---
cited_by:
  - '[[thoughts/inference/kv-transfer#5-reading-order|reading-map §5]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-07-29'
number: 49762
source:
  - https://github.com/vllm-project/vllm/blob/f37f03db4af69e4969242767734db3d9175055f7/vllm/distributed/kv_transfer/kv_connector/v1/nixl/push_worker.py
  - https://github.com/vllm-project/vllm/blob/f37f03db4af69e4969242767734db3d9175055f7/vllm/distributed/kv_transfer/kv_connector/v1/nixl/base_worker.py
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#49762 [KV Connector] Support NIXL P/D for hybrid MLA+SSM models '
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/49762
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/49762
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/49762.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
`register_kv_caches` marked a shared region’s FA view by whichever layer registered it first; a KDA-first ordering left dual-purpose regions flagged as head-sharded. Merge the MLA flag on the HMA dedup path so split handles replicate their… · full dump: [[thoughts/inference/vllm-prs/49762|#49762]]

