---
cited_by:
  - '[[thoughts/inference/kv-transfer#5-reading-order|reading-map §5]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/kv_cache_utils.py
merged: '2026-06-07'
number: 44454
source:
  - https://github.com/vllm-project/vllm/blob/4dcd10eb0d223a3ec4b2c96deaf3a48a96c8dcaa/vllm/v1/core/kv_cache_utils.py
state: merged
tags:
  - vllm
  - vllm-pr
  - hetero-tp-kv-layout
title: '#44454 [1/N][KV-Cache Layout Refactor] Refactor DSV4 KV cache config construction'
topic: hetero-tp-kv-layout
url: https://github.com/vllm-project/vllm/pull/44454
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/44454
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/44454.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Extracts the per-page-size layer bucketing logic out of `_get_kv_cache_config_deepseek_v4` into a reusable helper, `_bucket_layers_by_page_size`. Behavior-preserving: same allocation pattern (one `KVCacheTensor` per \`(slot_idx… · full dump: [[thoughts/inference/vllm-prs/44454|#44454]]

