basalt-org/basalt

Implements core ML operations (matrix multiplication, convolution, activation functions) with static graph compilation via Mojo's parametric functions and MLIR backend, enabling compile-time optimizations comparable to PyTorch performance. Includes custom Tensor implementations and benchmarking against standard frameworks, with a roadmap targeting GPU support, operator fusion, and ONNX compatibility.

442 stars. No commits in the last 6 months.