Spaces:
Runtime error
Runtime error
| { | |
| "results": { | |
| "harness|arc:challenge|25": { | |
| "acc": 0.5170648464163823, | |
| "acc_stderr": 0.0146028783885366, | |
| "acc_norm": 0.5281569965870307, | |
| "acc_norm_stderr": 0.014588204105102203 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6018721370244972, | |
| "acc_stderr": 0.004885116465550283, | |
| "acc_norm": 0.795857398924517, | |
| "acc_norm_stderr": 0.004022499210760732 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.32, | |
| "acc_stderr": 0.046882617226215034, | |
| "acc_norm": 0.32, | |
| "acc_norm_stderr": 0.046882617226215034 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.5259259259259259, | |
| "acc_stderr": 0.04313531696750575, | |
| "acc_norm": 0.5259259259259259, | |
| "acc_norm_stderr": 0.04313531696750575 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.5131578947368421, | |
| "acc_stderr": 0.04067533136309173, | |
| "acc_norm": 0.5131578947368421, | |
| "acc_norm_stderr": 0.04067533136309173 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.41, | |
| "acc_stderr": 0.049431107042371025, | |
| "acc_norm": 0.41, | |
| "acc_norm_stderr": 0.049431107042371025 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.5320754716981132, | |
| "acc_stderr": 0.03070948699255654, | |
| "acc_norm": 0.5320754716981132, | |
| "acc_norm_stderr": 0.03070948699255654 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.4930555555555556, | |
| "acc_stderr": 0.04180806750294938, | |
| "acc_norm": 0.4930555555555556, | |
| "acc_norm_stderr": 0.04180806750294938 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.29, | |
| "acc_stderr": 0.04560480215720684, | |
| "acc_norm": 0.29, | |
| "acc_norm_stderr": 0.04560480215720684 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.45, | |
| "acc_stderr": 0.05, | |
| "acc_norm": 0.45, | |
| "acc_norm_stderr": 0.05 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "acc": 0.31, | |
| "acc_stderr": 0.04648231987117316, | |
| "acc_norm": 0.31, | |
| "acc_norm_stderr": 0.04648231987117316 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "acc": 0.4277456647398844, | |
| "acc_stderr": 0.03772446857518026, | |
| "acc_norm": 0.4277456647398844, | |
| "acc_norm_stderr": 0.03772446857518026 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.17647058823529413, | |
| "acc_stderr": 0.0379328118530781, | |
| "acc_norm": 0.17647058823529413, | |
| "acc_norm_stderr": 0.0379328118530781 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.61, | |
| "acc_stderr": 0.04902071300001974, | |
| "acc_norm": 0.61, | |
| "acc_norm_stderr": 0.04902071300001974 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.4340425531914894, | |
| "acc_stderr": 0.03240038086792747, | |
| "acc_norm": 0.4340425531914894, | |
| "acc_norm_stderr": 0.03240038086792747 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.30701754385964913, | |
| "acc_stderr": 0.04339138322579861, | |
| "acc_norm": 0.30701754385964913, | |
| "acc_norm_stderr": 0.04339138322579861 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.4206896551724138, | |
| "acc_stderr": 0.0411391498118926, | |
| "acc_norm": 0.4206896551724138, | |
| "acc_norm_stderr": 0.0411391498118926 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.2830687830687831, | |
| "acc_stderr": 0.023201392938194978, | |
| "acc_norm": 0.2830687830687831, | |
| "acc_norm_stderr": 0.023201392938194978 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.24603174603174602, | |
| "acc_stderr": 0.038522733649243135, | |
| "acc_norm": 0.24603174603174602, | |
| "acc_norm_stderr": 0.038522733649243135 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.32, | |
| "acc_stderr": 0.04688261722621503, | |
| "acc_norm": 0.32, | |
| "acc_norm_stderr": 0.04688261722621503 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.5129032258064516, | |
| "acc_stderr": 0.028434533152681855, | |
| "acc_norm": 0.5129032258064516, | |
