---
cited_by:
  - '[[thoughts/inference/qwen3.8-serving#4-pr-reading-order-vllm-main-git-show-sha|qwen4next §4]]'
  - '[[thoughts/inference/qwen3.8-serving#5-sharp-edges-bugs-that-expose-invariants|qwen4next §5]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-10-07'
number: 59945
source:
  - https://github.com/vllm-project/vllm/blob/d9503dc503e75193d2128908b59c0d141d4b5fb1/vllm/models/qwen4_exp/amd/qsa.py
  - https://github.com/vllm-project/vllm/blob/d9503dc503e75193d2128908b59c0d141d4b5fb1/vllm/models/qwen4_exp/nvidia/qsa.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#59945 [Bugfix][Qwen4Exp] Derive the QSA head_dim from the unsharded head count'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/59945
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/59945
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/59945.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Qwen4Exp QSA computes its fallback `head_dim` from `num_heads`, which is already divided by the tensor-parallel size, so on TP > 1 the fallback is `tp_size` times too large. The fallback is latent: `Qwen4ExpTextConfig` defaults `head_dim`… · full dump: [[thoughts/inference/vllm-prs/59945|#59945]]

