Shypd.aiThe largest open index of AI toolsrllm is an open-source framework that democratizes Reinforcement Learning for LLMs, allowing users to train AI agents with minimal code changes. It works with any agent framework and offers CLI-first workflows for evaluation and training.
Trending
About
rllm is an open-source framework designed to democratize Reinforcement Learning (RL) for Large Language Models (LLMs), enabling the training of AI agents with minimal code changes. It integrates seamlessly with various agent frameworks like LangGraph, SmolAgent, and OpenAI Agents SDK, requiring only a client swap. The framework features a near-zero code change approach, where users can wrap their agent code with `@rllm.rollout` to automatically trace LLM calls. It supports a CLI-first workflow for evaluating and training agents on over 50 built-in benchmarks, such as `rllm eval gsm8k`. rllm-trained agents have demonstrated impressive performance, outperforming models significantly larger in size. It offers multiple RL algorithms, including GRPO and REINFORCE, and provides two training backends: `verl` for distributed multi-GPU training and `tinker` for single-machine setups, both with the same API.
Capabilities
Pricing & Plans
Open Source
Free
FAQs
Trending
