Skip to content

Dataset · SWE-bench

SWE-smith trajectories for coding agent

SWE-bench/SWE-smith-trajectories

5,017 SWE-agent + Claude 3.7 Sonnet trajectories used for fine-tuning Qwen 2.5 Coder Instruct.

What's inside

English text dataset containing 5,017 trajectories generated by SWE-agent and Claude 3.7 Sonnet on SWE-smith tasks. These data were used for fine-tuning Qwen 2.5 Coder Instruct, producing SWE-agent-LM-32B. The model achieved 40.2% on SWE-bench Verified without verifiers and multiple attempts — one attempt per task. Metadata indicates a size of 10K<n<100K.

What you can do with it

  • Fine-tune a coding agent or LoRA on agent trajectories.
  • Reproduce SWE-agent-LM-32B and its pipeline.
  • Evaluate the quality of SWE-agent trajectories on SWE-smith tasks.
Tasks
Text generation
Languages
English
Size
10K–100K rows
Downloads this month
13.6K
Likes
82

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("SWE-bench/SWE-smith-trajectories")