---
cited_by:
  - '[[thoughts/inference/kimi-k3-state#34-pd-dcp-kv-offload-59229-test-matrix|kimik3 §3.4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/kv_cache_utils.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/kv_cache_interface.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/attention/mla_attention.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/models/deepseek_v2.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/sparse_swa.py
merged: '2026-09-30'
number: 57652
source:
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/v1/core/kv_cache_utils.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/distributed/kv_transfer/kv_connector/v1/offloading/config.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/v1/kv_cache_interface.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/models/deepseek_v4/attention.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/model_executor/layers/attention/mla_attention.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/model_executor/models/deepseek_v2.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/models/deepseek_v4/compressor.py
  - https://github.com/vllm-project/vllm/blob/5463fe4962785cdc3383477bf3af6533e7647dfd/vllm/v1/attention/backends/mla/sparse_swa.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#57652 [KV-Offloading][TP] : Expand replicated_layout detection to multi-group MLA '
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/57652
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/57652
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/57652.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
KV offloading for MLA models + TP need only offload KV values from one TP rank. · full dump: [[thoughts/inference/vllm-prs/57652|#57652]]

