---
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: []
merged: '2026-10-08'
number: 60021
source:
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/amd/ple_layer.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/amd/model.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/amd/mtp.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/common/ops/__init__.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/nvidia/ngram_embedding.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/nvidia/ple_layer.py
  - https://github.com/vllm-project/vllm/blob/e2414ec1dbb177b54ac1bbce60e94769ebacbb28/vllm/models/qwen4_exp/common/ops/ple.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#60021 [ROCm][Qwen4Exp] Use fused PLE Triton kernels on the AMD backend'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/60021
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/60021
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/60021.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Ports the fused Qwen4Exp PLE Triton kernels from \[[#54517](https://github.com/aarnphm/aarnphm.github.io/issues/54517)\] (with the \[[#55375](https://github.com/aarnphm/aarnphm.github.io/issues/55375)\] stride fix) to the AMD backend, which still runs the older PyTorch PLE path. Follows up on \[[#54764](https://github.com/aarnphm/aarnphm.github.io/issues/54764)\]; supersedes \[[#58325](https://github.com/aarnphm/aarnphm.github.io/issues/58325)\]. · full dump: [[thoughts/inference/vllm-prs/60021|#60021]]

