---
cited_by:
  - '[[thoughts/inference/architectures#5-four-things-called-replay|README §5]]'
  - '[[thoughts/inference/deepseek-v4.1-state#33-decoder-side-bounded-replay-58132-59532-59894|dsv41 §3.3]]'
  - '[[thoughts/inference/deepseek-v4.1-serving#4-pr-reading-order|dsv41 §4]]'
  - '[[thoughts/inference/replayssm#4-which-models-use-it|replayssm §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/config/compilation.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/gpu/model_states/interface.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/worker/gpu/model_runner.py
merged: '2026-10-06'
number: 59532
source:
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/models/deepseek_v41/nvidia/model_state.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/models/deepseek_v41/nvidia/decoder_replay_cudagraph.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/models/deepseek_v41/decoder_replay_layers.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/config/compilation.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/v1/worker/gpu/model_states/interface.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/models/deepseek_v41/nvidia/model.py
  - https://github.com/vllm-project/vllm/blob/eb2901b7356dd6782b7418ac071d3cf56ce018f7/vllm/v1/worker/gpu/model_runner.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#59532 [Perf][DSv4.1] Run decoder replay layers in CUDA graphs and trim in PIECEWISE graphs'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/59532
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/59532
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/59532.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Run the decoder-side SWA bounded replay layers ([#58132](https://github.com/aarnphm/aarnphm.github.io/issues/58132)) in CUDA graphs of their own, and trim them in PIECEWISE graph steps too. [#58132](https://github.com/aarnphm/aarnphm.github.io/issues/58132) trims layers 21-39 to each prefill’s last 128 tokens only on eager steps (above the model graph’s… · full dump: [[thoughts/inference/vllm-prs/59532|#59532]]