| "acc_norm_stderr": 0.028434533152681855 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "acc": 0.33497536945812806, | |
| "acc_stderr": 0.033208527423483104, | |
| "acc_norm": 0.33497536945812806, | |
| "acc_norm_stderr": 0.033208527423483104 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "acc": 0.52, | |
| "acc_stderr": 0.050211673156867795, | |
| "acc_norm": 0.52, | |
| "acc_norm_stderr": 0.050211673156867795 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.5454545454545454, | |
| "acc_stderr": 0.038881769216741004, | |
| "acc_norm": 0.5454545454545454, | |
| "acc_norm_stderr": 0.038881769216741004 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.6616161616161617, | |
| "acc_stderr": 0.03371124142626303, | |
| "acc_norm": 0.6616161616161617, | |
| "acc_norm_stderr": 0.03371124142626303 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.6683937823834197, | |
| "acc_stderr": 0.03397636541089118, | |
| "acc_norm": 0.6683937823834197, | |
| "acc_norm_stderr": 0.03397636541089118 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.4717948717948718, | |
| "acc_stderr": 0.0253106392549339, | |
| "acc_norm": 0.4717948717948718, | |
| "acc_norm_stderr": 0.0253106392549339 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.25555555555555554, | |
| "acc_stderr": 0.026593939101844058, | |
| "acc_norm": 0.25555555555555554, | |
| "acc_norm_stderr": 0.026593939101844058 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.4831932773109244, | |
| "acc_stderr": 0.03246013680375308, | |
| "acc_norm": 0.4831932773109244, | |
| "acc_norm_stderr": 0.03246013680375308 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.2582781456953642, | |
| "acc_stderr": 0.035737053147634576, | |
| "acc_norm": 0.2582781456953642, | |
| "acc_norm_stderr": 0.035737053147634576 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.6440366972477064, | |
| "acc_stderr": 0.020528559278244214, | |
| "acc_norm": 0.6440366972477064, | |
| "acc_norm_stderr": 0.020528559278244214 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.35648148148148145, | |
| "acc_stderr": 0.03266478331527272, | |
| "acc_norm": 0.35648148148148145, | |
| "acc_norm_stderr": 0.03266478331527272 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.6372549019607843, | |
| "acc_stderr": 0.03374499356319355, | |
| "acc_norm": 0.6372549019607843, | |
| "acc_norm_stderr": 0.03374499356319355 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.6624472573839663, | |
| "acc_stderr": 0.03078154910202622, | |
| "acc_norm": 0.6624472573839663, | |
| "acc_norm_stderr": 0.03078154910202622 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.5381165919282511, | |
| "acc_stderr": 0.033460150119732274, | |
| "acc_norm": 0.5381165919282511, | |
| "acc_norm_stderr": 0.033460150119732274 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.5343511450381679, | |
| "acc_stderr": 0.04374928560599738, | |
| "acc_norm": 0.5343511450381679, | |
| "acc_norm_stderr": 0.04374928560599738 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.6776859504132231, | |
| "acc_stderr": 0.042664163633521685, | |
| "acc_norm": 0.6776859504132231, | |
| "acc_norm_stderr": 0.042664163633521685 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.5462962962962963, | |
| "acc_stderr": 0.04812917324536823, | |
| "acc_norm": 0.5462962962962963, | |
| "acc_norm_stderr": 0.04812917324536823 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.5521472392638037, | |
| "acc_stderr": 0.03906947479456606, | |
| "acc_norm": 0.5521472392638037, | |
| "acc_norm_stderr": 0.03906947479456606 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.45535714285714285, | |
| "acc_stderr": 0.047268355537191, | |
| "acc_norm": 0.45535714285714285, | |
| "acc_norm_stderr": 0.047268355537191 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.6796116504854369, | |
| "acc_stderr": 0.04620284082280042, | |
| "acc_norm": 0.6796116504854369, | |
| "acc_norm_stderr": 0.04620284082280042 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.7008547008547008, | |
