Spaces:
Runtime error
Runtime error
MLLM_leaderboard
/
eval-results
/TheBloke
/Llama-2-70B-fp16
/results_2023-07-31T16-40-00.231770.json
| { | |
| "results": { | |
| "harness|arc:challenge|25": { | |
| "acc": 0.6262798634812287, | |
| "acc_stderr": 0.014137708601759091, | |
| "acc_norm": 0.6732081911262798, | |
| "acc_norm_stderr": 0.013706665975587333 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6760605457080263, | |
| "acc_stderr": 0.00467020812857923, | |
| "acc_norm": 0.8733320055765784, | |
| "acc_norm_stderr": 0.0033192094001351187 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.33, | |
| "acc_stderr": 0.04725815626252605, | |
| "acc_norm": 0.33, | |
| "acc_norm_stderr": 0.04725815626252605 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.6296296296296297, | |
| "acc_stderr": 0.04171654161354544, | |
| "acc_norm": 0.6296296296296297, | |
| "acc_norm_stderr": 0.04171654161354544 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.8092105263157895, | |
| "acc_stderr": 0.031975658210325, | |
| "acc_norm": 0.8092105263157895, | |
| "acc_norm_stderr": 0.031975658210325 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.72, | |
| "acc_stderr": 0.04512608598542127, | |
| "acc_norm": 0.72, | |
| "acc_norm_stderr": 0.04512608598542127 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.7169811320754716, | |
| "acc_stderr": 0.027724236492700918, | |
| "acc_norm": 0.7169811320754716, | |
| "acc_norm_stderr": 0.027724236492700918 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.8472222222222222, | |
| "acc_stderr": 0.030085743248565666, | |
| "acc_norm": 0.8472222222222222, | |
| "acc_norm_stderr": 0.030085743248565666 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.51, | |
| "acc_stderr": 0.05024183937956912, | |
| "acc_norm": 0.51, | |
| "acc_norm_stderr": 0.05024183937956912 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.6, | |
| "acc_stderr": 0.049236596391733084, | |
| "acc_norm": 0.6, | |
| "acc_norm_stderr": 0.049236596391733084 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "acc": 0.37, | |
| "acc_stderr": 0.048523658709391, | |
| "acc_norm": 0.37, | |
| "acc_norm_stderr": 0.048523658709391 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "acc": 0.6416184971098265, | |
| "acc_stderr": 0.03656343653353159, | |
| "acc_norm": 0.6416184971098265, | |
| "acc_norm_stderr": 0.03656343653353159 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.37254901960784315, | |
| "acc_stderr": 0.04810840148082635, | |
| "acc_norm": 0.37254901960784315, | |
| "acc_norm_stderr": 0.04810840148082635 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.77, | |
| "acc_stderr": 0.04229525846816506, | |
| "acc_norm": 0.77, | |
| "acc_norm_stderr": 0.04229525846816506 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.6638297872340425, | |
| "acc_stderr": 0.030881618520676942, | |
| "acc_norm": 0.6638297872340425, | |
| "acc_norm_stderr": 0.030881618520676942 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.4473684210526316, | |
| "acc_stderr": 0.04677473004491199, | |
| "acc_norm": 0.4473684210526316, | |
| "acc_norm_stderr": 0.04677473004491199 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.6551724137931034, | |
| "acc_stderr": 0.03960933549451207, | |
| "acc_norm": 0.6551724137931034, | |
| "acc_norm_stderr": 0.03960933549451207 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.43386243386243384, | |
| "acc_stderr": 0.025525034382474894, | |
| "acc_norm": 0.43386243386243384, | |
| "acc_norm_stderr": 0.025525034382474894 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.47619047619047616, | |
| "acc_stderr": 0.04467062628403273, | |
| "acc_norm": 0.47619047619047616, | |
| "acc_norm_stderr": 0.04467062628403273 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.46, | |
| "acc_stderr": 0.05009082659620332, | |
| "acc_norm": 0.46, | |
| "acc_norm_stderr": 0.05009082659620332 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.8193548387096774, | |
| "acc_stderr": 0.02188617856717253, | |
| "acc_norm": 0.8193548387096774, | |
| "acc_norm_stderr": 0.02188617856717253 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "acc": 0.5123152709359606, | |
| "acc_stderr": 0.035169204442208966, | |
| "acc_norm": 0.5123152709359606, | |
