---
cited_by:
  - '[[thoughts/inference/speculative-drafting#1-read-order|reading-map §1]]'
  - '[[thoughts/papers/wang2026xpressparallelrefinementdiffusion|reading-map §2.4]]'
  - '[[thoughts/inference/speculative-drafting#3-vllm-status-at-a-glance|reading-map §3]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/config/speculative.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/gpu/spec_decode/__init__.py
  - file:///Users/aarnphm/workspace/vllm/.buildkite/test_areas/spec_decode.yaml
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/models/registry.py
  - file:///Users/aarnphm/workspace/vllm/vllm/config/vllm.py
merged: '2026-10-22'
number: 54448
source:
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/v1/worker/gpu/spec_decode/xpress/kernels.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/model_executor/models/qwen3_xpress.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/v1/worker/gpu/spec_decode/xpress/speculator.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/config/speculative.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/v1/worker/gpu/spec_decode/__init__.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/.buildkite/test_areas/spec_decode.yaml
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/model_executor/models/registry.py
  - https://github.com/vllm-project/vllm/blob/ce9467b757c024a55fea4d0919ac2e682f28a894/vllm/config/vllm.py
state: merged
tags:
  - vllm
  - vllm-pr
  - specdec
title: '#54448 [Core] Add XPress speculative decoding method'
topic: specdec
url: https://github.com/vllm-project/vllm/pull/54448
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/54448
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/54448.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Adds XPress, a causal refiner for block-diffusion drafters, as a V1 speculative decoding method alongside DFlash and DSpark. Paper: · full dump: [[thoughts/inference/vllm-prs/54448|#54448]]