| "acc_stderr": 0.02999695185834948, | |
| "acc_norm": 0.7008547008547008, | |
| "acc_norm_stderr": 0.02999695185834948 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.49, | |
| "acc_stderr": 0.05024183937956911, | |
| "acc_norm": 0.49, | |
| "acc_norm_stderr": 0.05024183937956911 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.669220945083014, | |
| "acc_stderr": 0.016824818462563746, | |
| "acc_norm": 0.669220945083014, | |
| "acc_norm_stderr": 0.016824818462563746 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.5057803468208093, | |
| "acc_stderr": 0.026917296179149116, | |
| "acc_norm": 0.5057803468208093, | |
| "acc_norm_stderr": 0.026917296179149116 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.29720670391061454, | |
| "acc_stderr": 0.015285313353641602, | |
| "acc_norm": 0.29720670391061454, | |
| "acc_norm_stderr": 0.015285313353641602 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.5032679738562091, | |
| "acc_stderr": 0.02862930519400354, | |
| "acc_norm": 0.5032679738562091, | |
| "acc_norm_stderr": 0.02862930519400354 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.5112540192926045, | |
| "acc_stderr": 0.028390897396863533, | |
| "acc_norm": 0.5112540192926045, | |
| "acc_norm_stderr": 0.028390897396863533 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.5185185185185185, | |
| "acc_stderr": 0.02780165621232366, | |
| "acc_norm": 0.5185185185185185, | |
| "acc_norm_stderr": 0.02780165621232366 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.3723404255319149, | |
| "acc_stderr": 0.028838921471251455, | |
| "acc_norm": 0.3723404255319149, | |
| "acc_norm_stderr": 0.028838921471251455 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.37809647979139505, | |
| "acc_stderr": 0.012384878406798095, | |
| "acc_norm": 0.37809647979139505, | |
| "acc_norm_stderr": 0.012384878406798095 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.44485294117647056, | |
| "acc_stderr": 0.030187532060329387, | |
| "acc_norm": 0.44485294117647056, | |
| "acc_norm_stderr": 0.030187532060329387 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.4591503267973856, | |
| "acc_stderr": 0.020160213617222516, | |
| "acc_norm": 0.4591503267973856, | |
| "acc_norm_stderr": 0.020160213617222516 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "acc": 0.6181818181818182, | |
| "acc_stderr": 0.046534298079135075, | |
| "acc_norm": 0.6181818181818182, | |
| "acc_norm_stderr": 0.046534298079135075 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "acc": 0.44081632653061226, | |
| "acc_stderr": 0.03178419114175364, | |
| "acc_norm": 0.44081632653061226, | |
| "acc_norm_stderr": 0.03178419114175364 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.6119402985074627, | |
| "acc_stderr": 0.03445789964362749, | |
| "acc_norm": 0.6119402985074627, | |
| "acc_norm_stderr": 0.03445789964362749 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.72, | |
| "acc_stderr": 0.045126085985421276, | |
| "acc_norm": 0.72, | |
| "acc_norm_stderr": 0.045126085985421276 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.4578313253012048, | |
| "acc_stderr": 0.0387862677100236, | |
| "acc_norm": 0.4578313253012048, | |
| "acc_norm_stderr": 0.0387862677100236 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.7017543859649122, | |
| "acc_stderr": 0.03508771929824565, | |
| "acc_norm": 0.7017543859649122, | |
| "acc_norm_stderr": 0.03508771929824565 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.31946144430844553, | |
| "mc1_stderr": 0.016322644182960498, | |
| "mc2": 0.48882404667849044, | |
| "mc2_stderr": 0.016077830165514555 | |
| }, | |
| "all": { | |
| "acc": 0.48449554561369323, | |
| "acc_stderr": 0.03506199979132179, | |
| "acc_norm": 0.48797143395387416, | |
| "acc_norm_stderr": 0.0350471304432163, | |
| "mc1": 0.31946144430844553, | |
| "mc1_stderr": 0.016322644182960498, | |
| "mc2": 0.48882404667849044, | |
| "mc2_stderr": 0.016077830165514555 | |
| } | |
| }, | |
