An Efficient and User-Friendly Scaling Library for Reinforcement Learning with Large Language Models

[EMNLP'25] s3 - ⚡ Efficient & Effective Search Agent Training via RL for RAG (Verifier-Powered RLVR for Search with Minimal Data)

An Awesome List of Agentic Model trained with Reinforcement Learning

Official repository for "RLVR-World: Training World Models with Reinforcement Learning" (NeurIPS 2025), https://arxiv.org/abs/2505.13934

An official implementation of "CapRL: Stimulating Dense Image Caption Capabilities via Reinforcement Learning"

A curated list of awesome resources about reward construction for AI agents. This repository covers cutting-edge research, and practical guides on defining and collecting rewards to build more intelligent and aligned AI agents.

a Python library that uses Reinforcement Learning (RL) to train LLMs.

A RL env with procedurally generated symbolic reasoning data

🐝 SwarmBench: Benchmarking LLMs' Swarm Intelligence

grpo to train long form QA and instructions with long-form reward model

Enemies for your LLM

MOTIF: Modular Thinking via Reinforcement Fine-tuning in LLMs

A curated collection of research papers on LLM Tool-Integrated Reasoning (TIR), where LLMs enhance reasoning by interacting with external tools such as calculators, search engines, and code interpreters.

CAP RLVR: Reinforcement Learning from Human Feedback for Legal Reasoning using Caselaw Access Project data. Complete GRPO training pipeline with OpenAI Gym environments, deterministic reward functions, and multi-stage curriculum learning for legal LLM development.

Finetuning of Qwen3 0.6B for MCQA tasks

RLHF and Verifiable Reward Models - Post training Research

A composable component orchestrator for Reinforcement Learning from Verifiable Rewards (RLVR) training of Large Language Models on reasoning tasks.

Comprehensive benchmarking framework for RLVR/RLHF libraries on GSM8K mathematical reasoning dataset

A curated list of papers on implicit-reward reinforcement learning for LLMs — no human feedback, no gold answers, no verifiable rewards.