open-llm-leaderboard-old/details_lmsys__vicuna-7b-v1.5
收藏数据集概述
数据集名称
- Evaluation run of lmsys/vicuna-7b-v1.5
数据集摘要
- 该数据集是在对模型 lmsys/vicuna-7b-v1.5 进行评估运行期间自动创建的,用于 Open LLM Leaderboard。
数据集组成
- 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_lmsys__vicuna-7b-v1.5", "harness_winogrande_5", split="train")
最新结果
- 这些是最新结果,来自 2023-10-21T21:05:37.153515 的运行: python { "all": { "em": 0.017932046979865772, "em_stderr": 0.0013590184569504276, "f1": 0.08961094798657747, "f1_stderr": 0.002014243406072028, "acc": 0.4016346602057357, "acc_stderr": 0.010076117588605417 }, "harness|drop|3": { "em": 0.017932046979865772, "em_stderr": 0.0013590184569504276, "f1": 0.08961094798657747, "f1_stderr": 0.002014243406072028 }, "harness|gsm8k|5": { "acc": 0.08188021228203184, "acc_stderr": 0.007552338527716956 }, "harness|winogrande|5": { "acc": 0.7213891081294396, "acc_stderr": 0.012599896649493878 } }
配置详情
-
harness_arc_challenge_25
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|arc:challenge|25_2023-08-17T12:09:52.202468.parquet
-
harness_drop_3
- 分割:2023_10_19T12_56_49.814418, 2023_10_21T21_05_37.153515, latest
- 路径:
**/details_harness|drop|3_2023-10-19T12-56-49.814418.parquet,**/details_harness|drop|3_2023-10-21T21-05-37.153515.parquet
-
harness_gsm8k_5
- 分割:2023_10_19T12_56_49.814418, 2023_10_21T21_05_37.153515, latest
- 路径:
**/details_harness|gsm8k|5_2023-10-19T12-56-49.814418.parquet,**/details_harness|gsm8k|5_2023-10-21T21-05-37.153515.parquet
-
harness_hellaswag_10
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hellaswag|10_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:多个路径,详见原文
-
harness_hendrycksTest_abstract_algebra_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-abstract_algebra|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_anatomy_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-anatomy|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_astronomy_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-astronomy|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_business_ethics_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-business_ethics|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_clinical_knowledge_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-clinical_knowledge|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_biology_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_biology|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_chemistry_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_chemistry|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_computer_science_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_computer_science|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_mathematics_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_mathematics|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_medicine_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_medicine|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_college_physics_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-college_physics|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_computer_security_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-computer_security|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_conceptual_physics_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-conceptual_physics|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_econometrics_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-econometrics|5_2023-08-17T12:09:52.202468.parquet
-
harness_hendrycksTest_electrical_engineering_5
- 分割:2023_08_17T12_09_52.202468, latest
- 路径:
**/details_harness|hendrycksTest-electrical_engineering|5_2023-08-17T12:09:52.202468.parquet
以上是数据集的详细概述,包括数据集名称、摘要、组成、加载示例、最新结果以及各个配置的详细信息。