| "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_general": { | |
| "model_name": "chavinlo/gpt4-x-alpaca", | |
| "model_sha": "6a571f458cab9a23d14324ec63e0abd1744c8353", | |
| "model_dtype": "torch.float16", | |
| "lighteval_sha": "efe93333f9f25e7d48cc67a6bf362e6d576f727b", | |
| "num_few_shot_default": 0, | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null | |
| }, | |
| "config_tasks": { | |
| "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" | |
| }, | |
| "summary_tasks": { | |
| "harness|arc:challenge|25": { | |
| "hashes": { | |
| "hash_examples": "17b0cae357c0259e", | |
| "hash_full_prompts": "045cbb916e5145c6", | |
| "hash_input_tokens": "2b0e07d4cdd3b0fe", | |
| "hash_cont_tokens": "ede2b335438f08e9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 4687, | |
| "padded": 4687, | |
| "non-padded": 0, | |
| "effective_few_shots": 25.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hellaswag|10": { | |
| "hashes": { | |
| "hash_examples": "e1768ecb99d7ecf0", | |
| "hash_full_prompts": "0b4c16983130f84f", | |
| "hash_input_tokens": "578edd77107cb2c3", | |
| "hash_cont_tokens": "b41cf1ad182d68d5" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 40168, | |
| "padded": 40113, | |
| "non-padded": 55, | |
| "effective_few_shots": 10.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "hashes": { | |
| "hash_examples": "280f9f325b40559a", | |
| "hash_full_prompts": "2f776a367d23aea2", | |
| "hash_input_tokens": "6a95a1511f8da075", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "hashes": { | |
| "hash_examples": "2f83a4f1cab4ba18", | |
| "hash_full_prompts": "516f74bef25df620", | |
| "hash_input_tokens": "24a78edc4d9a93aa", | |
| "hash_cont_tokens": "f11971a765cb609f" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 540, | |
| "padded": 540, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "hashes": { | |
| "hash_examples": "7d587b908da4d762", | |
| "hash_full_prompts": "faf4e80f65de93ca", | |
| "hash_input_tokens": "b11106668d6c0974", | |
| "hash_cont_tokens": "238bd86950544b29" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 608, | |
| "padded": 608, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "hashes": { | |
| "hash_examples": "33e51740670de686", | |
| "hash_full_prompts": "db01c3ef8e1479d4", | |
| "hash_input_tokens": "10180ba12a075cb0", | |
| "hash_cont_tokens": "f9d6d2a7d7e9a041" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "hashes": { | |
| "hash_examples": "f3366dbe7eefffa4", | |
| "hash_full_prompts": "49654f71d94b65c3", | |
| "hash_input_tokens": "73351ef4968750a2", | |
| "hash_cont_tokens": "6af58623d0d5fbcd" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1060, | |
| "padded": 1060, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "hashes": { | |
| "hash_examples": "ca2b6753a0193e7f", | |
| "hash_full_prompts": "2b460b75f1fdfefd", | |
| "hash_input_tokens": "a539150af234c668", | |
| "hash_cont_tokens": "875cde3af7a0ee14" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 576, | |
| "padded": 576, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "hashes": { | |
| "hash_examples": "22ff85f1d34f42d1", | |
| "hash_full_prompts": "242c9be6da583e95", | |
| "hash_input_tokens": "52e12e5a43bcee35", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "hashes": { | |
| "hash_examples": "30318289d717a5cf", | |
| "hash_full_prompts": "ed2bdb4e87c4b371", | |
| "hash_input_tokens": "d1f3721a5659f7ee", | |
| "hash_cont_tokens": "1ba0c71186b1505e" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "hashes": { | |
| "hash_examples": "4944d1f0b6b5d911", | |
| "hash_full_prompts": "770bc4281c973190", | |
| "hash_input_tokens": "f2d78f546b5595c2", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "hashes": { | |
| "hash_examples": "dd69cc33381275af", | |
| "hash_full_prompts": "ad2a53e5250ab46e", | |
| "hash_input_tokens": "c9cc19179f63d1d6", | |
| "hash_cont_tokens": "702fb6d82ff0d6ac" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 692, | |
