---
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/v1/attention/backends/mla/flashinfer_mla.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/attention/mla_attention.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/gpu/cp_utils.py
  - file:///Users/aarnphm/workspace/vllm/vllm/config/speculative.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/attention/backend.py
merged: '2026-08-17'
number: 52188
source:
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/attention/backends/mla/flashinfer_mla.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/worker/gpu/spec_decode/dflash/speculator.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/model_executor/layers/attention/mla_attention.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/worker/gpu/cp_utils.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/worker/gpu/spec_decode/dflash/cudagraph.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/config/speculative.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/attention/backend.py
  - https://github.com/vllm-project/vllm/blob/d1e3eee6fb8ed3623241ef5c8e3ac533f775bff9/vllm/v1/worker/gpu/spec_decode/speculator.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#52188 [Spec decode] Support Kimi-K3 DCP with DSpark'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/52188
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/52188
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/52188.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
This PR adds support for running Kimi-K3 decode context parallel with DSpark with FlashinferMLA and Tokenspeed as target causal attention backend and Tokenspeed as the draft non-causal backend. · full dump: [[thoughts/inference/vllm-prs/52188|#52188]]

