3 ms·
> The one being benchmarked here is heavily optimised for Apple Silicon. I don't think so, in the sense of a hand-optimized CUDA implementation. This just usin
by jdminhbg 3y ago
> The one being benchmarked here is heavily optimised for Apple Silicon.
I don't think so, in the sense of a hand-optimized CUDA implementation. This just using the MLX API in the same way that you'd use CUDA via PyTorch or something.