stalaei/sdft-tooluse-distil

SDFT工具使用（ToolAlpaca）—蒸馏格式数据集是一个用于文本生成任务的英语数据集，涉及自蒸馏、工具使用和函数调用等技术。数据集分为训练集（4046行）和评估集（97行），包含prompt、teacher_prompt和chat_template_kwargs三列。prompt列表示学生接收的聊天格式提示，teacher_prompt列表示教师接收的聊天格式提示，chat_template_kwargs列包含用于tokenizer.apply_chat_template的额外参数。数据集来源于ToolAlpaca和SDFT论文，用于模拟API多步工具使用跟踪。

The SDFT Tool-Use (ToolAlpaca) — distil format dataset is an English dataset for text-generation tasks, involving techniques such as self-distillation, tool-use, and function-calling. The dataset is divided into a training set (4046 rows) and an evaluation set (97 rows), containing columns for prompt, teacher_prompt, and chat_template_kwargs. The prompt column represents the chat-format prompt received by the student, the teacher_prompt column represents the chat-format prompt received by the teacher, and the chat_template_kwargs column contains extra kwargs for tokenizer.apply_chat_template. The dataset originates from ToolAlpaca and the SDFT paper, used for simulating multi-step tool-use traces with APIs.