SWE-smith-trajectories

<div align="center"> <a href="https://swesmith.com"> <img src="https://avatars.githubusercontent.com/u/189315905?s=200&v=4" alt="Logo" width="200"> <h1 align="center">SWE-smith Trajectories</h1> </a> </div> <p align="center"> <a href="https://github.com/SWE-bench/SWE-smith">Code</a> • <a href="https://arxiv.org/abs/2504.21798">Paper</a> • <a href="https://swesmith.com/">Site</a> </p> This dataset contains the 5017 trajectories we fine-tuned Qwen 2.5 Coder Instruct on, leading to [SWE-agent-LM-32B](https://huggingface.co/SWE-bench/SWE-agent-LM-32B), a coding LM agent that achieve 40.2% on SWE-bench Verified (no verifiers or multiple rollouts, just 1 attempt per instance). Trajectories were generated by running SWE-agent + Claude 3.7 Sonnet on task instances from the SWE-smith [dataset](https://huggingface.co/datasets/SWE-bench/SWE-smith).<div align="center"> <a href="https://swesmith.com"> <img src="https://avatars.githubusercontent.com/u/189315905?s=200&v=4" alt="标志" width="200"> <h1 align="center">SWE-smith 轨迹数据集</h1> </a> </div> <p align="center"> <a href="https://github.com/SWE-bench/SWE-smith">代码</a> • <a href="https://arxiv.org/abs/2504.21798">论文</a> • <a href="https://swesmith.com/">项目站点</a> </p> 本数据集包含我们用于微调通义千问2.5代码指令版（Qwen 2.5 Coder Instruct）的5017条执行轨迹，由此得到了代码大语言模型智能体[SWE-agent-LM-32B](https://huggingface.co/SWE-bench/SWE-agent-LM-32B)，该智能体在SWE-bench 验证集（SWE-bench Verified）上取得了40.2%的性能（无需验证器或多次推演，仅对每个任务实例进行单次尝试）。 本数据集的执行轨迹均通过在来自[SWE-smith数据集（SWE-smith dataset）](https://huggingface.co/datasets/SWE-bench/SWE-smith)的任务实例上运行SWE-agent与Claude 3.7 Sonnet生成。

搜集汇总

背景与挑战

背景概述

该数据集包含5017条轨迹，用于微调Qwen 2.5 Coder Instruct模型，生成SWE-agent-LM-32B模型，该模型在SWE-bench Verified上实现了40.2%的准确率。这些轨迹是通过在SWE-smith数据集的任务实例上运行SWE-agent和Claude 3.7 Sonnet生成的。