It needs a mlx fork because the lowest bit in mlx is 2 currently (for affine quantization).
That mlx is for apple hardware only, though? Or did I misunderstand something.
That mlx is for apple hardware only, though? Or did I misunderstand something.