osunlp/AgentCL

AgentCL数据集是一个用于智能体持续学习评估的框架，包含多个子集和构建的数据流。具体包括：agentboard_babyai、agentboard_scienceworld和mmlu_pro，这些是从现有公共数据集中直接提取的子集，未对原始数据进行修改。此外，还提供了构建的codeeval-pro和browsecomp_plus流。CodeEval-Pro部分来源于多个来源，包括配对任务输入（如humaneval_pro.json、mbpp_pro.json、bigcodebench_lite_pro.json）和原始任务测试代码输入（如humaneval-raw-test-json、mbpp-raw-test-json、bigcodebench-raw-test-json）。这些数据集用于支持智能体在持续学习环境下的评估任务。

The AgentCL dataset is an evaluation framework for continual learning in agents, comprising several subsets and constructed data streams. Specifically, it includes agentboard_babyai, agentboard_scienceworld, and mmlu_pro as ready-to-use subsets, which are direct subsets from existing public datasets with no modifications to the original data. Additionally, constructed streams codeeval-pro and browsecomp_plus are provided. The CodeEval-Pro component is built from sources such as paired-task inputs (e.g., humaneval_pro.json, mbpp_pro.json, bigcodebench_lite_pro.json) and raw-task test-code inputs (e.g., humaneval-raw-test-json, mbpp-raw-test-json, bigcodebench-raw-test-json). These datasets are designed to support the evaluation of agents in continual learning scenarios.