open-llm-leaderboard/details_EleutherAI__pythia-1.3b
收藏数据集概述
数据集简介
该数据集是在对模型 EleutherAI/pythia-1.3b 进行评估运行期间自动创建的,用于 Open LLM Leaderboard。
数据集结构
- 数据集包含 64 个配置,每个配置对应一个评估任务。
- 数据集从 2 次运行中创建,每次运行在每个配置中都有一个特定的分割,分割名称使用运行的时间戳。
- "train" 分割始终指向最新的结果。
- 额外的 "results" 配置存储所有运行的聚合结果,用于计算和显示 Open LLM Leaderboard 上的聚合指标。
数据加载示例
python from datasets import load_dataset data = load_dataset("open-llm-leaderboard/details_EleutherAI__pythia-1.3b", "harness_winogrande_5", split="train")
最新结果
以下是 2023-10-21T20:31:22.068379 运行 的最新结果: python { "all": { "em": 0.0010486577181208054, "em_stderr": 0.0003314581465219287, "f1": 0.040563129194630954, "f1_stderr": 0.0011177096979539825, "acc": 0.29182616042743625, "acc_stderr": 0.008309831271227 }, "harness|drop|3": { "em": 0.0010486577181208054, "em_stderr": 0.0003314581465219287, "f1": 0.040563129194630954, "f1_stderr": 0.0011177096979539825 }, "harness|gsm8k|5": { "acc": 0.009855951478392721, "acc_stderr": 0.00272107657704166 }, "harness|winogrande|5": { "acc": 0.5737963693764798, "acc_stderr": 0.013898585965412338 } }
配置详情
以下是数据集的配置详情:
配置列表
harness_arc_challenge_25harness_drop_3harness_gsm8k_5harness_hellaswag_10harness_hendrycksTest_5harness_hendrycksTest_abstract_algebra_5harness_hendrycksTest_anatomy_5harness_hendrycksTest_astronomy_5harness_hendrycksTest_business_ethics_5harness_hendrycksTest_clinical_knowledge_5harness_hendrycksTest_college_biology_5harness_hendrycksTest_college_chemistry_5harness_hendrycksTest_college_computer_science_5harness_hendrycksTest_college_mathematics_5harness_hendrycksTest_college_medicine_5harness_hendrycksTest_college_physics_5harness_hendrycksTest_computer_security_5harness_hendrycksTest_conceptual_physics_5harness_hendrycksTest_econometrics_5harness_hendrycksTest_electrical_engineering_5harness_hendrycksTest_elementary_mathematics_5harness_hendrycksTest_formal_logic_5harness_hendrycksTest_global_facts_5harness_hendrycksTest_high_school_biology_5harness_hendrycksTest_high_school_chemistry_5harness_hendrycksTest_high_school_computer_science_5harness_hendrycksTest_high_school_european_history_5harness_hendrycksTest_high_school_geography_5harness_hendrycksTest_high_school_government_and_politics_5harness_hendrycksTest_high_school_macroeconomics_5harness_hendrycksTest_high_school_mathematics_5harness_hendrycksTest_high_school_microeconomics_5harness_hendrycksTest_high_school_physics_5harness_hendrycksTest_high_school_psychology_5harness_hendrycksTest_high_school_statistics_5harness_hendrycksTest_high_school_us_history_5harness_hendrycksTest_high_school_world_history_5harness_hendrycksTest_human_aging_5harness_hendrycksTest_human_sexuality_5harness_hendrycksTest_international_law_5harness_hendrycksTest_jurisprudence_5harness_hendrycksTest_logical_fallacies_5harness_hendrycksTest_machine_learning_5harness_hendrycksTest_management_5harness_hendrycksTest_marketing_5harness_hendrycksTest_medical_genetics_5harness_hendrycksTest_miscellaneous_5harness_hendrycksTest_moral_disputes_5harness_hendrycksTest_moral_scenarios_5harness_hendrycksTest_nutrition_5harness_hendrycksTest_philosophy_5harness_hendrycksTest_prehistory_5harness_hendrycksTest_professional_accounting_5harness_hendrycksTest_professional_law_5harness_hendrycksTest_professional_medicine_5harness_hendrycksTest_professional_psychology_5harness_hendrycksTest_public_relations_5harness_hendrycksTest_security_studies_5harness_hendrycksTest_sociology_5harness_hendrycksTest_us_foreign_policy_5harness_hendrycksTest_virology_5harness_hendrycksTest_world_religions_5
数据文件路径
每个配置包含两个分割:
2023_07_19T15_01_09.572948latest
每个分割对应一个或多个数据文件路径,具体路径格式为 **/details_harness|任务名称|编号_时间戳.parquet。



