---
cited_by:
  - '[[thoughts/inference/kimi-k3-state#34-pd-dcp-kv-offload-59229-test-matrix|kimik3 §3.4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/attention/mla_attention.py
merged: '2026-10-06'
number: 56881
source:
  - https://github.com/vllm-project/vllm/blob/b2863c0de953204b64e4e3d184f96eac26bd5b46/vllm/model_executor/layers/attention/mla_attention.py
  - https://github.com/vllm-project/vllm/blob/b2863c0de953204b64e4e3d184f96eac26bd5b46/vllm/models/kimi_k3/nvidia/mla.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#56881 [Kimi-K3] Use the fused K/V pack kernel on the decode-context-parallel prefill path'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/56881
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/56881
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/56881.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Kimi-K3 prefill with a long context runs in chunks. For every chunk of every MLA layer, the model projects the gathered latent cache into per-head keys and values, casts them to FP8, and lays them out as the contiguous key and value… · full dump: [[thoughts/inference/vllm-prs/56881|#56881]]