| "padded": 692, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "hashes": { | |
| "hash_examples": "875dd26d22655b0d", | |
| "hash_full_prompts": "833a0d7b55aed500", | |
| "hash_input_tokens": "5046144e67e992e8", | |
| "hash_cont_tokens": "f7b8097afc16a47c" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 408, | |
| "padded": 408, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "hashes": { | |
| "hash_examples": "006451eedc0ededb", | |
| "hash_full_prompts": "94034c97e85d8f46", | |
| "hash_input_tokens": "4b14581ba4fc06fc", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "hashes": { | |
| "hash_examples": "8874ece872d2ca4c", | |
| "hash_full_prompts": "e40d15a34640d6fa", | |
| "hash_input_tokens": "1ee52c413b5b4cc4", | |
| "hash_cont_tokens": "aa0e8bc655f2f641" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 940, | |
| "padded": 940, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "hashes": { | |
| "hash_examples": "64d3623b0bfaa43f", | |
| "hash_full_prompts": "612f340fae41338d", | |
| "hash_input_tokens": "2914077c4dd3090a", | |
| "hash_cont_tokens": "a9b1f761089f6acc" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 456, | |
| "padded": 456, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "hashes": { | |
| "hash_examples": "e98f51780c674d7e", | |
| "hash_full_prompts": "10275b312d812ae6", | |
| "hash_input_tokens": "0f88a874342378de", | |
| "hash_cont_tokens": "2425a3f084a591ef" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 580, | |
| "padded": 580, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "hashes": { | |
| "hash_examples": "fc48208a5ac1c0ce", | |
| "hash_full_prompts": "5ec274c6c82aca23", | |
| "hash_input_tokens": "9889933f1dd02a23", | |
| "hash_cont_tokens": "eb2d5002052b5bc5" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1512, | |
| "padded": 1512, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "hashes": { | |
| "hash_examples": "5a6525665f63ea72", | |
| "hash_full_prompts": "07b92638c4a6b500", | |
| "hash_input_tokens": "dc309a94c4bfdd2f", | |
| "hash_cont_tokens": "9b30dc19c9b62f60" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 504, | |
| "padded": 504, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "hashes": { | |
| "hash_examples": "371d70d743b2b89b", | |
| "hash_full_prompts": "332fdee50a1921b4", | |
| "hash_input_tokens": "0801a0aebec3ba8c", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "hashes": { | |
| "hash_examples": "a79e1018b1674052", | |
| "hash_full_prompts": "e624e26ede922561", | |
| "hash_input_tokens": "5bc4aca8831d9c05", | |
| "hash_cont_tokens": "74217a4e2868536f" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1240, | |
| "padded": 1240, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "hashes": { | |
| "hash_examples": "44bfc25c389f0e03", | |
| "hash_full_prompts": "0e3e5f5d9246482a", | |
| "hash_input_tokens": "b92bd6b06fc3464c", | |
| "hash_cont_tokens": "bf39544be0ebf000" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 812, | |
| "padded": 812, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "hashes": { | |
| "hash_examples": "8b8cdb1084f24169", | |
| "hash_full_prompts": "c00487e67c1813cc", | |
| "hash_input_tokens": "a549346cde8165e9", | |
| "hash_cont_tokens": "43570b3948564b64" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "hashes": { | |
| "hash_examples": "11cd32d0ef440171", | |
| "hash_full_prompts": "318f4513c537c6bf", | |
| "hash_input_tokens": "f1f73dd687da18d7", | |
| "hash_cont_tokens": "674fc454bdc5ac93" | |
| }, | |
| "truncated": 660, | |
| "non-truncated": 0, | |
| "padded": 0, | |
| "non-padded": 660, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "hashes": { | |
| "hash_examples": "b60019b9e80b642f", | |
| "hash_full_prompts": "ee5789fcc1a81b1e", | |
| "hash_input_tokens": "e7e9cf91f9d6a081", | |
| "hash_cont_tokens": "03a5012b916274ea" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 792, | |
| "padded": 792, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "hashes": { | |
