---
cited_by:
  - '[[thoughts/inference/kimi-k3-prs#4-pr-reading-order|kimik3 §4]]'
date: '2026-11-11'
kind: pr
local: []
merged: '2026-09-10'
number: 56159
source:
  - https://github.com/vllm-project/vllm/blob/86aca6619161b308d675c753f228201e70b59ed7/vllm/models/kimi_k3/nvidia/kda.py
  - https://github.com/vllm-project/vllm/blob/86aca6619161b308d675c753f228201e70b59ed7/vllm/models/kimi_k3/nvidia/kda_metadata.py
  - https://github.com/vllm-project/vllm/blob/86aca6619161b308d675c753f228201e70b59ed7/vllm/models/kimi_k3/nvidia/ops/third_party/kda/chunk.py
  - https://github.com/vllm-project/vllm/blob/86aca6619161b308d675c753f228201e70b59ed7/vllm/models/kimi_k3/nvidia/ops/third_party/kda/fused_recurrent.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#56159 [Kimi K3 Perf] Avoid KDA mixed-batch gather/scatter, 5.2%~7.7% E2E Throughput Improvement'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/56159
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/56159
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/56159.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Avoid KDA mixed-batch gather/scatter, 5.2%\~7.7% E2E Throughput Improvement · full dump: [[thoughts/inference/vllm-prs/56159|#56159]]

