Spaces:
Runtime error
Runtime error
MLLM_leaderboard
/
eval-results
/PulsarAI
/Chat-AYB-Platypus2-13B
/results_2023-10-08T14-46-05.202813.json
| { | |
| "config_general": { | |
| "model_name": "PulsarAI/Chat-AYB-Platypus2-13B", | |
| "model_sha": "5a54eb9d5a66df4720ec52422f5627ccd94d5fd6", | |
| "model_size": "24.32 GB", | |
| "model_dtype": "torch.float16", | |
| "lighteval_sha": "0f318ecf002208468154899217b3ba7c6ae09374", | |
| "num_few_shot_default": 0, | |
| "num_fewshot_seeds": 1, | |
| "override_batch_size": 1, | |
| "max_samples": null, | |
| "job_id": "" | |
| }, | |
| "results": { | |
| "harness|arc:challenge|25": { | |
| "acc": 0.5750853242320819, | |
| "acc_stderr": 0.014445698968520765, | |
| "acc_norm": 0.6049488054607508, | |
| "acc_norm_stderr": 0.014285898292938163 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6450906193985262, | |
| "acc_stderr": 0.0047750796365670966, | |
| "acc_norm": 0.8402708623780123, | |
| "acc_norm_stderr": 0.0036560593900501065 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.3, | |
| "acc_stderr": 0.04605661864718381, | |
| "acc_norm": 0.3, | |
| "acc_norm_stderr": 0.04605661864718381 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.5037037037037037, | |
| "acc_stderr": 0.04319223625811331, | |
| "acc_norm": 0.5037037037037037, | |
| "acc_norm_stderr": 0.04319223625811331 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.6118421052631579, | |
| "acc_stderr": 0.03965842097512744, | |
| "acc_norm": 0.6118421052631579, | |
| "acc_norm_stderr": 0.03965842097512744 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.59, | |
| "acc_stderr": 0.049431107042371025, | |
| "acc_norm": 0.59, | |
| "acc_norm_stderr": 0.049431107042371025 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.5660377358490566, | |
| "acc_stderr": 0.03050329201334259, | |
| "acc_norm": 0.5660377358490566, | |
| "acc_norm_stderr": 0.03050329201334259 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.625, | |
| "acc_stderr": 0.04048439222695598, | |
| "acc_norm": 0.625, | |
| "acc_norm_stderr": 0.04048439222695598 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.41, | |
| "acc_stderr": 0.049431107042371025, | |
| "acc_norm": 0.41, | |
| "acc_norm_stderr": 0.049431107042371025 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.49, | |
| "acc_stderr": 0.05024183937956912, | |
| "acc_norm": 0.49, | |
| "acc_norm_stderr": 0.05024183937956912 | |
| }, | |
| "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": 0.5375722543352601, | |
| "acc_stderr": 0.0380168510452446, | |
| "acc_norm": 0.5375722543352601, | |
| "acc_norm_stderr": 0.0380168510452446 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.28431372549019607, | |
| "acc_stderr": 0.04488482852329017, | |
| "acc_norm": 0.28431372549019607, | |
| "acc_norm_stderr": 0.04488482852329017 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.66, | |
| "acc_stderr": 0.04760952285695237, | |
| "acc_norm": 0.66, | |
| "acc_norm_stderr": 0.04760952285695237 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.4808510638297872, | |
| "acc_stderr": 0.032662042990646775, | |
| "acc_norm": 0.4808510638297872, | |
| "acc_norm_stderr": 0.032662042990646775 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.3333333333333333, | |
| "acc_stderr": 0.044346007015849245, | |
| "acc_norm": 0.3333333333333333, | |
| "acc_norm_stderr": 0.044346007015849245 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.5379310344827586, | |
| "acc_stderr": 0.04154659671707548, | |
| "acc_norm": 0.5379310344827586, | |
| "acc_norm_stderr": 0.04154659671707548 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.3201058201058201, | |
| "acc_stderr": 0.0240268463928735, | |
| "acc_norm": 0.3201058201058201, | |
| "acc_norm_stderr": 0.0240268463928735 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.38095238095238093, | |
| "acc_stderr": 0.043435254289490965, | |
| "acc_norm": 0.38095238095238093, | |
| "acc_norm_stderr": 0.043435254289490965 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.35, | |
| "acc_stderr": 0.0479372485441102, | |
| "acc_norm": 0.35, | |
| "acc_norm_stderr": 0.0479372485441102 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.6258064516129033, | |
