Jacq

About

Jacq is a DSL and compiler that turns model and kernel descriptions written in .jac files into CUDA code. A model file produces a standalone CUDA inference runtime and a Python weight exporter, and kernel files can produce CUDA harnesses or callable library kernels. Validating, visualizing and generating code need only Rust, while running inference needs an NVIDIA GPU, CUDA, Python dependencies and weights. The repository also includes an experimental vLLM integration.