Performance is competitive with Python MLX. On small models, Ruby is within 0.55-1.54x of Python depending on model type and device. The heavy lifting happens in the same C++ / Metal runtime either way.

Ruby deserves better ML tooling. The language is expressive enough that model definitions can actually be more readable than their Python equivalents. gem install mlx to try it out.