| "acc_stderr": 0.027528904299845697, | |
| "acc_norm": 0.6258064516129033, | |
| "acc_norm_stderr": 0.027528904299845697 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "acc": 0.42857142857142855, | |
| "acc_stderr": 0.034819048444388045, | |
| "acc_norm": 0.42857142857142855, | |
| "acc_norm_stderr": 0.034819048444388045 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "acc": 0.6, | |
| "acc_stderr": 0.049236596391733084, | |
| "acc_norm": 0.6, | |
| "acc_norm_stderr": 0.049236596391733084 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.7393939393939394, | |
| "acc_stderr": 0.03427743175816524, | |
| "acc_norm": 0.7393939393939394, | |
| "acc_norm_stderr": 0.03427743175816524 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.7373737373737373, | |
| "acc_stderr": 0.031353050095330855, | |
| "acc_norm": 0.7373737373737373, | |
| "acc_norm_stderr": 0.031353050095330855 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.8652849740932642, | |
| "acc_stderr": 0.024639789097709443, | |
| "acc_norm": 0.8652849740932642, | |
| "acc_norm_stderr": 0.024639789097709443 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.5897435897435898, | |
| "acc_stderr": 0.024939313906940784, | |
| "acc_norm": 0.5897435897435898, | |
| "acc_norm_stderr": 0.024939313906940784 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.3333333333333333, | |
| "acc_stderr": 0.028742040903948496, | |
| "acc_norm": 0.3333333333333333, | |
| "acc_norm_stderr": 0.028742040903948496 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.5588235294117647, | |
| "acc_stderr": 0.0322529423239964, | |
| "acc_norm": 0.5588235294117647, | |
| "acc_norm_stderr": 0.0322529423239964 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.33112582781456956, | |
| "acc_stderr": 0.038425817186598696, | |
| "acc_norm": 0.33112582781456956, | |
| "acc_norm_stderr": 0.038425817186598696 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.7908256880733945, | |
| "acc_stderr": 0.017437937173343233, | |
| "acc_norm": 0.7908256880733945, | |
| "acc_norm_stderr": 0.017437937173343233 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.4398148148148148, | |
| "acc_stderr": 0.03385177976044812, | |
| "acc_norm": 0.4398148148148148, | |
| "acc_norm_stderr": 0.03385177976044812 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.7990196078431373, | |
| "acc_stderr": 0.028125972265654366, | |
| "acc_norm": 0.7990196078431373, | |
| "acc_norm_stderr": 0.028125972265654366 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.759493670886076, | |
| "acc_stderr": 0.027820781981149685, | |
| "acc_norm": 0.759493670886076, | |
| "acc_norm_stderr": 0.027820781981149685 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.6681614349775785, | |
| "acc_stderr": 0.03160295143776678, | |
| "acc_norm": 0.6681614349775785, | |
| "acc_norm_stderr": 0.03160295143776678 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.6412213740458015, | |
| "acc_stderr": 0.04206739313864908, | |
| "acc_norm": 0.6412213740458015, | |
| "acc_norm_stderr": 0.04206739313864908 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.7272727272727273, | |
| "acc_stderr": 0.04065578140908707, | |
| "acc_norm": 0.7272727272727273, | |
| "acc_norm_stderr": 0.04065578140908707 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.75, | |
| "acc_stderr": 0.04186091791394607, | |
| "acc_norm": 0.75, | |
| "acc_norm_stderr": 0.04186091791394607 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.6625766871165644, | |
| "acc_stderr": 0.03714908409935574, | |
| "acc_norm": 0.6625766871165644, | |
| "acc_norm_stderr": 0.03714908409935574 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.3482142857142857, | |
| "acc_stderr": 0.04521829902833585, | |
| "acc_norm": 0.3482142857142857, | |
| "acc_norm_stderr": 0.04521829902833585 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.7087378640776699, | |
| "acc_stderr": 0.044986763205729224, | |
| "acc_norm": 0.7087378640776699, | |
| "acc_norm_stderr": 0.044986763205729224 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.8376068376068376, | |
| "acc_stderr": 0.02416161812798774, | |
| "acc_norm": 0.8376068376068376, | |
| "acc_norm_stderr": 0.02416161812798774 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.63, | |
