---
cited_by:
  - '[[thoughts/inference/kv-transfer#2-what-each-layout-costs-the-transfer|reading-map §2]]'
  - '[[thoughts/inference/kv-transfer#5-reading-order|reading-map §5]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/docs/features/nixl_connector_usage.md
  - file:///Users/aarnphm/workspace/vllm/vllm/config/kv_transfer.py
merged: '2025-10-14'
number: 26556
source:
  - https://github.com/vllm-project/vllm/blob/7e6edb14698c1a760272dd44363a288aeb07b571/vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
  - https://github.com/vllm-project/vllm/blob/7e6edb14698c1a760272dd44363a288aeb07b571/docs/features/nixl_connector_usage.md
  - https://github.com/vllm-project/vllm/blob/7e6edb14698c1a760272dd44363a288aeb07b571/vllm/config/kv_transfer.py
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#26556 [NIXL][HeteroTP]Enable KV transfer from HND prefill to NHD decode'
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/26556
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/26556
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/26556.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Current codes only support prefill and decode with same KV cache layout. · full dump: [[thoughts/inference/vllm-prs/26556|#26556]]

