---
title: '57430'
pageLayout: default
slug: thoughts/inference/vllm-prs/57430
permalink: https://aarnphm.xyz/thoughts/inference/vllm-prs/57430.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
# [#57430](https://github.com/aarnphm/aarnphm.github.io/issues/57430) \[Quantization\] Support kimi-k3 routed expert quant

[vllm-project/vllm#57430](https://github.com/vllm-project/vllm/pull/57430) · kylesayrs · MERGED · created 2026-09-17T18:47:57Z · merged 2026-09-18T14:04:00Z

## Background

- Working on eeking out performance on kimi-k3, potentially by quantizing these moe linear adapter layers

## Testing

- Tested with `RedHatAI/Kimi-K3-NVFP4-REAP25`

---

## comments

**mgoin** (2026-09-17T21:47:09Z):

/ci run

**github-actions** (2026-09-17T21:47:27Z):

❌ This PR is 41 commits behind upstream `main`. Your branch must contain every commit currently on upstream `main`. No new CI build was started. Merge or rebase onto the latest `main`, then rerun `/ci run`. To test this branch at your own risk, use `/ci run --allow-stale`.

<!-- vllm-ci-command:5721677422 -->

**mgoin** (2026-09-17T23:30:23Z):

/ci run

**github-actions** (2026-09-17T23:30:41Z):

✅ Triggered [Buildkite CI #89739](https://buildkite.com/vllm/ci/builds/89739) for commit `9db2cca35418`.

