Write, debug, and optimize GPU kernels using Triton and Gluon DSLs.
npx skills add https://github.com/slowlyC/agent-gpu-skills --skill triton-skill-slowlyc
Please help me install this Agent Skill. Skill: triton-skill Source: https://github.com/slowlyC/agent-gpu-skills/tree/main/triton_skill Command: npx skills add https://github.com/slowlyC/agent-gpu-skills --skill triton-skill-slowlyc
This Skill eliminates the steep learning curve and manual effort of developing, debugging, and optimizing high-performance GPU kernels using Triton and Gluon, which require deep expertise in GPU architecture, compiler internals, and architecture-specific optimization patterns for modern NVIDIA GPUs.
Use the triton-skill to implement a FP8 persistent matmul kernel for Blackwell GPUs that uses TMA async copies and warp specialization to maximize memory bandwidth.
High-intent search queries and answers about installing and using this skill.
To write high-performance GPU kernels for LLM inference using Triton, leverage the Python DSL to implement fused operations like attention and matmul, applying tensor core and TMA operations to maximize memory bandwidth.
Optimize CUDA kernels for Hopper and Blackwell GPUs by converting them to Triton, applying architecture-specific patterns like WGMMA, tcgen05, persistent execution, and cluster launch control for maximum performance.
The best way to debug and profile Triton GPU kernel performance is by performing roofline performance analysis to identify memory bandwidth bottlenecks and optimize tensor core utilization for high-performance computing workloads.
Yes, you can use Gluon and Triton to implement FP8 persistent matmul kernels with TMA async copies and warp specialization, specifically optimizing for Blackwell GPUs to maximize memory bandwidth in high-throughput serving systems.
No, you do not need deep GPU compiler knowledge to use Triton for kernel optimization, as the Skill provides a comprehensive reference library with tutorials, source code, and production-grade implementations covering basic syntax to advanced features.
Your custom GPU kernel may underperform during machine learning accelerator development if it lacks kernel fusion, persistent execution patterns, or proper tensor core and TMA operations, which are required to achieve peak hardware utilization.
m0atDevelop, debug, and autotune Triton and Gluon GPU kernels for ML workloads.
IsNoobgrammerOptimize Triton and CUDA kernels using roofline-based trial-loop benchmarking.
ROCmGenerate and validate FlyDSL GPU kernels from kernel specifications using tile programming.
ShekswessGuide writing, debugging, and optimizing GPU kernels with NVIDIA CuTe DSL.
