---
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/ops/triton_decode_attention.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/cutlass_mla.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/flashattn_mla.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/flashinfer_mla.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/flashmla.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/triton_mla.py
merged: '2026-06-22'
number: 45111
source:
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/ops/triton_decode_attention.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/csrc/libtorch_stable/cache_kernels.cu
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/backends/mla/cutlass_mla.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/backends/mla/flashattn_mla.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/backends/mla/flashinfer_mla.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/backends/mla/flashmla.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/vllm/v1/attention/backends/mla/triton_mla.py
  - https://github.com/vllm-project/vllm/blob/aa4990a9a2024b3f93f1f26f828931f7301daa15/csrc/libtorch_stable/attention/mla/sm100_cutlass_mla_kernel.cu
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#45111 [Attention] Re-enable cross-layer KV cache layout for MLA via stride-aware kernels'
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/45111
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/45111
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/45111.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Re-enable cross-layer KV cache layout for MLA via stride-aware kernels · full dump: [[thoughts/inference/vllm-prs/45111|#45111]]

