Implement and optimize GPU kernels using PyTorch logic and DSLs.
npx skills add https://github.com/alibaba/atrex-kernel-agent --skill gpu-kernel-optimizer
Please help me install this Agent Skill. Skill: gpu-kernel-optimizer Source: https://github.com/alibaba/atrex-kernel-agent/tree/main Command: npx skills add https://github.com/alibaba/atrex-kernel-agent --skill gpu-kernel-optimizer
This Skill solves the complexity of manual GPU kernel optimization by providing a structured, profile-driven workflow that transforms PyTorch logic into high-performance GPU kernels while ensuring hardware-specific constraints are met.
Use the gpu-kernel-optimizer skill to initialize a new optimization workspace for the kernel file named kernel_demo.py on the H20 platform.
High-intent search queries and answers about installing and using this skill.
To optimize PyTorch logic for specific hardware, this Skill bridges it with high-performance DSLs like CuteDSL or FlyDSL. It enforces hardware-spec compliance by sourcing all architectural metrics from a local knowledge base.
Profile-driven GPU kernel optimization identifies and resolves bottlenecks using Nsight Compute or ROCm profiling tools. It facilitates iterative refinement through bottleneck analysis, Roofline modeling, and automated performance validation.
You can identify performance bottlenecks in a CUDA or Triton kernel by using Nsight Compute or ROCm profiling tools within an optimization workspace. This guides the iterative optimization loop until targets are achieved.
Yes, you need Python3, git, and jq installed to run the automated GPU kernel optimization workspace. These dependencies are required to manage the workspace and execute the iterative optimization loop.
The best way to perform Roofline modeling for AMD ROCm kernels is using this Skill's structured workflow. It runs Roofline analysis and manages the iterative optimization loop to achieve performance targets.
KuberwastakenRun multiple autonomous ML research agents on GPU machines with git branches.
m0atAnalyze CUDA kernel performance bottlenecks with Nsight Systems and Nsight Compute.
alibabaImplement and validate baseline GPU kernels using CuteDSL or FlyDSL.
m0atDevelop, debug, and autotune Triton and Gluon GPU kernels for ML workloads.
