open-llm-leaderboard-old/details_dhmeltzer__Llama-2-13b-hf-ds_eli5_1024_r_64_alpha_16_merged
收藏数据集概述
数据集创建背景
该数据集是在对模型 dhmeltzer/Llama-2-13b-hf-ds_eli5_1024_r_64_alpha_16_merged 进行评估运行期间自动创建的。
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2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" 存储了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_dhmeltzer__Llama-2-13b-hf-ds_eli5_1024_r_64_alpha_16_merged", "harness_winogrande_5", split="train")
最新结果
以下是 最新结果 来自运行 2023-10-28T17:04:10.334813:
python
{
"all": {
"em": 0.022651006711409395,
"em_stderr": 0.001523730780343799,
"f1": 0.09599203020134199,
"f1_stderr": 0.0020758446587035915,
"acc": 0.42256576668005047,
"acc_stderr": 0.00972711232409914
},
"harness|drop|3": {
"em": 0.022651006711409395,
"em_stderr": 0.001523730780343799,
"f1": 0.09599203020134199,
"f1_stderr": 0.0020758446587035915
},
"harness|gsm8k|5": {
"acc": 0.08112206216830932,
"acc_stderr": 0.007520395797922653
},
"harness|winogrande|5": {
"acc": 0.7640094711917916,
"acc_stderr": 0.011933828850275625
}
}
数据集配置
配置列表
-
harness_arc_challenge_25
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|arc:challenge|25_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_drop_3
- 分割:
2023_10_28T17_04_10.334813,latest - 路径:
**/details_harness|drop|3_2023-10-28T17-04-10.334813.parquet
- 分割:
-
harness_gsm8k_5
- 分割:
2023_10_28T17_04_10.334813,latest - 路径:
**/details_harness|gsm8k|5_2023-10-28T17-04-10.334813.parquet
- 分割:
-
harness_hellaswag_10
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hellaswag|10_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径: 多个路径,例如
**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_abstract_algebra_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-abstract_algebra|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_anatomy_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-anatomy|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_astronomy_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-astronomy|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_business_ethics_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-business_ethics|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_clinical_knowledge_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-clinical_knowledge|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_biology_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_biology|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_chemistry_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_chemistry|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_computer_science_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_computer_science|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_mathematics_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_mathematics|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_medicine_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_medicine|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_college_physics_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-college_physics|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_computer_security_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-computer_security|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_conceptual_physics_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-conceptual_physics|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_econometrics_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-econometrics|5_2023-10-03T17-10-34.313268.parquet
- 分割:
-
harness_hendrycksTest_electrical_engineering_5
- 分割:
2023_10_03T17_10_34.313268,latest - 路径:
**/details_harness|hendrycksTest-electrical_engineering|5_2023-10-03T17-10-34.313268.parquet
- 分割:
以上是数据集的详细概述,包括数据集的创建背景、组成、额外配置、数据加载示例、最新结果以及数据集配置列表。



