---
cited_by:
  - '[[thoughts/papers/kimiteam2026attentionresiduals#14-attnres-attention-residuals-arxiv-260315031-kimi-team-expansion-confirmed-from-the-paper-title|kimik3 §1.4]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-10-09'
number: 60717
source:
  - https://github.com/vllm-project/vllm/blob/88a419649e36f3412d2650c7162135a3b12bc5a1/vllm/models/kimi_k3/amd/ops/attn_res.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#60717 [ROCm][Kimi-K3] Use vLLM''s FP8 max in the fused AttnRes quantization'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/60717
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/60717
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/60717.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Fixes `test_amd_attn_res_fp8_preserves_prefix_and_quantized_output` on MI300 (fnuz FP8), which has failed since [#59069](https://github.com/aarnphm/aarnphm.github.io/issues/59069) landed. · full dump: [[thoughts/inference/vllm-prs/60717|#60717]]