| "acc_stderr": 0.04852365870939099, | |
| "acc_norm": 0.63, | |
| "acc_norm_stderr": 0.04852365870939099 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.7905491698595147, | |
| "acc_stderr": 0.014551310568143693, | |
| "acc_norm": 0.7905491698595147, | |
| "acc_norm_stderr": 0.014551310568143693 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.6184971098265896, | |
| "acc_stderr": 0.026152198619726803, | |
| "acc_norm": 0.6184971098265896, | |
| "acc_norm_stderr": 0.026152198619726803 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.4681564245810056, | |
| "acc_stderr": 0.016688553415612206, | |
| "acc_norm": 0.4681564245810056, | |
| "acc_norm_stderr": 0.016688553415612206 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.630718954248366, | |
| "acc_stderr": 0.027634176689602653, | |
| "acc_norm": 0.630718954248366, | |
| "acc_norm_stderr": 0.027634176689602653 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.6559485530546624, | |
| "acc_stderr": 0.02698147804364803, | |
| "acc_norm": 0.6559485530546624, | |
| "acc_norm_stderr": 0.02698147804364803 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.6790123456790124, | |
| "acc_stderr": 0.025976566010862744, | |
| "acc_norm": 0.6790123456790124, | |
| "acc_norm_stderr": 0.025976566010862744 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.4716312056737589, | |
| "acc_stderr": 0.029779450957303055, | |
| "acc_norm": 0.4716312056737589, | |
| "acc_norm_stderr": 0.029779450957303055 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.45436766623207303, | |
| "acc_stderr": 0.012716941720734815, | |
| "acc_norm": 0.45436766623207303, | |
| "acc_norm_stderr": 0.012716941720734815 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.5882352941176471, | |
| "acc_stderr": 0.02989616303312547, | |
| "acc_norm": 0.5882352941176471, | |
| "acc_norm_stderr": 0.02989616303312547 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.5735294117647058, | |
| "acc_stderr": 0.020007912739359375, | |
| "acc_norm": 0.5735294117647058, | |
| "acc_norm_stderr": 0.020007912739359375 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "acc": 0.6454545454545455, | |
| "acc_stderr": 0.045820048415054174, | |
| "acc_norm": 0.6454545454545455, | |
| "acc_norm_stderr": 0.045820048415054174 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "acc": 0.6530612244897959, | |
| "acc_stderr": 0.030472526026726496, | |
| "acc_norm": 0.6530612244897959, | |
| "acc_norm_stderr": 0.030472526026726496 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.6716417910447762, | |
| "acc_stderr": 0.033206858897443244, | |
| "acc_norm": 0.6716417910447762, | |
| "acc_norm_stderr": 0.033206858897443244 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.83, | |
| "acc_stderr": 0.03775251680686371, | |
| "acc_norm": 0.83, | |
| "acc_norm_stderr": 0.03775251680686371 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.4879518072289157, | |
| "acc_stderr": 0.03891364495835821, | |
| "acc_norm": 0.4879518072289157, | |
| "acc_norm_stderr": 0.03891364495835821 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.8187134502923976, | |
| "acc_stderr": 0.029547741687640038, | |
| "acc_norm": 0.8187134502923976, | |
| "acc_norm_stderr": 0.029547741687640038 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.3806609547123623, | |
| "mc1_stderr": 0.01699762787190793, | |
| "mc2": 0.5451670851918026, | |
| "mc2_stderr": 0.01582581744184166 | |
| }, | |
| "all": { | |
| "acc": 0.5793506755644279, | |
| "acc_stderr": 0.03418617983940258, | |
| "acc_norm": 0.5831649759747356, | |
| "acc_norm_stderr": 0.03416450490851953, | |
| "mc1": 0.3806609547123623, | |
| "mc1_stderr": 0.01699762787190793, | |
| "mc2": 0.5451670851918026, | |
| "mc2_stderr": 0.01582581744184166 | |
| } | |
| }, | |
| "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_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": "3722289b79076c44", | |
| "hash_cont_tokens": "e8abf848493b50f7" | |
| }, | |
| "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": "ececd684171f1ef2", | |
| "hash_cont_tokens": "9fe0a5c42e1532db" | |
| }, | |
| "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": "440a970fadecdc7b" | |
| }, | |
