Kronos

Kronos is yasp's embedded-compilation product. It takes a PyTorch model and builds a hardware-specific TensorRT engine on the GPU you're targeting — a Jetson, Drive AGX, or dGPU box — including models with ops that stock TensorRT can't convert on its own. The engine ships as a self-contained C++ package you can run, benchmark, and profile on the device.

A Kronos worker runs on the target hardware and does the actual build; you drive prepare, compile, and verify from your reference machine.

Guides

• Compile your first model — start a worker, then prepare, compile, and verify an engine in four commands.
• Run the demo on your hardware — unpack the build on the target GPU and run inference with the bundled yasp_inference_demo.
• Profile the model — get a per-layer GPU-time breakdown on the real device to find your hotspots.

Workers

• Kronos workers — pick the image for your hardware and deploy the worker with Docker (or Helm for a cluster).

Reference

• embedded reference — every flag for prepare, compile, and verify.