| "acc_norm_stderr": 0.035169204442208966 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "acc": 0.79, | |
| "acc_stderr": 0.040936018074033256, | |
| "acc_norm": 0.79, | |
| "acc_norm_stderr": 0.040936018074033256 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.8303030303030303, | |
| "acc_stderr": 0.029311188674983134, | |
| "acc_norm": 0.8303030303030303, | |
| "acc_norm_stderr": 0.029311188674983134 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.8787878787878788, | |
| "acc_stderr": 0.023253157951942084, | |
| "acc_norm": 0.8787878787878788, | |
| "acc_norm_stderr": 0.023253157951942084 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.9430051813471503, | |
| "acc_stderr": 0.016731085293607555, | |
| "acc_norm": 0.9430051813471503, | |
| "acc_norm_stderr": 0.016731085293607555 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.7410256410256411, | |
| "acc_stderr": 0.02221110681006167, | |
| "acc_norm": 0.7410256410256411, | |
| "acc_norm_stderr": 0.02221110681006167 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.35555555555555557, | |
| "acc_stderr": 0.029185714949857403, | |
| "acc_norm": 0.35555555555555557, | |
| "acc_norm_stderr": 0.029185714949857403 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.7647058823529411, | |
| "acc_stderr": 0.02755361446786381, | |
| "acc_norm": 0.7647058823529411, | |
| "acc_norm_stderr": 0.02755361446786381 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.4304635761589404, | |
| "acc_stderr": 0.04042809961395634, | |
| "acc_norm": 0.4304635761589404, | |
| "acc_norm_stderr": 0.04042809961395634 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.8733944954128441, | |
| "acc_stderr": 0.014257128686165169, | |
| "acc_norm": 0.8733944954128441, | |
| "acc_norm_stderr": 0.014257128686165169 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.6342592592592593, | |
| "acc_stderr": 0.032847388576472056, | |
| "acc_norm": 0.6342592592592593, | |
| "acc_norm_stderr": 0.032847388576472056 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.8970588235294118, | |
| "acc_stderr": 0.02132833757080437, | |
| "acc_norm": 0.8970588235294118, | |
| "acc_norm_stderr": 0.02132833757080437 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.8776371308016878, | |
| "acc_stderr": 0.021331741829746786, | |
| "acc_norm": 0.8776371308016878, | |
| "acc_norm_stderr": 0.021331741829746786 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.8026905829596412, | |
| "acc_stderr": 0.02670985334496796, | |
| "acc_norm": 0.8026905829596412, | |
| "acc_norm_stderr": 0.02670985334496796 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.8778625954198473, | |
| "acc_stderr": 0.028718776889342344, | |
| "acc_norm": 0.8778625954198473, | |
| "acc_norm_stderr": 0.028718776889342344 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.8760330578512396, | |
| "acc_stderr": 0.03008309871603521, | |
| "acc_norm": 0.8760330578512396, | |
| "acc_norm_stderr": 0.03008309871603521 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.8333333333333334, | |
| "acc_stderr": 0.03602814176392645, | |
| "acc_norm": 0.8333333333333334, | |
| "acc_norm_stderr": 0.03602814176392645 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.803680981595092, | |
| "acc_stderr": 0.031207970394709218, | |
| "acc_norm": 0.803680981595092, | |
| "acc_norm_stderr": 0.031207970394709218 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.5357142857142857, | |
| "acc_stderr": 0.04733667890053756, | |
| "acc_norm": 0.5357142857142857, | |
| "acc_norm_stderr": 0.04733667890053756 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.8349514563106796, | |
| "acc_stderr": 0.03675668832233188, | |
| "acc_norm": 0.8349514563106796, | |
| "acc_norm_stderr": 0.03675668832233188 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.905982905982906, | |
| "acc_stderr": 0.01911989279892498, | |
| "acc_norm": 0.905982905982906, | |
| "acc_norm_stderr": 0.01911989279892498 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.74, | |
| "acc_stderr": 0.04408440022768077, | |
| "acc_norm": 0.74, | |
| "acc_norm_stderr": 0.04408440022768077 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.8620689655172413, | |
| "acc_stderr": 0.012331009307795656, | |
| "acc_norm": 0.8620689655172413, | |
| "acc_norm_stderr": 0.012331009307795656 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.7774566473988439, | |
