---
cited_by:
  - '[[thoughts/inference/kimi-k3-prs#4-pr-reading-order|kimik3 §4]]'
date: '2026-11-11'
kind: pr
local:
  - file:///Users/aarnphm/workspace/vllm/vllm/config/vllm.py
  - file:///Users/aarnphm/workspace/vllm/vllm/v1/core/sched/scheduler.py
merged: '2026-08-11'
number: 51725
source:
  - https://github.com/vllm-project/vllm/blob/0914ed2e816fa1987951f22afba1c00b7786f110/vllm/config/vllm.py
  - https://github.com/vllm-project/vllm/blob/0914ed2e816fa1987951f22afba1c00b7786f110/vllm/v1/core/sched/scheduler.py
state: merged
tags:
  - vllm
  - vllm-pr
  - new-architectures
title: '#51725 [Perf] Adaptive budget for spec scheduled token, 55%~65% E2E TTFT Improvement'
topic: new-architectures
url: https://github.com/vllm-project/vllm/pull/51725
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/51725
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/51725.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
Let’s consider `max_num_seqs=1024`and `max_num_batched_tokens=8192` (by default) · full dump: [[thoughts/inference/vllm-prs/51725|#51725]]

