Spaces:
Runtime error
Runtime error
| { | |
| "config_general": { | |
| "model_name": "PulsarAI/Nebula-7B", | |
| "model_sha": "569f848698a468fb03d37033c67f3734bbaec127", | |
| "model_size": "13.99 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": 0.5418088737201365, | |
| "acc_stderr": 0.0145602203087147, | |
| "acc_norm": 0.5930034129692833, | |
| "acc_norm_stderr": 0.014356399418009121 | |
| }, | |
| "harness|hellaswag|10": { | |
| "acc": 0.6342362079267079, | |
| "acc_stderr": 0.004806593424942265, | |
| "acc_norm": 0.8345947022505477, | |
| "acc_norm_stderr": 0.0037078660457296048 | |
| }, | |
| "harness|hendrycksTest-abstract_algebra|5": { | |
| "acc": 0.34, | |
| "acc_stderr": 0.04760952285695235, | |
| "acc_norm": 0.34, | |
| "acc_norm_stderr": 0.04760952285695235 | |
| }, | |
| "harness|hendrycksTest-anatomy|5": { | |
| "acc": 0.5555555555555556, | |
| "acc_stderr": 0.04292596718256981, | |
| "acc_norm": 0.5555555555555556, | |
| "acc_norm_stderr": 0.04292596718256981 | |
| }, | |
| "harness|hendrycksTest-astronomy|5": { | |
| "acc": 0.5986842105263158, | |
| "acc_stderr": 0.039889037033362836, | |
| "acc_norm": 0.5986842105263158, | |
| "acc_norm_stderr": 0.039889037033362836 | |
| }, | |
| "harness|hendrycksTest-business_ethics|5": { | |
| "acc": 0.43, | |
| "acc_stderr": 0.04975698519562428, | |
| "acc_norm": 0.43, | |
| "acc_norm_stderr": 0.04975698519562428 | |
| }, | |
| "harness|hendrycksTest-clinical_knowledge|5": { | |
| "acc": 0.5773584905660377, | |
| "acc_stderr": 0.03040233144576954, | |
| "acc_norm": 0.5773584905660377, | |
| "acc_norm_stderr": 0.03040233144576954 | |
| }, | |
| "harness|hendrycksTest-college_biology|5": { | |
| "acc": 0.6944444444444444, | |
| "acc_stderr": 0.03852084696008534, | |
| "acc_norm": 0.6944444444444444, | |
| "acc_norm_stderr": 0.03852084696008534 | |
| }, | |
| "harness|hendrycksTest-college_chemistry|5": { | |
| "acc": 0.43, | |
| "acc_stderr": 0.049756985195624284, | |
| "acc_norm": 0.43, | |
| "acc_norm_stderr": 0.049756985195624284 | |
| }, | |
| "harness|hendrycksTest-college_computer_science|5": { | |
| "acc": 0.44, | |
| "acc_stderr": 0.04988876515698589, | |
| "acc_norm": 0.44, | |
| "acc_norm_stderr": 0.04988876515698589 | |
| }, | |
| "harness|hendrycksTest-college_mathematics|5": { | |
| "acc": 0.29, | |
| "acc_stderr": 0.045604802157206845, | |
| "acc_norm": 0.29, | |
| "acc_norm_stderr": 0.045604802157206845 | |
| }, | |
| "harness|hendrycksTest-college_medicine|5": { | |
| "acc": 0.5202312138728323, | |
| "acc_stderr": 0.03809342081273957, | |
| "acc_norm": 0.5202312138728323, | |
| "acc_norm_stderr": 0.03809342081273957 | |
| }, | |
| "harness|hendrycksTest-college_physics|5": { | |
| "acc": 0.3137254901960784, | |
| "acc_stderr": 0.04617034827006715, | |
| "acc_norm": 0.3137254901960784, | |
| "acc_norm_stderr": 0.04617034827006715 | |
| }, | |
| "harness|hendrycksTest-computer_security|5": { | |
| "acc": 0.68, | |
| "acc_stderr": 0.046882617226215034, | |
| "acc_norm": 0.68, | |
| "acc_norm_stderr": 0.046882617226215034 | |
| }, | |
| "harness|hendrycksTest-conceptual_physics|5": { | |
| "acc": 0.5063829787234042, | |
| "acc_stderr": 0.03268335899936336, | |
| "acc_norm": 0.5063829787234042, | |
| "acc_norm_stderr": 0.03268335899936336 | |
| }, | |
| "harness|hendrycksTest-econometrics|5": { | |
| "acc": 0.4649122807017544, | |
| "acc_stderr": 0.04692008381368909, | |
| "acc_norm": 0.4649122807017544, | |
| "acc_norm_stderr": 0.04692008381368909 | |
| }, | |
| "harness|hendrycksTest-electrical_engineering|5": { | |
| "acc": 0.503448275862069, | |
| "acc_stderr": 0.041665675771015785, | |
| "acc_norm": 0.503448275862069, | |
| "acc_norm_stderr": 0.041665675771015785 | |
| }, | |
| "harness|hendrycksTest-elementary_mathematics|5": { | |
| "acc": 0.38095238095238093, | |
