---
cited_by:
  - '[[thoughts/inference/architectures#3-what-changed-in-vllms-kv-cache-manager-to-hold-these|README §3]]'
  - '[[thoughts/inference/architectures#6-reading-order|README §6]]'
  - '[[thoughts/inference/kimi-k3-state#32-prefix-caching-for-recurrent-state|kimik3 §3.2]]'
  - '[[thoughts/inference/kimi-k3-prs#4-pr-reading-order|kimik3 §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/single_type_kv_cache_manager.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/sched/scheduler.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/kv_cache_interface.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/block_pool.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/kv_cache_coordinator.py
merged: '2026-09-06'
number: 53614
source:
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/v1/core/single_type_kv_cache_manager.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/v1/core/sched/scheduler.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/models/kimi_k3/nvidia/kda_metadata.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/v1/kv_cache_interface.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/v1/core/block_pool.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/v1/core/kv_cache_coordinator.py
  - https://github.com/vllm-project/vllm/blob/144e79c8106da23141ac010394b782f730cc7fe8/vllm/models/kimi_k3/nvidia/kda.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#53614 [Kimi K3] Support internal prefix checkpoints with partial prefix caching and spec-decoding'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/53614
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/53614
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/53614.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
speculative decoding / Eagle block rewind; · full dump: [[thoughts/inference/vllm-prs/53614|#53614]]

