RT by @awnihannun: Excited to release mlx-lstsq (install `pip install mlx-lstsq`), a library for fast least-squares solving in MLX (via custom kernels) (opens in new tab)
Excited to release mlx-lstsq (install `pip install mlx-lstsq`), a library for fast least-squares solving in MLX (via custom kernels) - beating numpy, scipy, and torch baselines. Uses Cholesky Decomp on the Gram matrix to solve the normal equations. Works at extreme scales:
Read the original article