skills.rest
linkedinlinkedinOfficial·3 Agent Skills Included

Liger-Kernel

Build, patch, and optimize Triton kernels for faster LLM training

Develops production-ready Triton kernels from PyTorch references, adds Liger Kernel support to new HuggingFace models, and tunes existing kernels for speed and memory. Removes the guesswork of GPU kernel engineering with staged analyze-generate-validate pipelines, automated benchmarks, and convergence tests. Helps engineers ship faster LLM training code with up to 20% higher throughput and 60% lower memory usage.
6,558|578
↗
npx skills add linkedin/Liger-Kernel --all -g -y
Available:

Directs the AI agent to the three workflow skills in this repository (kernel development, model auto-patching, and kernel performance optimization) and explains how to route requests through their staged pipelines.

All Skills in This Repository (3)

Pure Emerald Level Indicators
📦 In Repo
linkedinlinkedin

liger-kernel-dev

Automate Triton kernel development for Liger Kernel with a three-stage pipeline.

Official
Advanced
📦 In Repo
linkedinlinkedin

liger-autopatch

Generate and validate Liger Kernel monkey-patches for HuggingFace Transformers models.

Official
Advanced
📦 In Repo
linkedinlinkedin

liger-kernel-perf

Profile, generate variants, and validate Liger Kernel Triton kernels for NVIDIA GPUs.

Official
Advanced

Frequently Asked Questions

FAQPage Schema
How to install Liger-Kernel skills?▼

Run `npx skills add linkedin/Liger-Kernel --all -g -y` in your terminal to install all skills in this suite globally.

How to create a new Triton kernel with AI?▼

The liger-kernel-dev skill turns any PyTorch operation into a production-ready Triton kernel through a 3-stage analyze, generate, and validate pipeline with tests and benchmarks.

How to add Liger Kernel support for a new HuggingFace model?▼

The liger-autopatch skill reads the model's source code, builds an architecture profile, and generates the monkey-patch, forward function, tests, and documentation automatically.

Can AI optimize an existing Triton kernel?▼

Yes. The liger-kernel-perf skill profiles the kernel, diagnoses bottlenecks, benchmarks multiple optimization variants, and applies the fastest correct version.

Do these skills require an NVIDIA GPU?▼

Kernel development and performance optimization require an NVIDIA GPU for benchmarking and validation, while the auto-patch skill mainly needs a standard Python environment.

Related Repositories in Software Engineering

View All in Software Engineering→
openclawopenclaw
388.4k

openclaw

Run a personal AI assistant across your devices and chat apps

Official📦 120 Skills
obraobra
280.0k

superpowers

Gives coding agents a disciplined workflow from idea to merged code

Community📦 15 Skills
facebookfacebook
248.4k

react

AI agent skills for building, testing, and porting React core

Official📦 14 Skills