---
cited_by:
  - '[[thoughts/inference/agent-tooling#1-read-these-first-in-order|reading-map §1]]'
  - '[[thoughts/inference/agent-review#43-monokernels-rfc-60904-as-the-isolation-bet-made-concrete|reading-map §4.3]]'
date: '2026-11-11'
kind: rfc
local: []
number: 60904
source: []
state: open
tags:
  - vllm
  - vllm-pr
  - sprint-skills-agent-friendliness
title: '#60904 [RFC]: A standard for Specialized (MonoKernels) Decodes in vllm/models'
topic: sprint-skills-agent-friendliness
url: https://github.com/vllm-project/vllm/issues/60904
created: '2026-11-11'
modified: '2026-11-11'
published: '2026-11-11'
pageLayout: default
slug: hinterland/papers/maps/prs/60904
permalink: https://aarnphm.xyz/hinterland/papers/maps/prs/60904.md
generator:
  quartz: v4.6.0
  hostedProvider: Cloudflare
  baseUrl: aarnphm.xyz
full: https://aarnphm.xyz/llms-full.txt
---
A Specialized Decode implementation is one where most/all operations in a model’s forward pass are manually implemented in a way that optimizes for the model’s specific shapes and execution patterns on certain hardware. Typically, these… · full dump: [[thoughts/inference/vllm-prs/60904|#60904]]

