AI and models · Independent · 22 Oct 2026

verl-project/verl

verl/HybridFlow: A Flexible and Efficient RL Post-Training Framework

← AI and models

In simple English

This library helps researchers train large language models using reinforcement learning. It makes the training process faster and uses fewer computer resources. You can plug it into popular training engines and language models without rewriting your code. It is free to use and open source. You will need powerful graphics cards and technical knowledge to run it.

What it is

Verl is a reinforcement learning framework for large language models that decouples computation and data dependencies. It allows you to run complex post-training algorithms across clusters of graphics processing units.

What it is good for

It is good for AI researchers and engineers who want to train reasoning models or align language models using algorithms like PPO and GRPO. It removes the memory bottlenecks that usually happen when switching between generating text and updating model weights.

Stars24k
LicenceApache-2.0
Latestv0.9.1 (2026-09-20)checked 22 Oct 2026
Last code change2026-10-22
Written inPython
Setup effort
4 / 5
Payoff
5 / 5
Good
  • Integrates cleanly with existing tools like PyTorch FSDP and vLLM.
  • Supports modern reinforcement learning algorithms such as GRPO and PPO.
  • Includes memory-efficient engines that reduce communication overhead across servers.
Watch for
  • Requires multiple high-end graphics processing units with large amounts of memory.
  • Steep learning curve for anyone who is not an experienced machine learning engineer.
  • Setup involves complex distributed computing configurations and dependencies.
Replaces

Writing custom, inefficient training loops for reinforcement learning on large language models.

Works well with

vLLM (a fast and easy-to-use library for serving language models with high throughput)

Similar repositories
  • OpenRLHF/OpenRLHF10k stars

    An open-source reinforcement learning framework for language models that also supports distributed training engines.

  • huggingface/trl20k stars

    A popular library by Hugging Face for training transformer language models with reinforcement learning and supervised fine-tuning.

Also reviewed on Repo Radar
  • ggml-org/llama.cpp132k stars

    A C and C++ library and command-line tool for running language models locally.

  • EleutherAI/lm-evaluation-harness14k stars

    Runs standard question sets against a language model and reports the scores.

  • mudler/LocalAI50k stars

    Serves text, vision, speech, image and video models from one self-hosted service that imitates the paid interfaces.

Install
git clone https://github.com/verl-project/verl.git
cd verl
pip install -e .

Independent: reviewed for Repo Radar outside the daily editions. Every independent repository

On Stack Radar: AI & Automation › AI Models & Inference › Model Fine-Tuning.

Get the next edition in your inbox

A dozen repositories, opened and checked. The licence read, the last release dated, and the ones that did not make it named with the reason. It is the half most lists leave out.

No tracking pixels. One click to leave. The archive stays free either way.

We use your address to send the edition and nothing else. Confirm by email, leave in one click. How we handle it.