# vLLM adds day-zero reinforcement learning support with Megatron

> Original title: Day-0 RL support, too: vLLM + Megatron.

- Source: vLLM
- Published: 2026-11-04T16:40:00.000Z
- Bad Influence: https://badinfluence.news/items/v1c0ssl2xs63ozsl7jynbbr4g
- Original: https://x.com/vllm_project/status/2118020448554463488

## Summary

vLLM says it now supports reinforcement learning from day zero with Megatron, covering training–inference consistency for Kimi Delta Attention. The work also includes cache and state management for serving long-running agents, according to the post by @inferact.

## Post

Day-0 RL support, too: vLLM + Megatron.

@inferact’s work covers training–inference consistency for Kimi Delta Attention, alongside the cache and state management needed to serve long-running agents.

From serving the model to continuing its training, the open stack is ready for the next step.

![](https://badinfluence.news/api/img-proxy?u=https%3A%2F%2Fpbs.twimg.com%2Fmedia%2FHWS4X9ebMAASoQL.jpg&mode=thumb&exp=1794009600&sig=f2dd4864da2ee0d9)