| "hash_examples": "d221ec983d143dc3", | |
| "hash_full_prompts": "ac42d888e1ce1155", | |
| "hash_input_tokens": "a61a1670f854d9e1", | |
| "hash_cont_tokens": "50ab225c2f535210" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 772, | |
| "padded": 772, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "hashes": { | |
| "hash_examples": "59c2915cacfd3fbb", | |
| "hash_full_prompts": "c6bd9d25158abd0e", | |
| "hash_input_tokens": "8a77cb7763f28110", | |
| "hash_cont_tokens": "c583432ad27fcfe0" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1560, | |
| "padded": 1560, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "hashes": { | |
| "hash_examples": "1f8ac897608de342", | |
| "hash_full_prompts": "5d88f41fc2d643a8", | |
| "hash_input_tokens": "fcfcfae391f8faa1", | |
| "hash_cont_tokens": "1194078d4e38c984" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1080, | |
| "padded": 1080, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "hashes": { | |
| "hash_examples": "ead6a0f2f6c83370", | |
| "hash_full_prompts": "bfc393381298609e", | |
| "hash_input_tokens": "a29454cc1feb23ef", | |
| "hash_cont_tokens": "f47f041de50333b9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 952, | |
| "padded": 952, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "hashes": { | |
| "hash_examples": "c3f2025990afec64", | |
| "hash_full_prompts": "fc78b4997e436734", | |
| "hash_input_tokens": "b6734a25556d75dc", | |
| "hash_cont_tokens": "6296151cf7fee15c" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 604, | |
| "padded": 604, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "hashes": { | |
| "hash_examples": "21f8aab618f6d636", | |
| "hash_full_prompts": "d5c76aa40b9dbc43", | |
| "hash_input_tokens": "5720438e29473426", | |
| "hash_cont_tokens": "a490d3db0ea5935a" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 2180, | |
| "padded": 2180, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "hashes": { | |
| "hash_examples": "2386a60a11fc5de3", | |
| "hash_full_prompts": "4c5c8be5aafac432", | |
| "hash_input_tokens": "486321d5858de240", | |
| "hash_cont_tokens": "6830ef7d0325d7ef" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 864, | |
| "padded": 864, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "hashes": { | |
| "hash_examples": "74961543be40f04f", | |
| "hash_full_prompts": "5d5ca4840131ba21", | |
| "hash_input_tokens": "50c9ff438c85a69e", | |
| "hash_cont_tokens": "cdd0b3dc06d933e5" | |
| }, | |
| "truncated": 816, | |
| "non-truncated": 0, | |
| "padded": 0, | |
| "non-padded": 816, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "hashes": { | |
| "hash_examples": "2ad2f6b7198b2234", | |
| "hash_full_prompts": "11845057459afd72", | |
| "hash_input_tokens": "473919e64d1b8c80", | |
| "hash_cont_tokens": "e0203e3fc1bb0500" | |
| }, | |
| "truncated": 8, | |
| "non-truncated": 940, | |
| "padded": 940, | |
| "non-padded": 8, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "hashes": { | |
| "hash_examples": "1a7199dc733e779b", | |
| "hash_full_prompts": "756b9096b8eaf892", | |
| "hash_input_tokens": "47a65c81fd7ed010", | |
| "hash_cont_tokens": "142a4a8a1138a214" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 892, | |
| "padded": 892, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "hashes": { | |
| "hash_examples": "7acb8fdad97f88a6", | |
| "hash_full_prompts": "731a52ff15b8cfdb", | |
| "hash_input_tokens": "aedfcd41cbd2fcc9", | |
| "hash_cont_tokens": "bc54813e809b796d" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 524, | |
| "padded": 524, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "hashes": { | |
| "hash_examples": "1300bfd0dfc59114", | |
| "hash_full_prompts": "db2aefbff5eec996", | |
| "hash_input_tokens": "ed5f2414144d7b72", | |
| "hash_cont_tokens": "63435df622d5437b" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 484, | |
| "padded": 484, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "hashes": { | |
| "hash_examples": "083b1e4904c48dc2", | |
