open-llm-leaderboard-old/details_wei123602__llama2-13b-FINETUNE3_TEST2
收藏数据集概述
数据集创建
- 数据集是在模型 wei123602/llama2-13b-FINETUNE3_TEST2 在 Open LLM Leaderboard 上的评估运行期间自动创建的。
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 3 次运行中创建,每次运行可以在每个配置中找到一个特定的分割,分割名称使用运行的时间戳。
- "train" 分割总是指向最新的结果。
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 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_wei123602__llama2-13b-FINETUNE3_TEST2", "harness_winogrande_5", split="train")
最新结果
- 这些是最新的结果,来自 2023-10-28T09:53:17.709619 的运行: python { "all": { "em": 0.2633179530201342, "em_stderr": 0.004510450588757746, "f1": 0.3047556627516783, "f1_stderr": 0.004459334625484884, "acc": 0.4441419290522286, "acc_stderr": 0.010548755752104734 }, "harness|drop|3": { "em": 0.2633179530201342, "em_stderr": 0.004510450588757746, "f1": 0.3047556627516783, "f1_stderr": 0.004459334625484884 }, "harness|gsm8k|5": { "acc": 0.12585291887793784, "acc_stderr": 0.009136212598406319 }, "harness|winogrande|5": { "acc": 0.7624309392265194, "acc_stderr": 0.01196129890580315 } }
配置详情
-
harness_arc_challenge_25
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|arc:challenge|25_2023-09-14T13-51-34.438102.parquet
-
harness_drop_3
- 分割: 2023_10_28T06_56_58.916586, 2023_10_28T09_53_17.709619, latest
- 路径:
**/details_harness|drop|3_2023-10-28T06-56-58.916586.parquet,**/details_harness|drop|3_2023-10-28T09-53-17.709619.parquet
-
harness_gsm8k_5
- 分割: 2023_10_28T06_56_58.916586, 2023_10_28T09_53_17.709619, latest
- 路径:
**/details_harness|gsm8k|5_2023-10-28T06-56-58.916586.parquet,**/details_harness|gsm8k|5_2023-10-28T09-53-17.709619.parquet
-
harness_hellaswag_10
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hellaswag|10_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径: 多个路径,详见原文
-
harness_hendrycksTest_abstract_algebra_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-abstract_algebra|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_anatomy_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-anatomy|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_astronomy_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-astronomy|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_business_ethics_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-business_ethics|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_clinical_knowledge_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-clinical_knowledge|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_biology_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_biology|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_chemistry_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_chemistry|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_computer_science_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_computer_science|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_mathematics_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_mathematics|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_medicine_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_medicine|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_college_physics_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-college_physics|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_computer_security_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-computer_security|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_conceptual_physics_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-conceptual_physics|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_econometrics_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-econometrics|5_2023-09-14T13-51-34.438102.parquet
-
harness_hendrycksTest_electrical_engineering_5
- 分割: 2023_09_14T13_51_34.438102, latest
- 路径:
**/details_harness|hendrycksTest-electrical_engineering|5_2023-09-14T13-51-34.438102.parquet
以上是数据集的详细概述,包括数据集的创建背景、结构、加载示例、最新结果以及各个配置的详细信息。



