five

open-llm-leaderboard/details_CHIH-HUNG__llama-2-13b-FINETUNE5_4w-r16-gate_up_down

收藏
Hugging Face2023-10-29 更新2024-03-04 收录
下载链接:
https://hf-mirror.com/datasets/open-llm-leaderboard/details_CHIH-HUNG__llama-2-13b-FINETUNE5_4w-r16-gate_up_down
下载链接
链接失效反馈
官方服务:
资源简介:
该数据集是在模型CHIH-HUNG/llama-2-13b-FINETUNE5_4w-r16-gate_up_down在Open LLM Leaderboard上的评估运行期间自动创建的。数据集由64个配置组成,每个配置对应一个被评估的任务。数据集是从2次运行中创建的,每次运行在每个配置中表示为特定的分割,分割名称使用运行的时间戳。train分割始终指向最新的结果。此外,还有一个名为results的配置,存储了所有运行的聚合结果,这些结果用于计算和显示Open LLM Leaderboard上的聚合指标。README还提供了一个如何使用Python中的datasets库加载运行细节的示例。
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 CHIH-HUNG/llama-2-13b-FINETUNE5_4w-r16-gate_up_downOpen LLM Leaderboard 上的自动创建的。

数据集结构

  • 配置数量:64个配置,每个配置对应一个评估任务。
  • 数据来源:数据集从2次运行中创建,每次运行可以在每个配置中作为一个特定的分片找到,分片名称使用运行的时间戳。
  • 最新结果:"train" 分片始终指向最新的结果。
  • 汇总结果:一个额外的配置 "results" 存储所有运行的汇总结果,用于计算和显示在 Open LLM Leaderboard 上的聚合指标。

数据加载示例

python from datasets import load_dataset data = load_dataset("open-llm-leaderboard/details_CHIH-HUNG__llama-2-13b-FINETUNE5_4w-r16-gate_up_down", "harness_winogrande_5", split="train")

最新结果

以下是 2023-10-29T07:06:26.845938 运行的最新结果: python { "all": { "em": 0.20836828859060402, "em_stderr": 0.004159269440162747, "f1": 0.2507906879194633, "f1_stderr": 0.004162090421371717, "acc": 0.43807672814244847, "acc_stderr": 0.01035305451841861 }, "harness|drop|3": { "em": 0.20836828859060402, "em_stderr": 0.004159269440162747, "f1": 0.2507906879194633, "f1_stderr": 0.004162090421371717 }, "harness|gsm8k|5": { "acc": 0.11372251705837756, "acc_stderr": 0.008744810131034056 }, "harness|winogrande|5": { "acc": 0.7624309392265194, "acc_stderr": 0.011961298905803166 } }

配置详情

以下是数据集的部分配置详情:

  • harness_arc_challenge_25

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|arc:challenge|25_2023-10-10T10-01-17.783068.parquet
  • harness_drop_3

    • 分片:2023_10_29T07_06_26.845938, latest
    • 路径:**/details_harness|drop|3_2023-10-29T07-06-26.845938.parquet
  • harness_gsm8k_5

    • 分片:2023_10_29T07_06_26.845938, latest
    • 路径:**/details_harness|gsm8k|5_2023-10-29T07-06-26.845938.parquet
  • harness_hellaswag_10

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hellaswag|10_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:多个路径,例如 **/details_harness|hendrycksTest-abstract_algebra|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_abstract_algebra_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_anatomy_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-anatomy|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_astronomy_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-astronomy|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_business_ethics_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-business_ethics|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_clinical_knowledge_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_biology_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_biology|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_chemistry_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_chemistry|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_computer_science_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_computer_science|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_mathematics_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_mathematics|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_medicine_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_medicine|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_college_physics_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-college_physics|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_computer_security_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-computer_security|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_conceptual_physics_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_econometrics_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-econometrics|5_2023-10-10T10-01-17.783068.parquet
  • harness_hendrycksTest_electrical_engineering_5

    • 分片:2023_10_10T10_01_17.783068, latest
    • 路径:**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-10T10-01-17.783068.parquet

以上是数据集的概述和部分配置详情,更多配置请参考原始数据集详情页面。

二维码
社区交流群
二维码
科研交流群
商业服务