five

open-llm-leaderboard/details_codellama__CodeLlama-7b-Instruct-hf

收藏
Hugging Face2023-10-14 更新2024-03-04 收录
下载链接:
https://hf-mirror.com/datasets/open-llm-leaderboard/details_codellama__CodeLlama-7b-Instruct-hf
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集是在Open LLM Leaderboard上对模型codellama/CodeLlama-7b-Instruct-hf进行评估时自动创建的。数据集由64个配置组成,每个配置对应一个评估任务。数据集由3次运行生成,每次运行的结果作为特定的分割存储在配置中,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,还有一个名为results的配置,存储了所有运行的聚合结果,并用于在Open LLM Leaderboard上计算和显示聚合指标。

该数据集是在Open LLM Leaderboard上对模型codellama/CodeLlama-7b-Instruct-hf进行评估时自动创建的。数据集由64个配置组成,每个配置对应一个评估任务。数据集由3次运行生成,每次运行的结果作为特定的分割存储在配置中,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,还有一个名为results的配置,存储了所有运行的聚合结果,并用于在Open LLM Leaderboard上计算和显示聚合指标。
提供机构:
open-llm-leaderboard
原始信息汇总

数据集概述

数据集来源

该数据集是在评估模型 codellama/CodeLlama-7b-Instruct-hfOpen LLM Leaderboard 上的自动创建的。

数据集组成

数据集由64个配置组成,每个配置对应一个评估任务。数据集从3次运行中创建,每次运行可以在每个配置中找到特定的分割,分割名称使用运行的时间戳。"train" 分割始终指向最新的结果。

额外配置

一个额外的配置 "results" 存储所有运行的聚合结果,用于计算和显示在 Open LLM Leaderboard 上的聚合指标。

数据加载示例

以下是加载特定运行详细信息的示例代码: python from datasets import load_dataset data = load_dataset("open-llm-leaderboard/details_codellama__CodeLlama-7b-Instruct-hf", "harness_winogrande_5", split="train")

最新结果

以下是来自2023-10-14T16:13:27.845445运行的最新结果(注意可能包含其他任务的结果,每个任务的结果可以在 "results" 和 "latest" 分割中找到): python { "all": { "em": 0.0008389261744966443, "em_stderr": 0.00029649629898012493, "f1": 0.05166841442953039, "f1_stderr": 0.0012678878311342997, "acc": 0.36261266786861684, "acc_stderr": 0.010449619353516184 }, "harness|drop|3": { "em": 0.0008389261744966443, "em_stderr": 0.00029649629898012493, "f1": 0.05166841442953039, "f1_stderr": 0.0012678878311342997 }, "harness|gsm8k|5": { "acc": 0.07960576194086429, "acc_stderr": 0.00745592433867628 }, "harness|winogrande|5": { "acc": 0.6456195737963694, "acc_stderr": 0.013443314368356088 } }

配置详情

以下是数据集的配置详情:

  • config_name: harness_arc_challenge_25

    • split: 2023_08_25T17_04_00.078187
      • path: **/details_harness|arc:challenge|25_2023-08-25T17:04:00.078187.parquet
    • split: 2023_08_26T03_58_42.829453
      • path: **/details_harness|arc:challenge|25_2023-08-26T03:58:42.829453.parquet
    • split: latest
      • path: **/details_harness|arc:challenge|25_2023-08-26T03:58:42.829453.parquet
  • config_name: harness_drop_3

    • split: 2023_10_14T16_13_27.845445
      • path: **/details_harness|drop|3_2023-10-14T16-13-27.845445.parquet
    • split: latest
      • path: **/details_harness|drop|3_2023-10-14T16-13-27.845445.parquet
  • config_name: harness_gsm8k_5

    • split: 2023_10_14T16_13_27.845445
      • path: **/details_harness|gsm8k|5_2023-10-14T16-13-27.845445.parquet
    • split: latest
      • path: **/details_harness|gsm8k|5_2023-10-14T16-13-27.845445.parquet
  • config_name: harness_hellaswag_10

    • split: 2023_08_25T17_04_00.078187
      • path: **/details_harness|hellaswag|10_2023-08-25T17:04:00.078187.parquet
    • split: 2023_08_26T03_58_42.829453
      • path: **/details_harness|hellaswag|10_2023-08-26T03:58:42.829453.parquet
    • split: latest
      • path: **/details_harness|hellaswag|10_2023-08-26T03:58:42.829453.parquet
  • config_name: harness_hendrycksTest_5