| "acc_stderr": 0.02501074911613759, | |
| "acc_norm": 0.38095238095238093, | |
| "acc_norm_stderr": 0.02501074911613759 | |
| }, | |
| "harness|hendrycksTest-formal_logic|5": { | |
| "acc": 0.40476190476190477, | |
| "acc_stderr": 0.043902592653775635, | |
| "acc_norm": 0.40476190476190477, | |
| "acc_norm_stderr": 0.043902592653775635 | |
| }, | |
| "harness|hendrycksTest-global_facts|5": { | |
| "acc": 0.34, | |
| "acc_stderr": 0.04760952285695235, | |
| "acc_norm": 0.34, | |
| "acc_norm_stderr": 0.04760952285695235 | |
| }, | |
| "harness|hendrycksTest-high_school_biology|5": { | |
| "acc": 0.6516129032258065, | |
| "acc_stderr": 0.027104826328100944, | |
| "acc_norm": 0.6516129032258065, | |
| "acc_norm_stderr": 0.027104826328100944 | |
| }, | |
| "harness|hendrycksTest-high_school_chemistry|5": { | |
| "acc": 0.47783251231527096, | |
| "acc_stderr": 0.035145285621750094, | |
| "acc_norm": 0.47783251231527096, | |
| "acc_norm_stderr": 0.035145285621750094 | |
| }, | |
| "harness|hendrycksTest-high_school_computer_science|5": { | |
| "acc": 0.56, | |
| "acc_stderr": 0.049888765156985884, | |
| "acc_norm": 0.56, | |
| "acc_norm_stderr": 0.049888765156985884 | |
| }, | |
| "harness|hendrycksTest-high_school_european_history|5": { | |
| "acc": 0.696969696969697, | |
| "acc_stderr": 0.03588624800091707, | |
| "acc_norm": 0.696969696969697, | |
| "acc_norm_stderr": 0.03588624800091707 | |
| }, | |
| "harness|hendrycksTest-high_school_geography|5": { | |
| "acc": 0.7525252525252525, | |
| "acc_stderr": 0.030746300742124495, | |
| "acc_norm": 0.7525252525252525, | |
| "acc_norm_stderr": 0.030746300742124495 | |
| }, | |
| "harness|hendrycksTest-high_school_government_and_politics|5": { | |
| "acc": 0.8082901554404145, | |
| "acc_stderr": 0.028408953626245282, | |
| "acc_norm": 0.8082901554404145, | |
| "acc_norm_stderr": 0.028408953626245282 | |
| }, | |
| "harness|hendrycksTest-high_school_macroeconomics|5": { | |
| "acc": 0.5333333333333333, | |
| "acc_stderr": 0.02529460802398647, | |
| "acc_norm": 0.5333333333333333, | |
| "acc_norm_stderr": 0.02529460802398647 | |
| }, | |
| "harness|hendrycksTest-high_school_mathematics|5": { | |
| "acc": 0.2814814814814815, | |
| "acc_stderr": 0.02742001935094527, | |
| "acc_norm": 0.2814814814814815, | |
| "acc_norm_stderr": 0.02742001935094527 | |
| }, | |
| "harness|hendrycksTest-high_school_microeconomics|5": { | |
| "acc": 0.5588235294117647, | |
| "acc_stderr": 0.032252942323996406, | |
| "acc_norm": 0.5588235294117647, | |
| "acc_norm_stderr": 0.032252942323996406 | |
| }, | |
| "harness|hendrycksTest-high_school_physics|5": { | |
| "acc": 0.3443708609271523, | |
| "acc_stderr": 0.03879687024073327, | |
| "acc_norm": 0.3443708609271523, | |
| "acc_norm_stderr": 0.03879687024073327 | |
| }, | |
| "harness|hendrycksTest-high_school_psychology|5": { | |
| "acc": 0.7743119266055046, | |
| "acc_stderr": 0.017923087667803064, | |
| "acc_norm": 0.7743119266055046, | |
| "acc_norm_stderr": 0.017923087667803064 | |
| }, | |
| "harness|hendrycksTest-high_school_statistics|5": { | |
| "acc": 0.39351851851851855, | |
| "acc_stderr": 0.03331747876370312, | |
| "acc_norm": 0.39351851851851855, | |
| "acc_norm_stderr": 0.03331747876370312 | |
| }, | |
| "harness|hendrycksTest-high_school_us_history|5": { | |
| "acc": 0.7156862745098039, | |
| "acc_stderr": 0.031660096793998116, | |
| "acc_norm": 0.7156862745098039, | |
| "acc_norm_stderr": 0.031660096793998116 | |
| }, | |
| "harness|hendrycksTest-high_school_world_history|5": { | |
| "acc": 0.7510548523206751, | |
| "acc_stderr": 0.028146970599422644, | |
| "acc_norm": 0.7510548523206751, | |
| "acc_norm_stderr": 0.028146970599422644 | |
| }, | |
| "harness|hendrycksTest-human_aging|5": { | |
| "acc": 0.6681614349775785, | |
