Develop, debug, and autotune Triton and Gluon GPU kernels for ML workloads.
npx skills add https://github.com/m0at/claudemd --skill triton-skill
Please help me install this Agent Skill. Skill: triton-skill Source: https://github.com/m0at/claudemd/tree/main/skills/triton Command: npx skills add https://github.com/m0at/claudemd --skill triton-skill
Streamlines the development, debugging, and optimization of Triton and Gluon GPU kernels by consolidating local source access, tutorials, and kernel references for quick learning and productive work.
Run a quick tutorial to implement and test a vector-add Triton kernel.
High-intent search queries and answers about installing and using this skill.
To develop Triton GPU kernels, you use operations like tl.load, tl.store, and tl.dot within triton.jit to build and optimize custom ML workloads. You can debug and validate kernel correctness using provided tutorials and production references.
Porting CUDA kernels to Triton involves translating CUDA logic into Triton's Python-based DSL using block-level operations like tl.dot and tl.store. The Skill provides practical workflows and references to guide the porting and performance profiling process.
Autotuning in Triton GPU kernels optimizes performance by searching over configurable kernel parameters like block sizes and instruction counts. It is supported directly in the development workflow to achieve device-specific optimizations for complex operations like fused attention.
Yes, accessing local Triton source code, tutorials, and kernel references is required. This local environment supports learning Gluon operations, gluon.jit, and MLIR concepts for effective kernel-level optimization and development.
Yes, you can optimize complex fused attention kernels using Triton and Gluon. The Skill supports guided kernel development from simple vector additions to advanced fused attention with autotuning and device-specific performance optimizations.
Triton streamlines GPU kernel development by consolidating local source access, tutorials, and references into a productive workflow. It allows you to write, autotune, and port CUDA kernels using higher-level abstractions like tl.dot and triton.jit.
mindspore-aiImplement GPU kernels in Triton CUDA using Python and the @triton.jit decorator.
xchang1121Implement Triton CUDA kernel templates for elementwise, reduction, and matmul workloads.
mindspore-aiIntegrate Triton CUDA kernels with PyTorch models for custom operators.
ShekswessAuthor, debug, and optimize GPU kernels with Triton in Python.