| "acc_stderr": 0.02239421566194282, | |
| "acc_norm": 0.7774566473988439, | |
| "acc_norm_stderr": 0.02239421566194282 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.4547486033519553, | |
| "acc_stderr": 0.016653875777524012, | |
| "acc_norm": 0.4547486033519553, | |
| "acc_norm_stderr": 0.016653875777524012 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.7810457516339869, | |
| "acc_stderr": 0.02367908986180772, | |
| "acc_norm": 0.7810457516339869, | |
| "acc_norm_stderr": 0.02367908986180772 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.7877813504823151, | |
| "acc_stderr": 0.023222756797435115, | |
| "acc_norm": 0.7877813504823151, | |
| "acc_norm_stderr": 0.023222756797435115 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.8364197530864198, | |
| "acc_stderr": 0.020581466138257114, | |
| "acc_norm": 0.8364197530864198, | |
| "acc_norm_stderr": 0.020581466138257114 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.5673758865248227, | |
| "acc_stderr": 0.02955545423677884, | |
| "acc_norm": 0.5673758865248227, | |
| "acc_norm_stderr": 0.02955545423677884 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.5319426336375489, | |
| "acc_stderr": 0.012744149704869645, | |
| "acc_norm": 0.5319426336375489, | |
| "acc_norm_stderr": 0.012744149704869645 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.75, | |
| "acc_stderr": 0.026303648393696036, | |
| "acc_norm": 0.75, | |
| "acc_norm_stderr": 0.026303648393696036 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.7565359477124183, | |
| "acc_stderr": 0.01736247376214662, | |
| "acc_norm": 0.7565359477124183, | |
| "acc_norm_stderr": 0.01736247376214662 | |
| }, | |
| "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": 0.7918367346938775, | |
| "acc_stderr": 0.0259911176728133, | |
| "acc_norm": 0.7918367346938775, | |
| "acc_norm_stderr": 0.0259911176728133 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.900497512437811, | |
| "acc_stderr": 0.021166216304659393, | |
| "acc_norm": 0.900497512437811, | |
| "acc_norm_stderr": 0.021166216304659393 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.92, | |
| "acc_stderr": 0.0272659924344291, | |
| "acc_norm": 0.92, | |
| "acc_norm_stderr": 0.0272659924344291 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.5301204819277109, | |
| "acc_stderr": 0.03885425420866767, | |
| "acc_norm": 0.5301204819277109, | |
| "acc_norm_stderr": 0.03885425420866767 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.8538011695906432, | |
| "acc_stderr": 0.027097290118070806, | |
| "acc_norm": 0.8538011695906432, | |
| "acc_norm_stderr": 0.027097290118070806 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.3108935128518972, | |
| "mc1_stderr": 0.016203316673559696, | |
| "mc2": 0.44923493721887353, | |
| "mc2_stderr": 0.01390226410719232 | |
| }, | |
| "all": { | |
| "acc": 0.6967225637378714, | |
| "acc_stderr": 0.030867069907791145, | |
| "acc_norm": 0.7008615431872544, | |
| "acc_norm_stderr": 0.030836865817034945, | |
| "mc1": 0.3108935128518972, | |
| "mc1_stderr": 0.016203316673559696, | |
| "mc2": 0.44923493721887353, | |
| "mc2_stderr": 0.01390226410719232 | |
| } | |
| }, | |
| "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": "TheBloke/Llama-2-70B-fp16", | |
| "model_sha": "b25061ef1b440e970d15d4ac99bc42937cd442a2", | |
| "model_dtype": "torch.float16", | |
| "lighteval_sha": "03c2fad20ff7f5334c33cfee459024b8d7e4a109", | |
| "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": "61571bf68d6d89aa", | |
| "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": "29906669b1c7054a", | |
| "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": "c54ff61ad0273dd7", | |
| "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": "be31a1e22aef5f90", | |
| "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": "277a7b1fad566940", | |
| "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": "ba552605bc116de5", | |
| "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": "428c7563d0b98ab9", | |
| "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": "da036601573942e2", | |
| "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": "94e0196d6aded13d", | |
| "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": "6e4d0f4a8d36690b", | |
