home/categories/framework-internals/currentf-sparse-vllm-skills-add-sparse-method-skill-md
framework-internalsdevelopment

add-sparse-method

Add or refactor a first-class Sparse-vLLM sparse method alongside vanilla, SnapKV, OmniKV, QuEST, and DeltaKV. Use when Codex needs to introduce a new `vllm_sparse_method`, move method logic out of `attention.py` or `utils/`, add method-specific cache metadata or decode-time view building, wire config and registration, and preserve the repo's cache-manager-first architecture.

CURRENTF
maintainer
CURRENTF
Updated 3/13/2026
Stars
41
Forks
6
quick start

Installation and usage

Add or refactor a first-class Sparse-vLLM sparse method alongside vanilla, SnapKV, OmniKV, QuEST, and DeltaKV. Use when Codex needs to introduce a new `vllm_sparse_method`, move method logic out of `attention.py` or `utils/`, add method-specific cache metadata or decode-time view building, wire config and registration, and preserve the repo's cache-manager-first architecture.

Installation
$ install --globalskills.sh
Usage

Once installed, you can use this skill by running the following command in your terminal:

skills use add-sparse-method