---
cited_by:
  - '[[thoughts/papers/kimiteam2026kimik3openfrontier#18-quant-formats|kimik3 §1.8]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py
merged: '2026-09-07'
number: 55407
source:
  - https://github.com/vllm-project/vllm/blob/6748217fb949952b4db418d09e64952cd7d8e352/vllm/model_executor/layers/quantization/utils/flashinfer_fp4_moe.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#55407 [Bugfix] Fix Kimi K3 NVFP4 MoE weight conversion OOM'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/55407
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/55407
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/55407.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Currently when running Kimi k3 nvfp4, it throws \~15000 lines of warning like the following due to fragmented memory during quantization post processing. In some cases, it also fatally fails with OOM. · full dump: [[thoughts/inference/vllm-prs/55407|#55407]]

