open-llm-leaderboard/details_heegyu__LIMA2-13b-hf

该数据集是在模型 heegyu/LIMA2-13b-hf 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，每个配置对应一个评估任务。数据集是从 2 次运行中生成的，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。一个名为 results 的额外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。

This dataset was automatically generated during the evaluation run of the model heegyu/LIMA2-13b-hf on the Open LLM Leaderboard. The dataset comprises 64 configurations, each corresponding to a single evaluation task. It is compiled from two evaluation runs, where each run is represented as a dedicated split under each configuration, with the split names using the timestamp of the corresponding run as their identifiers. The "train" split always points to the most recent evaluation results. An additional configuration named "results" stores the aggregated results across all runs, which are utilized to compute and display the aggregate metrics on the Open LLM Leaderboard. The README also provides an example of how to load detailed run information using the `datasets` library in Python.

原始信息汇总

数据集概述

数据集来源

该数据集是在对模型 heegyu/LIMA2-13b-hf 进行评估运行期间自动创建的，评估结果发布在 Open LLM Leaderboard 上。

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。
• 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

python from datasets import load_dataset data = load_dataset("open-llm-leaderboard/details_heegyu__LIMA2-13b-hf", "harness_winogrande_5", split="train")

最新结果

以下是 2023-10-22T00:28:18.061876 运行 的最新结果： python { "all": { "em": 0.2590184563758389, "em_stderr": 0.004486510640529356, "f1": 0.3212950922818803, "f1_stderr": 0.004447928613953936, "acc": 0.3950291202646285, "acc_stderr": 0.009430155888357935 }, "harness|drop|3": { "em": 0.2590184563758389, "em_stderr": 0.004486510640529356, "f1": 0.3212950922818803, "f1_stderr": 0.004447928613953936 }, "harness|gsm8k|5": { "acc": 0.0576194086429113, "acc_stderr": 0.006418593319822863 }, "harness|winogrande|5": { "acc": 0.7324388318863457, "acc_stderr": 0.012441718456893009 } }

配置详情

• harness_arc_challenge_25

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|arc:challenge|25_2023-08-09T15:19:08.555277.parquet

• harness_drop_3

  • 分片: 2023_10_22T00_28_18.061876, latest
  • 路径: **/details_harness|drop|3_2023-10-22T00-28-18.061876.parquet

• harness_gsm8k_5

  • 分片: 2023_10_22T00_28_18.061876, latest
  • 路径: **/details_harness|gsm8k|5_2023-10-22T00-28-18.061876.parquet

• harness_hellaswag_10

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hellaswag|10_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: 多个路径，例如 **/details_harness|hendrycksTest-abstract_algebra|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_abstract_algebra_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-abstract_algebra|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_anatomy_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-anatomy|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_astronomy_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-astronomy|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_business_ethics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-business_ethics|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_clinical_knowledge_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-clinical_knowledge|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_biology_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_biology|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_chemistry_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_chemistry|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_computer_science_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_computer_science|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_mathematics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_mathematics|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_medicine_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_medicine|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_college_physics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-college_physics|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_computer_security_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-computer_security|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_conceptual_physics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-conceptual_physics|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_econometrics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-econometrics|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_electrical_engineering_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-electrical_engineering|5_2023-08-09T15:19:08.555277.parquet

• harness_hendrycksTest_elementary_mathematics_5

  • 分片: 2023_08_09T15_19_08.555277, latest
  • 路径: **/details_harness|hendrycksTest-elementary_mathematics|5_2023-08-09T15:19:08.555277.parquet

搜集汇总

背景与挑战

背景概述

该数据集是模型heegyu/LIMA2-13b-hf在Open LLM Leaderboard评估过程中自动生成的，包含64个配置对应不同评估任务，从2次运行中产生，并以时间戳分割存储运行细节。数据集的核心用途是聚合评估结果，通过results配置计算排行榜的聚合指标，便于性能跟踪和比较。