| "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": "50421e30bef398f9" | |
| }, | |
| "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": "7ecd60c25b9bfe5b" | |
| }, | |
| "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": "50421e30bef398f9" | |
| }, | |
| "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": "081bb2b524defd1c", | |
| "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": "b1cc6e7e9fcd3827" | |
| }, | |
| "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": "bd87bf0c060fd925" | |
| }, | |
| "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": "eb8932890e0605db" | |
| }, | |
| "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": "1ddcb86d28cde266" | |
| }, | |
| "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": "176c8dcff38c5f8f" | |
| }, | |
| "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": "a20a96b44dcc5b30", | |
| "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_european_history|5": { | |
| "hashes": { | |
| "hash_examples": "11cd32d0ef440171", | |
| "hash_full_prompts": "318f4513c537c6bf", | |
| "hash_input_tokens": "5002f4ac8b1562ca", | |
| "hash_cont_tokens": "674fc454bdc5ac93" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 660, | |
| "padded": 656, | |
| "non-padded": 4, | |
| "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": "873d2aab226ba1d8" | |
| }, | |
| "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": "d7907b61bcb8c123" | |
| }, | |
| "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": "0d56317b3e5eedb5" | |
| }, | |
| "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": "09ba1243e7390c0f" | |
| }, | |
| "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": "400e55b56ee6fbd7", | |
| "hash_cont_tokens": "9cc29889c3d3f77d" | |
| }, | |
| "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": "c639cce12a46ebad", | |
| "hash_cont_tokens": "cdd0b3dc06d933e5" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 816, | |
| "padded": 816, | |
| "non-padded": 0, | |
| "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": "b9762065cce6f3a6", | |
| "hash_cont_tokens": "e02816433ff28daf" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 948, | |
| "padded": 948, | |
| "non-padded": 0, | |
| "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": "8ea8c5ff76a15bca" | |
| }, | |
| "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": "3e9e0bdc248fd88a" | |
| }, | |
| "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": "55b12fb138c6a064" | |
| }, | |
| "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": "3e9bfc0362e97330" | |
| }, | |
| "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": "23b2dc6ee2da4cfc" | |
| }, | |
| "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": "d6458d743d875837" | |
| }, | |
| "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": "922a195f53a35662" | |
| }, | |
| "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": "c9f7583fff66d361", | |
| "hash_cont_tokens": "2e590029ef41fbcd" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 6136, | |
| "padded": 6136, | |
| "non-padded": 0, | |
| "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": "40a933f829116f8d", | |
| "hash_cont_tokens": "7cfee54dbddd5a98" | |
| }, | |
| "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": "a86677b2a45c20e1" | |
| }, | |
| "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": "0d756ccaae031757" | |
| }, | |
| "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": "32a03f1f22a6e103", | |
| "hash_cont_tokens": "b2229bc2cfbf594b" | |
| }, | |
| "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": "50421e30bef398f9" | |
| }, | |
| "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": "af8b3658088cb37f" | |
| }, | |
| "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": "060118bef6de4e0a" | |
| }, | |
| "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": "f5da56a132aab151" | |
| }, | |
| "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": "5c73a7dce6ccf737", | |
| "hash_cont_tokens": "71d56183130fecbd" | |
| }, | |
| "total_evaluation_time_secondes": "6380.6676704883575", | |
| "truncated": 0, | |
| "non-truncated": 111019, | |
| "padded": 110926, | |
| "non-padded": 93, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |