---
cited_by:
  - '[[thoughts/papers/kimiteam2026attentionresiduals#14-attnres-attention-residuals-arxiv-260315031-kimi-team-expansion-confirmed-from-the-paper-title|kimik3 §1.4]]'
  - '[[thoughts/inference/kimi-k3-prs#4-pr-reading-order|kimik3 §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/csrc/libtorch_stable/ops.h
  - file:///Users/aarnphm/workspace/vllm/csrc/libtorch_stable/torch_bindings.cpp
  - file:///Users/aarnphm/workspace/vllm/vllm/_custom_ops.py
merged: '2026-08-31'
number: 54261
source:
  - https://github.com/vllm-project/vllm/blob/d6d665854314f0aa90ad6ef32a3382136c71314c/csrc/libtorch_stable/kimi_k3/attn_res_kernel.cu
  - https://github.com/vllm-project/vllm/blob/d6d665854314f0aa90ad6ef32a3382136c71314c/vllm/models/kimi_k3/nvidia/ops/attn_res.py
  - https://github.com/vllm-project/vllm/blob/d6d665854314f0aa90ad6ef32a3382136c71314c/csrc/libtorch_stable/ops.h
  - https://github.com/vllm-project/vllm/blob/d6d665854314f0aa90ad6ef32a3382136c71314c/csrc/libtorch_stable/torch_bindings.cpp
  - https://github.com/vllm-project/vllm/blob/d6d665854314f0aa90ad6ef32a3382136c71314c/vllm/_custom_ops.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#54261 [Kimi-K3][Perf] Make native CUDA AttnRes the SM100 default'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/54261
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/54261
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/54261.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Make the native CUDA Kimi-K3 AttnRes implementation the default for dense · full dump: [[thoughts/inference/vllm-prs/54261|#54261]]

