---
cited_by:
  - '[[thoughts/inference/architectures#5-four-things-called-replay|README §5]]'
  - '[[thoughts/inference/architectures#6-reading-order|README §6]]'
  - '[[thoughts/inference/kimi-k3-state#33-spec-decode-with-recurrent-state|kimik3 §3.3]]'
  - '[[thoughts/inference/kimi-k3-prs#4-pr-reading-order|kimik3 §4]]'
  - '[[thoughts/inference/replayssm#4-which-models-use-it|replayssm §4]]'
  - '[[thoughts/inference/replayssm-serving#5-pr-reading-order|replayssm §5]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/gpu/model_states/mamba_hybrid.py
  - file:///Users/aarnphm/workspace/vllm/vllm/config/vllm.py
  - file:///Users/aarnphm/workspace/vllm/vllm/model_executor/layers/mamba/mamba_utils.py
merged: '2026-08-17'
number: 51855
source:
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/models/kimi_k3/nvidia/ops/recoverssm.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/models/kimi_k3/nvidia/kda_metadata.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/v1/worker/gpu/model_states/recoverssm.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/models/kimi_k3/nvidia/kda.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/v1/worker/gpu/model_states/mamba_hybrid.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/config/vllm.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/models/kimi_k3/nvidia/model.py
  - https://github.com/vllm-project/vllm/blob/70afdedc1081d28c3eaae53bece8292298484c86/vllm/model_executor/layers/mamba/mamba_utils.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#51855 [K3] support recoverssm for K3'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/51855
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/51855
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/51855.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
This PR adds RecoverSSM, a Kimi-K3-specific state recovery path for speculative KDA decoding on NVIDIA Model Runner V2. It is enabled through `--use-replayssm`, is intended for Kimi-K3 serving with DSpark, and supports… · full dump: [[thoughts/inference/vllm-prs/51855|#51855]]