    • split: 2023_08_25T17_04_00.078187
      • path: **/details_harness|hendrycksTest-abstract_algebra|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-anatomy|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-astronomy|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-business_ethics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-clinical_knowledge|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_biology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_chemistry|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_computer_science|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_mathematics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_medicine|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-college_physics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-computer_security|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-conceptual_physics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-econometrics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-electrical_engineering|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-elementary_mathematics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-formal_logic|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-global_facts|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_biology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_chemistry|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_computer_science|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_european_history|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_geography|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_mathematics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_microeconomics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_physics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_psychology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_statistics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_us_history|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-high_school_world_history|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-human_aging|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-human_sexuality|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-international_law|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-jurisprudence|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-logical_fallacies|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-machine_learning|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-management|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-marketing|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-medical_genetics|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-miscellaneous|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-moral_disputes|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-moral_scenarios|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-nutrition|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-philosophy|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-prehistory|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-professional_accounting|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-professional_law|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-professional_medicine|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-professional_psychology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-public_relations|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-security_studies|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-sociology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-us_foreign_policy|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-virology|5_2023-08-25T17:04:00.078187.parquet
      • path: **/details_harness|hendrycksTest-world_religions|5_2023-08-25T17:04:00.078187.parquet
    • split: 2023_08_26T03_58_42.829453
      • path: **/details_harness|hendrycksTest-abstract_algebra|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-anatomy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-astronomy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-business_ethics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-clinical_knowledge|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_biology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_chemistry|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_computer_science|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_medicine|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-computer_security|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-conceptual_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-econometrics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-electrical_engineering|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-elementary_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-formal_logic|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-global_facts|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_biology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_chemistry|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_computer_science|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_european_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_geography|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_microeconomics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_psychology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_statistics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_us_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_world_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-human_aging|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-human_sexuality|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-international_law|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-jurisprudence|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-logical_fallacies|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-machine_learning|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-management|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-marketing|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-medical_genetics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-miscellaneous|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-moral_disputes|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-moral_scenarios|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-nutrition|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-philosophy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-prehistory|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_accounting|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_law|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_medicine|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_psychology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-public_relations|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-security_studies|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-sociology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-us_foreign_policy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-virology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-world_religions|5_2023-08-26T03:58:42.829453.parquet
