SonicBoom
A native tensor runtime. No Python. One .so.
SonicBoom is written from the ground up in C++ and CUDA, exposed through a stable C ABI, with GNU Guile as its first-class language. No Python interpreter, pip or wheel tree is involved in inference.
Where to start
- Getting started — build SonicBoom and run a model from Guile.
- The Guile binding — the API reference, generated from the modules’ documentation.
- Architecture — how the pieces fit together.
- Coming from PyTorch — look up a PyTorch name and find the SonicBoom function or the few lines that do the same.
- Benchmarks — measured numbers.
- Roadmap — what is done and what comes next.