AReno Toolkit for Single-Node LLM RL Training Released

AReno is a local LLM post-training toolkit developed by engineers from Ant Group's ASystem team. It is optimized to deliver maximum performance on a single node, enabling fast post-training without the need to build a separate cluster.

Key features include:

• Plug-and-play: Various post-training methods can be used immediately without setting up a separate cluster or launcher.
• Lightweight: A standalone package that uses only essential libraries such as PyTorch and FlashAttention, without external backends.
• Agentic RL support: Agent functions can be executed via an OpenAI-compatible proxy, and training can be performed based on token and reward data.
• Extensibility: New algorithms, model adapters, reward functions, and more can be easily registered without modifying the core.