---
cited_by:
  - '[[thoughts/papers/qiu2026designqwen38nextarchitectureevaluation#11-qsa--qwen-sparse-attention|qwen4next §1.1]]'
  - '[[thoughts/inference/qwen3.8-serving#4-pr-reading-order-vllm-main-git-show-sha|qwen4next §4]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-10-05'
number: 60023
source:
  - https://github.com/vllm-project/vllm/blob/4ff028d77e063e0055849ffec3855c6634179158/vllm/models/qwen4_exp/nvidia/qsa.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#60023 [Bugfix][Qwen4Exp] Honor --kv-cache-dtype-skip-layers in QSA attention'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/60023
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/60023
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/60023.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
`Qwen4ExpQSAAttention` takes its KV cache dtype straight from `cache_config.cache_dtype`, so · full dump: [[thoughts/inference/vllm-prs/60023|#60023]]

