---
cited_by:
  - '[[thoughts/inference/deepseek-v4.1-serving#18-moe--gate|dsv41 §1.8]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-09-16'
number: 57204
source:
  - https://github.com/vllm-project/vllm/blob/30b847c1fa5461a590606cbc9b8dae4a490c6cdd/vllm/models/deepseek_v4/nvidia/model.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#57204 [Perf][DSV4.1] Remove MegaMoE padding and shared padding workaround'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/57204
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/57204
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/57204.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Remove unnecessary MegaMoE intermediate-width padding for DeepSeek-V4.1-Flash and revert the production workaround added in [#56568](https://github.com/aarnphm/aarnphm.github.io/issues/56568). Routed and shared experts both retain the checkpoint width of 2304, so native shared-expert fusion remains… · full dump: [[thoughts/inference/vllm-prs/57204|#57204]]