| "hash_full_prompts": "0f89ee3fe03d6a21", | |
| "hash_input_tokens": "692eaacb5b747264", | |
| "hash_cont_tokens": "e3a8cd951b6e3469" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 432, | |
| "padded": 432, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "hashes": { | |
| "hash_examples": "709128f9926a634c", | |
| "hash_full_prompts": "98a04b1f8f841069", | |
| "hash_input_tokens": "2cbce4edca937588", | |
| "hash_cont_tokens": "5e6ee2ff0404f23c" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 652, | |
| "padded": 648, | |
| "non-padded": 4, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "hashes": { | |
| "hash_examples": "88f22a636029ae47", | |
| "hash_full_prompts": "2e1c8d4b1e0cc921", | |
| "hash_input_tokens": "c2f38b19bab1aa2c", | |
| "hash_cont_tokens": "c81919424db3b267" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 448, | |
| "padded": 448, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "hashes": { | |
| "hash_examples": "8c8a1e07a2151dca", | |
| "hash_full_prompts": "f51611f514b265b0", | |
| "hash_input_tokens": "fde277bc547bc3d8", | |
| "hash_cont_tokens": "a01d6d39a83c4597" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 412, | |
| "padded": 412, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "hashes": { | |
| "hash_examples": "2668953431f91e96", | |
| "hash_full_prompts": "77562bef997c7650", | |
| "hash_input_tokens": "87b232bbebce39db", | |
| "hash_cont_tokens": "6aeaed4d823c98aa" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 936, | |
| "padded": 936, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "hashes": { | |
| "hash_examples": "9c2dda34a2ea4fd2", | |
| "hash_full_prompts": "202139046daa118f", | |
| "hash_input_tokens": "58c21af9da3e126e", | |
| "hash_cont_tokens": "50421e30bef398f9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "hashes": { | |
| "hash_examples": "41adb694024809c2", | |
| "hash_full_prompts": "bffec9fc237bcf93", | |
| "hash_input_tokens": "d1f5c770d368e9c6", | |
| "hash_cont_tokens": "9b0ab02a64603081" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 3132, | |
| "padded": 3132, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "hashes": { | |
| "hash_examples": "3171c13ba3c594c4", | |
| "hash_full_prompts": "170831fc36f1d59e", | |
| "hash_input_tokens": "98d6db15a50aaa8e", | |
| "hash_cont_tokens": "3b8bbe9108e55ce9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1384, | |
| "padded": 1354, | |
| "non-padded": 30, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "hashes": { | |
| "hash_examples": "9873e077e83e0546", | |
| "hash_full_prompts": "08f4ceba3131a068", | |
| "hash_input_tokens": "2aabd8c7337502f8", | |
| "hash_cont_tokens": "2eae753a177d5460" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 3580, | |
| "padded": 3580, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "hashes": { | |
| "hash_examples": "7db1d8142ec14323", | |
| "hash_full_prompts": "4c0e68e3586cb453", | |
| "hash_input_tokens": "17f8c8f2d4a0a9b1", | |
| "hash_cont_tokens": "29771089bd3c65c6" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1224, | |
| "padded": 1224, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "hashes": { | |
| "hash_examples": "9b455b7d72811cc8", | |
| "hash_full_prompts": "e467f822d8a0d3ff", | |
| "hash_input_tokens": "dfc6df491d991966", | |
| "hash_cont_tokens": "9f6ff69d23a48783" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1244, | |
| "padded": 1244, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "hashes": { | |
| "hash_examples": "8be90d0f538f1560", | |
| "hash_full_prompts": "152187949bcd0921", | |
| "hash_input_tokens": "cffe8139e00da9dd", | |
| "hash_cont_tokens": "a789a13af22308bf" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1296, | |
| "padded": 1296, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "hashes": { | |
| "hash_examples": "8d377597916cd07e", | |
| "hash_full_prompts": "0eb7345d6144ee0d", | |
| "hash_input_tokens": "4a69ed6ee55918fb", | |
