---
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/model_executor/layers/attention/mla_attention.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/kv_connector_model_runner_mixin.py
  - file:///Users/aarnphm/workspace/vllm/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/indexer.py
merged: '2026-03-16'
number: 37090
source:
  - https://github.com/vllm-project/vllm/blob/ca1954d58c49e3a3209ec86d743a99f3a605028b/vllm/model_executor/layers/attention/mla_attention.py
  - https://github.com/vllm-project/vllm/blob/ca1954d58c49e3a3209ec86d743a99f3a605028b/vllm/v1/worker/kv_connector_model_runner_mixin.py
  - https://github.com/vllm-project/vllm/blob/ca1954d58c49e3a3209ec86d743a99f3a605028b/vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
  - https://github.com/vllm-project/vllm/blob/ca1954d58c49e3a3209ec86d743a99f3a605028b/vllm/v1/attention/backends/mla/indexer.py
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#37090 [Bugfix] Disable cross-layer KV cache for MLA attention backends'
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/37090
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/37090
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/37090.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Fixes [#37032](https://github.com/aarnphm/aarnphm.github.io/issues/37032) · full dump: [[thoughts/inference/vllm-prs/37090|#37090]]

