ThinkRL
Open Source RLHF Library

ThinkRL Documentation

A modular, high-performance, reasoning-centric library for Reinforcement Learning from Human and AI Feedback (RLHF & RLAIF).

Get StartedRead the DocsView on GitHub
Made with love from India 🇮🇳 for the world
Get Started
Install ThinkRL and run your first training
pip install -e ".[all]"
Algorithms
VAPO, DAPO, GRPO, COPO, PAPO, PPO, REINFORCE

State-of-the-art algorithms for RLHF training

Training Guides
SFT, CoT, DPO, PRM, and multimodal training

Step-by-step guides for every training scenario

Features
Reasoning, optimization, and integrations

PRM, STaR, LoRA, vLLM, and more

Advanced
Custom rewards, LoRA merging, vLLM setup

Deep dives into advanced functionality

Source Code
Explore the codebase on GitHub

Contribute, report issues, or star the repo

Quick Install

Get started with ThinkRL in seconds

# Clone and setup
git clone https://github.com/ellanorai/ThinkRL.git
cd ThinkRL
# Create venv and install
python -m venv .venv
source .venv/bin/activate
pip install -e .
# Or use the make shortcut
make install
# Or with all extras
pip install -e ".[all]"