Find your MAX model's speed under real load.
What it does
Measures a running MAX model's throughput and latency using workloads matched to the user's performance question.
When to use it
Use to measure MAX model performance, sweep concurrency or request rates, size deployments, or troubleshoot failed benchmark runs.
How to use it
Give it a MAX performance question; it runs a matching benchmark, reports throughput and latency, and saves results as JSON.
Uses
What you need · 3
Requires a MAX installation through pip or pixi; commands in a pixi project use the pixi run prefix.
Requires a running model endpoint before benchmarking.
GPU utilization and peak-memory collection require NVIDIA hardware, with the benchmark running on the same machine as the server.
Access · 1
This skill
max serve
Starts or reconfigures MAX serving
Sends inference requests to MAX