| "acc_stderr": 0.03160295143776679, | |
| "acc_norm": 0.6681614349775785, | |
| "acc_norm_stderr": 0.03160295143776679 | |
| }, | |
| "harness|hendrycksTest-human_sexuality|5": { | |
| "acc": 0.6793893129770993, | |
| "acc_stderr": 0.04093329229834278, | |
| "acc_norm": 0.6793893129770993, | |
| "acc_norm_stderr": 0.04093329229834278 | |
| }, | |
| "harness|hendrycksTest-international_law|5": { | |
| "acc": 0.7355371900826446, | |
| "acc_stderr": 0.040261875275912046, | |
| "acc_norm": 0.7355371900826446, | |
| "acc_norm_stderr": 0.040261875275912046 | |
| }, | |
| "harness|hendrycksTest-jurisprudence|5": { | |
| "acc": 0.6851851851851852, | |
| "acc_stderr": 0.04489931073591311, | |
| "acc_norm": 0.6851851851851852, | |
| "acc_norm_stderr": 0.04489931073591311 | |
| }, | |
| "harness|hendrycksTest-logical_fallacies|5": { | |
| "acc": 0.6932515337423313, | |
| "acc_stderr": 0.03623089915724146, | |
| "acc_norm": 0.6932515337423313, | |
| "acc_norm_stderr": 0.03623089915724146 | |
| }, | |
| "harness|hendrycksTest-machine_learning|5": { | |
| "acc": 0.38392857142857145, | |
| "acc_stderr": 0.04616143075028547, | |
| "acc_norm": 0.38392857142857145, | |
| "acc_norm_stderr": 0.04616143075028547 | |
| }, | |
| "harness|hendrycksTest-management|5": { | |
| "acc": 0.6990291262135923, | |
| "acc_stderr": 0.04541609446503947, | |
| "acc_norm": 0.6990291262135923, | |
| "acc_norm_stderr": 0.04541609446503947 | |
| }, | |
| "harness|hendrycksTest-marketing|5": { | |
| "acc": 0.8205128205128205, | |
| "acc_stderr": 0.025140935950335442, | |
| "acc_norm": 0.8205128205128205, | |
| "acc_norm_stderr": 0.025140935950335442 | |
| }, | |
| "harness|hendrycksTest-medical_genetics|5": { | |
| "acc": 0.66, | |
| "acc_stderr": 0.04760952285695237, | |
| "acc_norm": 0.66, | |
| "acc_norm_stderr": 0.04760952285695237 | |
| }, | |
| "harness|hendrycksTest-miscellaneous|5": { | |
| "acc": 0.8007662835249042, | |
| "acc_stderr": 0.014283378044296422, | |
| "acc_norm": 0.8007662835249042, | |
| "acc_norm_stderr": 0.014283378044296422 | |
| }, | |
| "harness|hendrycksTest-moral_disputes|5": { | |
| "acc": 0.6242774566473989, | |
| "acc_stderr": 0.02607431485165708, | |
| "acc_norm": 0.6242774566473989, | |
| "acc_norm_stderr": 0.02607431485165708 | |
| }, | |
| "harness|hendrycksTest-moral_scenarios|5": { | |
| "acc": 0.2659217877094972, | |
| "acc_stderr": 0.014776765066438885, | |
| "acc_norm": 0.2659217877094972, | |
| "acc_norm_stderr": 0.014776765066438885 | |
| }, | |
| "harness|hendrycksTest-nutrition|5": { | |
| "acc": 0.6274509803921569, | |
| "acc_stderr": 0.027684181883302888, | |
| "acc_norm": 0.6274509803921569, | |
| "acc_norm_stderr": 0.027684181883302888 | |
| }, | |
| "harness|hendrycksTest-philosophy|5": { | |
| "acc": 0.662379421221865, | |
| "acc_stderr": 0.026858825879488544, | |
| "acc_norm": 0.662379421221865, | |
| "acc_norm_stderr": 0.026858825879488544 | |
| }, | |
| "harness|hendrycksTest-prehistory|5": { | |
| "acc": 0.7067901234567902, | |
| "acc_stderr": 0.025329888171900926, | |
| "acc_norm": 0.7067901234567902, | |
| "acc_norm_stderr": 0.025329888171900926 | |
| }, | |
| "harness|hendrycksTest-professional_accounting|5": { | |
| "acc": 0.44680851063829785, | |
| "acc_stderr": 0.02965823509766691, | |
| "acc_norm": 0.44680851063829785, | |
| "acc_norm_stderr": 0.02965823509766691 | |
| }, | |
| "harness|hendrycksTest-professional_law|5": { | |
| "acc": 0.4406779661016949, | |
| "acc_stderr": 0.012680037994097065, | |
| "acc_norm": 0.4406779661016949, | |
| "acc_norm_stderr": 0.012680037994097065 | |
| }, | |
| "harness|hendrycksTest-professional_medicine|5": { | |
| "acc": 0.5441176470588235, | |
| "acc_stderr": 0.03025437257397671, | |
| "acc_norm": 0.5441176470588235, | |
| "acc_norm_stderr": 0.03025437257397671 | |
| }, | |