    • split: latest
      • path: **/details_harness|hendrycksTest-abstract_algebra|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-anatomy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-astronomy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-business_ethics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-clinical_knowledge|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_biology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_chemistry|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_computer_science|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_medicine|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-college_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-computer_security|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-conceptual_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-econometrics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-electrical_engineering|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-elementary_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-formal_logic|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-global_facts|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_biology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_chemistry|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_computer_science|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_european_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_geography|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_government_and_politics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_macroeconomics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_mathematics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_microeconomics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_physics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_psychology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_statistics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_us_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-high_school_world_history|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-human_aging|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-human_sexuality|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-international_law|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-jurisprudence|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-logical_fallacies|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-machine_learning|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-management|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-marketing|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-medical_genetics|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-miscellaneous|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-moral_disputes|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-moral_scenarios|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-nutrition|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-philosophy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-prehistory|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_accounting|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_law|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_medicine|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-professional_psychology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-public_relations|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-security_studies|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-sociology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-us_foreign_policy|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-virology|5_2023-08-26T03:58:42.829453.parquet
      • path: **/details_harness|hendrycksTest-world_religions|5_2023-08-26T03:58:42.829453.parquet
搜集汇总
数据集介绍
main_image_url
构建方式
在大型语言模型评估领域,模型性能的量化分析依赖于系统化的评测框架。该数据集是Open LLM Leaderboard在对CodeLlama-7B-Instruct-hf模型进行自动化评估过程中自动生成的产物,其构建逻辑紧密围绕评测任务展开。数据集包含64个配置,每个配置对应一个独立的评估任务,如ARC挑战、DROP、GSM8K等。数据源自三次独立的评估运行,每次运行的结果以时间戳为标识,作为特定分割(split)存储于各配置中,而'train'分割则始终指向最新一次运行的结果。此外,一个名为'results'的独立配置专门存储所有运行的聚合指标,用于在排行榜上展示模型的综合表现。
特点
该数据集的核心特点在于其结构化的多任务评估体系与版本追溯能力。它涵盖了从常识推理(如HellaSwag)到数学求解(如GSM8K)等广泛的语言理解与生成任务,每个任务均设有独立的配置,便于研究者聚焦分析模型在特定领域的能力。数据集的每次运行结果均以时间戳分割保存,支持对模型性能随时间演变的纵向追踪,而'train'分割自动指向最新结果,确保了数据引用的时效性。'results'配置整合了所有任务的聚合指标,包括准确率、F1分数等统计量及其标准误差,为模型间的横向对比提供了标准化依据。
使用方法
研究者可通过Hugging Face的datasets库便捷地加载该数据集,以获取模型在特定任务上的详细评估结果。例如,使用load_dataset函数指定数据集名称及目标配置(如'harness_winogrande_5'),并通过split参数选择'train'以获取最新运行数据,或选择具体时间戳分割以回溯历史版本。对于需要整体性能概览的场景,可直接加载'results'配置,其中包含所有任务的聚合指标。此外,数据集以Parquet格式存储,支持高效的数据读取与过滤操作,便于进行深入的性能分析或可视化展示。
背景与挑战
背景概述
随着大语言模型在自然语言处理领域的飞速发展,如何公正、全面地评估模型性能成为学界与工业界共同关注的核心议题。Open LLM Leaderboard由HuggingFace团队于2023年创建,旨在为开源社区提供标准化的模型评估基准。该数据集记录了CodeLlama-7B-Instruct-hf模型在多个任务上的详细评测结果,涵盖ARC挑战、DROP、GSM8K、HellaSwag及MMLU等广泛认可的基准测试。通过多次运行与细粒度的结果存储,该数据集不仅揭示了模型在常识推理、数学问题求解与知识理解等方面的能力边界,还为研究者提供了可复现的评估流程,推动了开源大模型评测体系的透明化与规范化。
当前挑战
该数据集所面临的挑战体现在两个层面。在领域问题层面,大语言模型的评测需覆盖多样化的认知能力,然而现有任务如DROP的精确匹配准确率极低(0.08%),揭示了模型在复杂推理与数值理解上的显著短板。构建过程中,评测结果需跨多次运行进行整合与版本管理,不同时间戳下的数据分裂可能导致结果一致性难以保证。此外,任务配置的多样性(如64种配置)要求数据加载与索引机制具备高灵活性,而存储格式(如parquet)的兼容性与大规模数据的快速检索亦构成技术难点,需持续优化以支持高效分析与比较。
常用场景
经典使用场景
在大型语言模型评估领域,Open LLM Leaderboard 的评测数据集已成为衡量模型综合能力的黄金标准。该数据集专为 CodeLlama-7B-Instruct-hf 模型的系统化评估而设计,涵盖 ARC、HellaSwag、MMLU、GSM8K 等多项经典基准任务。研究者可借助其分任务配置与多次运行记录,深入剖析模型在常识推理、数学求解、知识理解等维度的表现,从而实现对模型能力的精细化诊断与横向对比。
衍生相关工作
该数据集催生了多项关于模型能力边界与评估方法论的重要研究。例如,研究者基于其细粒度结果分析了指令微调对模型推理能力的提升效果,并衍生出针对特定任务(如代码生成与数学解题)的专项评估基准。同时,评测过程中对随机性与多次运行差异的深入探讨,也推动了关于模型鲁棒性评估标准与统计显著性检验方法的革新。
数据集最近研究
最新研究方向
在大型语言模型评测领域,Open LLM Leaderboard 已成为衡量模型综合能力的重要基准平台。CodeLlama-7b-Instruct-hf 作为面向代码生成与指令遵循的专用模型,其在该排行榜上的评估数据揭示了当前研究的前沿趋势:研究者正聚焦于多任务泛化能力的量化分析,涵盖从常识推理(如 Winogrande)到数学解题(GSM8K)再到文本理解(DROP)的多元维度。该数据集通过标准化评测框架与动态更新的运行记录,为追踪模型在不同时间节点、不同任务上的性能演变提供了可靠依据。其意义在于,不仅推动了代码智能模型的透明化比较,更促进了开源社区对模型短板与改进方向的精准洞察,成为连接模型开发与实证评估的桥梁。
以上内容由遇见数据集搜集并总结生成
二维码
社区交流群
二维码
科研交流群
商业服务