---
cited_by:
  - '[[thoughts/inference/kv-layout#3-what-each-layout-costs-the-kernels|reading-map §3]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/flashinfer_mla_sparse.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/sparse_utils.py
merged: '2026-08-31'
number: 54465
source:
  - https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/v1/attention/backends/mla/flashinfer_mla_sparse.py
  - https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/models/deepseek_v32/common/kernels.py
  - https://github.com/vllm-project/vllm/blob/570735520942b6428ae98f1eb442c6d88adae55e/vllm/v1/attention/backends/mla/sparse_utils.py
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#54465 [Bugfix][MLA] Fix BLHNC addressing for FlashInfer sparse MLA'
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/54465
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/54465
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/54465.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Fix sparse MLA cache addressing when `FLASHINFER_MLA_SPARSE` uses a block-outermost KV-cache layout such as BLHNC. · full dump: [[thoughts/inference/vllm-prs/54465|#54465]]

