Find and validate faster GPU kernels with Magpie.
What it does
Benchmarks inference workloads and evaluates GPU kernel variants so users can identify bottlenecks and validate optimizations end to end.
When to use it
Use for inference benchmarking, trace analysis, bottleneck discovery, kernel comparison, or validation of GPU optimizations.
How to use it
Give it an inference workload or kernel implementations in a project; it produces benchmark reports, evaluated candidates, and end-to-end optimization results.
What you provide
- An existing project
What you need · 3
Executing Magpie workloads requires an intended Python environment running Python 3.10 or newer.
Workload execution requires a working Magpie installation; the skill installs it if missing. Preparation-only plans do not require it.
Before workload execution, check GPU visibility and verify hardware compatibility rather than assuming support.
Access · 1
This skill
magpie-eval
Installs Magpie from GitHub