| "harness|hendrycksTest-professional_psychology|5": { | |
| "acc": 0.5980392156862745, | |
| "acc_stderr": 0.019835176484375383, | |
| "acc_norm": 0.5980392156862745, | |
| "acc_norm_stderr": 0.019835176484375383 | |
| }, | |
| "harness|hendrycksTest-public_relations|5": { | |
| "acc": 0.6181818181818182, | |
| "acc_stderr": 0.046534298079135075, | |
| "acc_norm": 0.6181818181818182, | |
| "acc_norm_stderr": 0.046534298079135075 | |
| }, | |
| "harness|hendrycksTest-security_studies|5": { | |
| "acc": 0.5142857142857142, | |
| "acc_stderr": 0.03199615232806286, | |
| "acc_norm": 0.5142857142857142, | |
| "acc_norm_stderr": 0.03199615232806286 | |
| }, | |
| "harness|hendrycksTest-sociology|5": { | |
| "acc": 0.7661691542288557, | |
| "acc_stderr": 0.029929415408348377, | |
| "acc_norm": 0.7661691542288557, | |
| "acc_norm_stderr": 0.029929415408348377 | |
| }, | |
| "harness|hendrycksTest-us_foreign_policy|5": { | |
| "acc": 0.83, | |
| "acc_stderr": 0.0377525168068637, | |
| "acc_norm": 0.83, | |
| "acc_norm_stderr": 0.0377525168068637 | |
| }, | |
| "harness|hendrycksTest-virology|5": { | |
| "acc": 0.4819277108433735, | |
| "acc_stderr": 0.03889951252827216, | |
| "acc_norm": 0.4819277108433735, | |
| "acc_norm_stderr": 0.03889951252827216 | |
| }, | |
| "harness|hendrycksTest-world_religions|5": { | |
| "acc": 0.7660818713450293, | |
| "acc_stderr": 0.03246721765117826, | |
| "acc_norm": 0.7660818713450293, | |
| "acc_norm_stderr": 0.03246721765117826 | |
| }, | |
| "harness|truthfulqa:mc|0": { | |
| "mc1": 0.31334149326805383, | |
| "mc1_stderr": 0.0162380650690596, | |
| "mc2": 0.45561649492894496, | |
| "mc2_stderr": 0.014644899277894422 | |
| }, | |
| "all": { | |
| "acc": 0.570596346471807, | |
| "acc_stderr": 0.034371584431446715, | |
| "acc_norm": 0.5748599572103322, | |
| "acc_norm_stderr": 0.03434950734212607, | |
| "mc1": 0.31334149326805383, | |
| "mc1_stderr": 0.0162380650690596, | |
| "mc2": 0.45561649492894496, | |
| "mc2_stderr": 0.014644899277894422 | |
| } | |
| }, | |
| "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": "e43adcaa871b1364", | |
| "hash_cont_tokens": "289aa98c400841d8" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 4687, | |
| "padded": 4684, | |
| "non-padded": 3, | |
| "effective_few_shots": 25.0, | |
| "num_truncated_few_shots": 0 | |
| }, | |
| "harness|hellaswag|10": { | |
| "hashes": { | |
| "hash_examples": "e1768ecb99d7ecf0", | |
| "hash_full_prompts": "0b4c16983130f84f", | |
| "hash_input_tokens": "08da6b3d0798f3e5", | |
| "hash_cont_tokens": "ac460260c3e6efc9" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 40168, | |
| "padded": 40039, | |
| "non-padded": 129, | |
| "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": "5e2b26eb9b4d08bf", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "d33cda9df28030eb", | |
| "hash_cont_tokens": "a52a4f60d98cbe5c" | |
| }, | |
| "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": "0dd50c500d64c57d", | |
| "hash_cont_tokens": "10f7d8eeba97841d" | |
| }, | |
| "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": "40b524d0df3defc2", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "1f87d12d677e0dfd", | |
| "hash_cont_tokens": "edef9975ba9165b5" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1060, | |
| "padded": 1056, | |
| "non-padded": 4, | |
| "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": "dd6d69d8b13afbeb", | |
| "hash_cont_tokens": "0aa103ec6602280b" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 576, | |
| "padded": 572, | |
| "non-padded": 4, | |
| "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": "d45f3c401a00e97e", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "c04f21d954ae67b2", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "e7de03b4e1a407d8", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "9ce9516475f0b09c", | |
