C++17 library of hand-written x86 SIMD kernels for the arithmetic at the core of neural-network inference: matrix multiplication, the matrix-vector product that dominates generating text with a language model one token at a time, and activation and normalization functions.
python machine-learning cpp simd high-performance-computing performance-optimization microkernels ml-primitives
-
Updated
Oct 21, 2026 - C++