bengaliAI/BFCL
收藏资源简介:
伯克利函数调用排行榜(BFCL)是一个用于评估大型语言模型(LLM)调用函数能力的实时排行榜。该数据集从我们的学习成果中构建而成,代表了大多数用户的功能调用用例,例如在代理、企业工作流程的一部分等。我们的评估数据集涵盖了多个类别,并在多种语言中应用。BFCL V1 主要分为 Python 和 Non-Python 两个类别,包含 Simple、Multiple Function、Parallel Function、Parallel Multiple Function、Chatting Capability、Function Relevance Detection、REST API、SQL、Java 和 JavaScript 等子类别。BFCL V2 Live 使用真实世界数据,重点关注多个函数场景和相关性/不相关性检测。BFCL V3 引入了多步骤和多轮函数调用场景。数据集以多个 JSON 文件的形式组织,可以使用提供的辅助函数加载。数据集在 Apache 2.0 许可下发布,并欢迎贡献。
The Berkeley Function Calling Leaderboard (BFCL) is a live leaderboard to evaluate the ability of different LLMs to call functions. The dataset includes multiple releases, each with different categories and compositions. BFCL V1 includes Python and Non-Python categories, with subcategories such as Simple, Multiple Function, Parallel Function, Parallel Multiple Function, Chatting Capability, Function Relevance Detection, REST API, SQL, Java, and JavaScript. BFCL V2 Live uses real-world data and focuses on multiple function scenarios and relevance/irrelevance detection. BFCL V3 introduces multi-step and multi-turn function calling scenarios. The dataset is organized in multiple JSON files and can be loaded using the provided helper functions. The dataset is released under Apache 2.0 license and is open for contributions.




