open-llm-leaderboard-old/details_janhq__supermario-slerp-v2

该数据集是在评估模型janhq/supermario-slerp-v2时自动创建的，主要用于在Open LLM Leaderboard上进行评估。数据集包含63个配置，每个配置对应一个评估任务。数据集由1次运行生成，每次运行的结果作为特定配置中的一个分割，分割名称使用运行的时间戳。train分割始终指向最新的结果。此外，results配置存储了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。

原始信息汇总

数据集概述

数据集创建

• 创建背景：该数据集是在评估模型 janhq/supermario-slerp-v2 在 Open LLM Leaderboard 上的运行过程中自动创建的。
• 数据集组成：包含 63 个配置，每个配置对应一个评估任务。
• 创建次数：数据集从 1 次运行中创建，每个运行在每个配置中作为一个特定的分片存在，分片名称使用运行的时间戳。"train" 分片始终指向最新的结果。

数据集结构

• 额外配置："results" 配置存储了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_janhq__supermario-slerp-v2", "harness_winogrande_5", split="train")

最新结果

• 最新结果时间戳：2023-12-12T07:20:29.210830
• 结果示例： python { "all": { "acc": 0.6523192880412831, "acc_stderr": 0.031929152104876686, "acc_norm": 0.6535004820907845, "acc_norm_stderr": 0.03257318036897926, "mc1": 0.47613219094247244, "mc1_stderr": 0.017483547156961574, "mc2": 0.6295900737174474, "mc2_stderr": 0.015194573521166509 }, "harness|arc:challenge|25": { "acc": 0.6680887372013652, "acc_stderr": 0.01376098820088054, "acc_norm": 0.6936860068259386, "acc_norm_stderr": 0.013470584417276513 }, "harness|hellaswag|10": { "acc": 0.6837283409679347, "acc_stderr": 0.004640699483543311, "acc_norm": 0.8659629555865366, "acc_norm_stderr": 0.003399958334372066 }, ... }

配置详情

• 配置名称：harness_arc_challenge_25

  • 数据文件：
    • 分片：2023_12_12T07_20_29.210830
      • 路径：**/details_harness|arc:challenge|25_2023-12-12T07-20-29.210830.parquet
    • 分片：latest
      • 路径：**/details_harness|arc:challenge|25_2023-12-12T07-20-29.210830.parquet

• 配置名称：harness_gsm8k_5

  • 数据文件：
    • 分片：2023_12_12T07_20_29.210830
      • 路径：**/details_harness|gsm8k|5_2023-12-12T07-20-29.210830.parquet
    • 分片：latest
      • 路径：**/details_harness|gsm8k|5_2023-12-12T07-20-29.210830.parquet

• 配置名称：harness_hellaswag_10

  • 数据文件：
    • 分片：2023_12_12T07_20_29.210830
      • 路径：**/details_harness|hellaswag|10_2023-12-12T07-20-29.210830.parquet
    • 分片：latest
      • 路径：**/details_harness|hellaswag|10_2023-12-12T07-20-29.210830.parquet

• 配置名称：harness_hendrycksTest_5

  • 数据文件：
    • 分片：2023_12_12T07_20_29.210830
      • 路径：
        • **/details_harness|hendrycksTest-abstract_algebra|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-anatomy|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-astronomy|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-business_ethics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-clinical_knowledge|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_biology|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_chemistry|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_computer_science|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_mathematics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_medicine|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-college_physics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-computer_security|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-conceptual_physics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-econometrics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-electrical_engineering|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-elementary_mathematics|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-formal_logic|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-global_facts|5_2023-12-12T07-20-29.210830.parquet
        • **/details_harness|hendrycksTest-high_school_biology|5_2023-12-12T07-20-29.210830.parquet
        • ...