/Tech32d ago

Hugging Face's Lewis Tunstall open-sources Repo2RLEnv to automatically convert GitHub issues and pull requests into RL environments

It generates simulated software engineering environments from any GitHub repository

6116136313.4K

#1136

Original post

Lewis Tunstall@_lewtun#1719inTech

Today's best coding models from Qwen, DeepSeek, Minimax etc are trained on 1000s of concurrent RL environments to simulate SWE tasks in real git repos.

Very excited to open source a tool that unlocks this capability for the whole AI community: point Repo2RLEnv at any GitHub repo to automatically convert issues and PRs into verifiable tasks for RL.

Adithya S K@adithya_s_k

Introducing Repo2RLEnv

Turn any repository into runnable, verifiable coding environments built from real PRs and commits for coding-agent evaluation or RL training

> uv pip install repo2rlenv

6:48 AM · May 28, 2026 · 8.2K Views

Sentiment

Users praised the open-source tool converting GitHub repos into RL environments for coding agents because it makes model progress feel concrete and helps models handle new repo structures via open-sourcing.

Pos

100.0%

Neg

0.0%

2 comments with sentiment.

Cluster Engagement

Digg Deeper

No Digg Deeper questions have been answered for this story yet.

Posts from X

Most Activity

VIEWS4.4KBOOKMARKS10LIKES44RETWEETS4

Anindyadeep@anindyadeeps

The amount of alpha HF folks are pulling off from projects like this, ml-intern, high quality blogs is insane dude.

Adithya S K@adithya_s_k

Introducing Repo2RLEnv

Turn any repository into runnable, verifiable coding environments built from real PRs and commits for coding-agent evaluation or RL training

> uv pip install repo2rlenv

32d4.4K4410

REPLIES1

elie@eliebakouch

@_lewtun do you have ablation to see the quality of the RL env you can extract from it compare to existing pipeline? 👀

Lewis Tunstall@_lewtun

Today's best coding models from Qwen, DeepSeek, Minimax etc are trained on 1000s of concurrent RL environments to simulate SWE tasks in real git repos.

Very excited to open source a tool that unlocks this capability for the whole AI community: point Repo2RLEnv at any GitHub repo to automatically convert issues and PRs into verifiable tasks for RL.

32d85380

Pang Shuo@pangshuo1981

@_lewtun Exactly. This is where the model progress starts to feel concrete. That is exactly where DeepSeek feels more useful: the builder productivity side of DeepSeek. I am building around the same practical AI workflow direction here: https://aevrynai.com/register?invite=uDaiK5WU

32d24

Aaron ⚡️@TheEcomNomad

@_lewtun open sourcing the translation layer is the move. most teams train on proprietary tasks, so models can't handle new repo structures. now we can actually benchmark apples to apples instead of guessing why performance tanks on real codebases.

32d13

Strata@ChainZenit

@_lewtun Not seeing the real repo scale in those test envs yet

32d9