SWE-bench task data

Dockerfile generator for the SWE-bench benchmark. This repo is just for the original SWE-bench benchmark and does not contain any data related to the newer benchmarks we released like SWE-bench Multilingual or SWE-bench Multimodal.

Usage

# From HuggingFace dataset
dockerfile-gen

# From local JSON/JSONL file
dockerfile-gen

# Specific instances
dockerfile-gen

Output

Dockerfile and eval.sh are regenerated in place under tasks/<instance_id>/, from that task's task.yaml.

Install

pip install -e .