---
cited_by:
  - '[[thoughts/inference/deepseek-v4.1-serving#18-moe--gate|dsv41 §1.8]]'
  - '[[thoughts/inference/deepseek-v4.1-serving#5-sharp-edges-bugs-and-fixes-that-expose-invariants|dsv41 §5]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/.buildkite/test_areas/lm_eval.yaml
  - file:///Users/aarnphm/workspace/vllm/.buildkite/test_areas/quantization.yaml
merged: '2026-10-08'
number: 59655
source:
  - https://github.com/vllm-project/vllm/blob/8faac3dd2af39fd631f018277387f01c89ad5782/vllm/models/deepseek_v41/quant_config.py
  - https://github.com/vllm-project/vllm/blob/8faac3dd2af39fd631f018277387f01c89ad5782/.buildkite/test_areas/lm_eval.yaml
  - https://github.com/vllm-project/vllm/blob/8faac3dd2af39fd631f018277387f01c89ad5782/.buildkite/test_areas/quantization.yaml
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#59655 [Bugfix] [Spec Decode]   DeepSeek V4/V4.1 NVFP4: keep ignored MTP/DSpark experts on MXFP4'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/59655
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/59655
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/59655.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
With `nvidia/DeepSeek-V4.1-Flash-NVFP4`, DSpark speculative decoding accepts no draft tokens. Acceptance length is · full dump: [[thoughts/inference/vllm-prs/59655|#59655]]

