---
cited_by:
  - '[[thoughts/inference/deepseek-v4.1-serving#4-pr-reading-order|dsv41 §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backends/mla/indexer.py
  - file:///Users/aarnphm/workspace/vllm/vllm/config/attention.py
merged: '2026-09-29'
number: 58671
source:
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/v1/attention/ops/rocm_paged_mxfp4_indexer.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/v1/attention/backends/mla/rocm_paged_mxfp4_indexer.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/models/deepseek_v41/amd/rocm.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/model_executor/layers/rocm_paged_mxfp4_indexer.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/models/deepseek_v41/attention.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/v1/attention/backends/mla/indexer.py
  - https://github.com/vllm-project/vllm/blob/63b9931b7399c9bfa7a8f7cdd60e2991ba87fbf9/vllm/config/attention.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#58671 [ROCm][DSv4.1] Paged MXFP4 sparse indexer on aiter''s MQA-logits kernel'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/58671
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/58671
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/58671.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Regular MXFP4 (`indexer_kv_dtype="mxfp4"`): dense logits over the whole context, read straight from the paged K cache. · full dump: [[thoughts/inference/vllm-prs/58671|#58671]]