| "hash_cont_tokens": "5129a9cfb30c5239" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1128, | |
| "padded": 1128, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "hashes": { | |
| "hash_examples": "cd9dbc52b3c932d6", | |
| "hash_full_prompts": "36ac764272bfb182", | |
| "hash_input_tokens": "6cc713f12b5890de", | |
| "hash_cont_tokens": "2e590029ef41fbcd" | |
| }, | |
| "truncated": 604, | |
| "non-truncated": 5532, | |
| "padded": 5524, | |
| "non-padded": 612, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "hashes": { | |
| "hash_examples": "b20e4e816c1e383e", | |
| "hash_full_prompts": "7b8d69ea2acaf2f7", | |
| "hash_input_tokens": "b4044fc92756c377", | |
| "hash_cont_tokens": "cd82e108370cece8" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1088, | |
| "padded": 1088, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "hashes": { | |
| "hash_examples": "d45b73b22f9cc039", | |
| "hash_full_prompts": "fe8937e9ffc99771", | |
| "hash_input_tokens": "b019784da8db089a", | |
| "hash_cont_tokens": "61ef0c8a87f9c92d" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 2448, | |
| "padded": 2448, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "hashes": { | |
| "hash_examples": "0d25072e1761652a", | |
| "hash_full_prompts": "f9adc39cfa9f42ba", | |
| "hash_input_tokens": "f47f37c7c9bfc601", | |
| "hash_cont_tokens": "568f585a259965c1" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 440, | |
| "padded": 440, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "hashes": { | |
| "hash_examples": "62bb8197e63d60d4", | |
| "hash_full_prompts": "869c9c3ae196b7c3", | |
| "hash_input_tokens": "4d282718d6142410", | |
| "hash_cont_tokens": "d70cfe096d4fb7bd" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 980, | |
| "padded": 980, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "hashes": { | |
| "hash_examples": "e7959df87dea8672", | |
| "hash_full_prompts": "1a1fc00e17b3a52a", | |
| "hash_input_tokens": "fbc6026e500537bc", | |
| "hash_cont_tokens": "c3a3bdfd177eed5b" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 804, | |
| "padded": 804, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "hashes": { | |
| "hash_examples": "4a56a01ddca44dca", | |
| "hash_full_prompts": "0c7a7081c71c07b6", | |
| "hash_input_tokens": "150dd1ff81ff642e", | |
| "hash_cont_tokens": "2568d0e8e36fa959" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 400, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "hashes": { | |
| "hash_examples": "451cc86a8c4f4fe9", | |
| "hash_full_prompts": "01e95325d8b738e4", | |
| "hash_input_tokens": "fcbac3e735545969", | |
| "hash_cont_tokens": "c178cccd753d9bc5" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 664, | |
| "padded": 664, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "hashes": { | |
| "hash_examples": "3b29cfaf1a81c379", | |
| "hash_full_prompts": "e0d79a15083dfdff", | |
| "hash_input_tokens": "ffc962a38441ef13", | |
| "hash_cont_tokens": "0a3a3ea5ef49d19c" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 684, | |
| "padded": 684, | |
| "non-padded": 0, | |
| "effective_few_shots": 5.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "hashes": { | |
| "hash_examples": "23176c0531c7b867", | |
| "hash_full_prompts": "36a6d90e75d92d4a", | |
| "hash_input_tokens": "9ffb65d225ae550f", | |
| "hash_cont_tokens": "6d1691881e252df0" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 9996, | |
| "padded": 9996, | |
| "non-padded": 0, | |
| "effective_few_shots": 0.0, | |
| "num_truncated_few_shots": 0 | |
| } | |
| }, | |
| "summary_general": { | |
| "hashes": { | |
| "hash_examples": "d84d18e9a963753d", | |
| "hash_full_prompts": "12b540783521a8e6", | |
| "hash_input_tokens": "1c61d6705b299f5c", | |
| "hash_cont_tokens": "f4b7b7f3a2788768" | |
| }, | |
| "total_evaluation_time_secondes": "7065.078456878662", | |
| "truncated": 2088, | |
| "non-truncated": 108931, | |
| "padded": 108834, | |
| "non-padded": 2185, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |