hash_cont_tokens": "bc54813e809b796d" | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "hash_examples": "1300bfd0dfc59114", | |
| "hash_full_prompts": "db2aefbff5eec996", | |
| "hash_input_tokens": "d93ba9d9d38e4397", | |
| "hash_cont_tokens": "dc45b45fcda18e5d" | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "hash_examples": "083b1e4904c48dc2", | |
| "hash_full_prompts": "0f89ee3fe03d6a21", | |
| "hash_input_tokens": "9eeaccd2698b4f5a", | |
| "hash_cont_tokens": "e3a8cd951b6e3469" | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "hash_examples": "709128f9926a634c", | |
| "hash_full_prompts": "98a04b1f8f841069", | |
| "hash_input_tokens": "b4f08f544f2b7576", | |
| "hash_cont_tokens": "1e80dbd30f6453d5" | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "hash_examples": "88f22a636029ae47", | |
| "hash_full_prompts": "2e1c8d4b1e0cc921", | |
| "hash_input_tokens": "900c2a51f1174b9f", | |
| "hash_cont_tokens": "9b37da7777378ca9" | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "hash_examples": "8c8a1e07a2151dca", | |
| "hash_full_prompts": "f51611f514b265b0", | |
| "hash_input_tokens": "6b36efb4689c6eca", | |
| "hash_cont_tokens": "a01d6d39a83c4597" | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "hash_examples": "2668953431f91e96", | |
| "hash_full_prompts": "77562bef997c7650", | |
| "hash_input_tokens": "2aaac78a0cfed47a", | |
| "hash_cont_tokens": "6aeaed4d823c98aa" | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "hash_examples": "9c2dda34a2ea4fd2", | |
| "hash_full_prompts": "202139046daa118f", | |
| "hash_input_tokens": "886ca823b41c094a", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "hash_examples": "41adb694024809c2", | |
| "hash_full_prompts": "bffec9fc237bcf93", | |
| "hash_input_tokens": "72fd71de7675e7d0", | |
| "hash_cont_tokens": "9b0ab02a64603081" | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "hash_examples": "3171c13ba3c594c4", | |
| "hash_full_prompts": "170831fc36f1d59e", | |
| "hash_input_tokens": "f3ca0dd8e7a1eb09", | |
| "hash_cont_tokens": "8badf768f7b0467a" | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "hash_examples": "9873e077e83e0546", | |
| "hash_full_prompts": "08f4ceba3131a068", | |
| "hash_input_tokens": "3e793631e951f23c", | |
| "hash_cont_tokens": "32ae620376b2bbba" | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "hash_examples": "7db1d8142ec14323", | |
| "hash_full_prompts": "4c0e68e3586cb453", | |
| "hash_input_tokens": "59753c2144ea93af", | |
| "hash_cont_tokens": "3071def75bacc404" | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "hash_examples": "9b455b7d72811cc8", | |
| "hash_full_prompts": "e467f822d8a0d3ff", | |
| "hash_input_tokens": "bd8d3dbed15a8c34", | |
| "hash_cont_tokens": "9f6ff69d23a48783" | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "hash_examples": "8be90d0f538f1560", | |
| "hash_full_prompts": "152187949bcd0921", | |
| "hash_input_tokens": "3573cd87facbb7c5", | |
| "hash_cont_tokens": "de469d2b981e32a3" | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "hash_examples": "8d377597916cd07e", | |
| "hash_full_prompts": "0eb7345d6144ee0d", | |
| "hash_input_tokens": "17e721bc1a7cbb47", | |
| "hash_cont_tokens": "c46f74d2dfc7b13b" | |
| }, | |
| "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": "fe35cfa9c6ca802e" | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "hash_examples": "d45b73b22f9cc039", | |
| "hash_full_prompts": "fe8937e9ffc99771", | |
| "hash_input_tokens": "0dfb73a8eb3f692c", | |
| "hash_cont_tokens": "f020fbddf72c8652" | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "hash_examples": "0d25072e1761652a", | |
| "hash_full_prompts": "f9adc39cfa9f42ba", | |
| "hash_input_tokens": "1710c6ba4c9f3cbd", | |
| "hash_cont_tokens": "568f585a259965c1" | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "hash_examples": "62bb8197e63d60d4", | |
| "hash_full_prompts": "869c9c3ae196b7c3", | |
| "hash_input_tokens": "d49711415961ced7", | |
| "hash_cont_tokens": "cc6fd7cccd64cd5d" | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "hash_examples": "e7959df87dea8672", | |
| "hash_full_prompts": "1a1fc00e17b3a52a", | |
| "hash_input_tokens": "828999f7624cbe7e", | |
| "hash_cont_tokens": "c3a3bdfd177eed5b" | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "hash_examples": "4a56a01ddca44dca", | |
| "hash_full_prompts": "0c7a7081c71c07b6", | |
| "hash_input_tokens": "42054621e718dbee", | |
| "hash_cont_tokens": "2568d0e8e36fa959" | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "hash_examples": "451cc86a8c4f4fe9", | |
| "hash_full_prompts": "01e95325d8b738e4", | |
| "hash_input_tokens": "6c4f0aa4dc859c04", | |
| "hash_cont_tokens": "926cf60b0891f374" | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "hash_examples": "3b29cfaf1a81c379", | |
| "hash_full_prompts": "e0d79a15083dfdff", | |
| "hash_input_tokens": "6c75d44e092ff24f", | |
| "hash_cont_tokens": "c525a5de974c1ea3" | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "hash_examples": "23176c0531c7b867", | |
| "hash_full_prompts": "36a6d90e75d92d4a", | |
| "hash_input_tokens": "2738d7ed7075faa7", | |
| "hash_cont_tokens": "c014154380b74b9e" | |
| } | |
| } | |
| } |