---
cited_by:
  - '[[thoughts/inference/architectures#3-what-changed-in-vllms-kv-cache-manager-to-hold-these|README §3]]'
  - '[[thoughts/inference/deepseek-v4.1-serving#5-sharp-edges-bugs-and-fixes-that-expose-invariants|dsv41 §5]]'
  - '[[thoughts/inference/review-deepseek#grill-questions-answer-key-follows-each|grill-key § grill questions (answer key…]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-10-05'
number: 58560
source:
  - https://github.com/vllm-project/vllm/blob/49e0f4797853da0d1da7fdfe794f327fb5d5843d/vllm/models/deepseek_v41/compressor.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#58560 [Bugfix][DSv4.1] Keep the compressor ring out of the null block'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/58560
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/58560
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/58560.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
`_ring_slot_mapping_kernel` (`vllm/models/deepseek_v41/compressor.py`) maps each token to `block_table[req, 0] * capacity + pos % capacity`. Warmup and CUDA-graph capture run dummy batches whose block tables are zeroed, so their compressor… · full dump: [[thoughts/inference/vllm-prs/58560|#58560]]