| "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": "614054d17109a25d", | |
| "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": "1d633b3cc0524ba8", | |
| "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": "5421d9a1af86cbd4", | |
| "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": "5e6b70ecb333cf18", | |
| "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": "c2ef11a87264ceed", | |
| "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": "ecaccd912a4c3978", | |
| "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": "1590c84291399be8", | |
| "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": "3269597f715b0da1", | |
| "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": "a2800d20f3ab8d7c", | |
| "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": "94ed44b3772505ad", | |
| "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": "24423acb928db768", | |
| "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": "831ff35c474e5cef", | |
| "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": "8c34e0f2bda77358", | |
| "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": "7c5547c7da5bc793", | |
| "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": "f62991cb6a496b05", | |
| "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": "4cef2aff6e3d59ed", | |
| "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": "6e2577ea4082ed2b", | |
| "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": "c5fc9aeb1079c8e4", | |
| "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": "555fc385cffa84ca", | |
| "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": "febd23cbf9973b7f", | |
| "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": "424b02981230ee83", | |
| "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": "054824cc474caef5", | |
| "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": "541a75f071dcf579", | |
| "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": "04269e5c5a257dd9", | |
| "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": "d93ba9d9d38e4397", | |
| "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": "9eeaccd2698b4f5a", | |
| "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": "b4f08f544f2b7576", | |
| "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": "900c2a51f1174b9f", | |
| "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": "6b36efb4689c6eca", | |
| "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": "2aaac78a0cfed47a", | |
| "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": "886ca823b41c094a", | |
| "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": "72fd71de7675e7d0", | |
| "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": "f3ca0dd8e7a1eb09", | |
| "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": "3e793631e951f23c", | |
| "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": "59753c2144ea93af", | |
| "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": "bd8d3dbed15a8c34", | |
| "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": "3573cd87facbb7c5", | |
| "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": "17e721bc1a7cbb47", | |
| "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": "9178e10bd0763ec4", | |
| "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": "f5a22012a54f70ea", | |
| "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": "0dfb73a8eb3f692c", | |
| "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": "1710c6ba4c9f3cbd", | |
| "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": "d49711415961ced7", | |
| "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": "828999f7624cbe7e", | |
| "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": "42054621e718dbee", | |
| "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": "6c4f0aa4dc859c04", | |
| "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": "6c75d44e092ff24f", | |
| "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": "2738d7ed7075faa7", | |
| "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": "6fecf578c508db6a", | |
| "hash_cont_tokens": "f4b7b7f3a2788768" | |
| }, | |
| "total_evaluation_time_secondes": "26263.30303668976", | |
| "truncated": 2088, | |
| "non-truncated": 108931, | |
| "padded": 108834, | |
| "non-padded": 2185, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |