---
cited_by:
  - '[[thoughts/inference/qwen3.8-serving#4-pr-reading-order-vllm-main-git-show-sha|qwen4next §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/short_conv_attn.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mamba_attn.py
merged: '2026-09-28'
number: 58114
source:
  - https://github.com/vllm-project/vllm/blob/20b52e9f5b5793d56586b91df9b93cbe30a24040/vllm/v1/attention/backends/short_conv_attn.py
  - https://github.com/vllm-project/vllm/blob/20b52e9f5b5793d56586b91df9b93cbe30a24040/vllm/v1/attention/backends/mamba_attn.py
  - https://github.com/vllm-project/vllm/blob/20b52e9f5b5793d56586b91df9b93cbe30a24040/vllm/models/qwen4_exp/amd/ple_layer.py
  - https://github.com/vllm-project/vllm/blob/20b52e9f5b5793d56586b91df9b93cbe30a24040/vllm/models/qwen4_exp/nvidia/ple_layer.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#58114 [Perf][Qwen3.8] Reduce PLE metadata construction overhead'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/58114
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/58114
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/58114.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
PLE metadata construction still synchronizes in mixed speculative batches and builds temporary tensors that its consumers do not need. This reduces per-step builder work while preserving request ordering, accepted-token routing, and… · full dump: [[thoughts/inference/vllm-prs/58114|#58114]]