| "hash_cont_tokens": "1979021dbc698754" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 692, | |
| "padded": 684, | |
| "non-padded": 8, | |
| "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": "f749592a0d6c967d", | |
| "hash_cont_tokens": "7cf7fe2bab00acbd" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 408, | |
| "padded": 404, | |
| "non-padded": 4, | |
| "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": "1a6dccf2066f3598", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "6ce98c8aec8e7514", | |
| "hash_cont_tokens": "903f64eed2b0d217" | |
| }, | |
| "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": "7794b03bf6b9bb11", | |
| "hash_cont_tokens": "721ae6c5302c4bf2" | |
| }, | |
| "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": "e47ff85e05850517", | |
| "hash_cont_tokens": "15a738960ed3e587" | |
| }, | |
| "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": "2ce6901704311790", | |
| "hash_cont_tokens": "c96470462fc71683" | |
| }, | |
| "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": "fa49c3faa72a3955", | |
| "hash_cont_tokens": "0e1ce025c9d6ee7e" | |
| }, | |
| "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": "38992a391c7040d5", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 400, | |
| "padded": 396, | |
| "non-padded": 4, | |
| "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": "4944fad6e0578120", | |
| "hash_cont_tokens": "e34d57f7d3c4ca16" | |
| }, | |
| "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": "bec955dfccee0331", | |
| "hash_cont_tokens": "e8482d44df4b3740" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 812, | |
| "padded": 796, | |
| "non-padded": 16, | |
| "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": "2ccfe020e0a8e824", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "5e5e8bf3808e0ead", | |
| "hash_cont_tokens": "d63e679a49418339" | |
| }, | |
| "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": "6a624d76e1b40f9d", | |
| "hash_cont_tokens": "d78483e286d06f1a" | |
| }, | |
| "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": "8340aed0285230f4", | |
| "hash_cont_tokens": "691cdff71ff5fe57" | |
| }, | |
| "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": "ca47137b1f3a769c", | |
| "hash_cont_tokens": "d5ad4c5bdca967ad" | |
| }, | |
| "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": "c9d341ab62890f30", | |
| "hash_cont_tokens": "8f631ca5687dd0d4" | |
| }, | |
| "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": "62573d06618ae7df", | |
| "hash_cont_tokens": "7321048a28451473" | |
| }, | |
| "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": "ddddcaae96263221", | |
| "hash_cont_tokens": "bb137581f269861c" | |
| }, | |
| "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": "ef9c1ae343139fdd", | |
| "hash_cont_tokens": "b455cab2675bd863" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 2180, | |
| "padded": 2161, | |
| "non-padded": 19, | |
| "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": "eb4abd87b0e863cc", | |
| "hash_cont_tokens": "1b3196fec7e58037" | |
| }, | |
| "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": "63548c7fa9ba7a78", | |
| "hash_cont_tokens": "a331dedc2aa01b3e" | |
| }, | |
| "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": "83c5da18bfa50812", | |
| "hash_cont_tokens": "d0fbe030b8c8c2bf" | |
| }, | |
| "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": "c93c778cb8c58a32", | |
| "hash_cont_tokens": "1dd29c3755494850" | |
| }, | |
| "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": "1daed91f54b42f7d", | |
| "hash_cont_tokens": "c85573f663c10691" | |
| }, | |
| "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": "cfdae69f75ee8670", | |
| "hash_cont_tokens": "d263804ba918154f" | |
| }, | |
