Repo2RLEnv PR Diff: GitHub Pull Request Patches for RL Training

Repo2RLEnv PR Diff is a dataset generated by the Repo2RLEnv tool, which turns real GitHub repositories into verifiable reinforcement learning environments. The dataset includes tasks derived from 25 source repositories, such as BurntSushi/ripgrep, axios/axios, and huggingface/datasets. It was created by author AdithyaSK and last updated on 2026-05-26.

Use Cases

• Training RL agents to generate code patches based on task specifications and instructions.
• Benchmarking code generation models on real-world pull request diffs from popular repositories.
• Creating verifiable environments for testing automated program repair systems.

Strengths

• Derived from 25 real-world, popular open-source GitHub repositories.
• Includes task specifications, instructions, oracle patches, and test scripts for each task.
• Generated by a tool (Repo2RLEnv) designed to create verifiable RL environments.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Generated from 25 specified GitHub repositories.

Collection Method

Generated by the Repo2RLEnv tool.

Freshness

Last updated 2026-05-26 18:48:09; freshness should be verified.