Tune model throughput on AMD Instinct GPUs.
What it does
Sets up Hyperloom and hands off AMD Instinct inference workloads for benchmarked throughput optimization.
When to use it
Use to improve inference throughput on AMD Instinct GPUs, set up Hyperloom, or resume an optimization session, not for plain serving or isolated benchmarks.
How to use it
Ask to set up an optimization run; it guides credential and run-mode setup and saves workspace configuration before optimization.
Uses
What you need · 11
Requires an MI300X, MI325X, or MI355X GPU host with /dev/kfd and /dev/dri present and either amd-smi or rocm-smi working.
Requires ROCm on the GPU host; the agent must confirm its shell is on that host before installation.
Access · 2
This skill
hyperloom-inference-optimizer
Installs Hyperloom and its skills
Hyperloom
Starts hours-long GPU optimization