| "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": "173979adbb5ab44e", | |
| "hash_cont_tokens": "581986691a84ece8" | |
| }, | |
| "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": "7b7d06271aff55ff", | |
| "hash_cont_tokens": "55a858b28bbda458" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 652, | |
| "padded": 652, | |
| "non-padded": 0, | |
| "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": "ca062cfd7c7fddcb", | |
| "hash_cont_tokens": "e99d3d3efd4ac7a3" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 448, | |
| "padded": 445, | |
| "non-padded": 3, | |
| "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": "fc47171ffb714da3", | |
| "hash_cont_tokens": "13d9dc56bca34726" | |
| }, | |
| "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": "aa29e9d883670c8f", | |
| "hash_cont_tokens": "2700ea26933916a2" | |
| }, | |
| "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": "88ad044b653ecaa5", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "f9e7e01573277484", | |
| "hash_cont_tokens": "7bf4341c79587250" | |
| }, | |
| "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": "03728b9e48594c28", | |
| "hash_cont_tokens": "38a48e9de6976f00" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 1384, | |
| "padded": 1360, | |
| "non-padded": 24, | |
| "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": "04a903966514d177", | |
| "hash_cont_tokens": "761c4dc187689d89" | |
| }, | |
| "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": "a2176d3ac6f01cf0", | |
| "hash_cont_tokens": "65005bd7d6f6012a" | |
| }, | |
| "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": "a96dc872948245a8", | |
| "hash_cont_tokens": "0b47934fb6314dec" | |
| }, | |
| "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": "e0b03637947e9efa", | |
| "hash_cont_tokens": "3f20acd855ee0a29" | |
| }, | |
| "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": "0b4c6d0e49c47ab4", | |
| "hash_cont_tokens": "8f122ba881355d4b" | |
| }, | |
| "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": "bcbdbbde22ec73e3", | |
| "hash_cont_tokens": "90d5df417c4d3fd3" | |
| }, | |
| "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": "c54d753563114d45", | |
| "hash_cont_tokens": "4a2d2988884f7f70" | |
| }, | |
| "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": "9e6e34f48034edc0", | |
| "hash_cont_tokens": "e0a952cb8a9c81de" | |
| }, | |
| "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": "634feb3f97d1064d", | |
| "hash_cont_tokens": "1fa77a8dff3922b8" | |
| }, | |
| "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": "ca8497342e5b1d57", | |
| "hash_cont_tokens": "81fc9cb3cbdd52db" | |
| }, | |
| "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": "ae361375c940a0fb", | |
| "hash_cont_tokens": "2a0493252ed2cf43" | |
| }, | |
| "truncated": 0, | |
| "non-truncated": 804, | |
| "padded": 800, | |
| "non-padded": 4, | |
| "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": "e8bdf33cf82d89f5", | |
| "hash_cont_tokens": "17b868b63507f9a3" | |
| }, | |
| "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": "32ce831e0ba2d2e2", | |
| "hash_cont_tokens": "5ab892d003b00c98" | |
| }, | |
| "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": "4ed9b68c5694211b", | |
| "hash_cont_tokens": "15a5e5dbdfbb8568" | |
| }, | |
| "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": "a30fbd9af05d717a", | |
| "hash_cont_tokens": "5a8d4bb398b1c3c0" | |
| }, | |
| "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": "3d86ffeb7677bd9d", | |
| "hash_cont_tokens": "35527140510ee91a" | |
| }, | |
| "total_evaluation_time_secondes": "4129.603856563568", | |
| "truncated": 0, | |
| "non-truncated": 111019, | |
| "padded": 110793, | |
| "non-padded": 226, | |
| "num_truncated_few_shots": 0 | |
| } | |
| } |