---
cited_by:
  - '[[thoughts/papers/xie2026mhcmanifoldconstrainedhyperconnections#16-mhc-manifold-constrained-hyper-connections-with-v41s-delayedsingle-pass-pre|dsv41 §1.6]]'
  - '[[thoughts/inference/deepseek-v4.1-serving#4-pr-reading-order|dsv41 §4]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-09-14'
number: 56633
source:
  - https://github.com/vllm-project/vllm/blob/5372e72a9884c141874465f0eb8054a4a81da6bb/vllm/model_executor/kernels/mhc/tilelang.py
  - https://github.com/vllm-project/vllm/blob/5372e72a9884c141874465f0eb8054a4a81da6bb/vllm/models/deepseek_v41/nvidia/model.py
  - https://github.com/vllm-project/vllm/blob/5372e72a9884c141874465f0eb8054a4a81da6bb/vllm/model_executor/kernels/mhc/tilelang_kernels.py
  - https://github.com/vllm-project/vllm/blob/5372e72a9884c141874465f0eb8054a4a81da6bb/vllm/model_executor/kernels/mhc/warmup.py
  - https://github.com/vllm-project/vllm/blob/5372e72a9884c141874465f0eb8054a4a81da6bb/vllm/models/deepseek_v41/nvidia/dspark.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#56633 [Perf][DSv4.1] Fold the mHC post block into the delayed pre projection'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/56633
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/56633
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/56633.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
DeepSeek-V4.1 runs an mHC post block and then the next sublayer’s delayed pre block at every seam. Each seam therefore pays for three things: a post kernel, a split-k projection that re-reads the residual streams the post kernel just… · full dump: [[thoughts/inference/vllm-prs/56633|#56633]]

