Seeing some WILD improvements to M5 Ultra performance for MLX after just one week.
Measured recent oMLX upstream builds, token prefill has essentially doubled across the board with Flash-Next.
Only one week has passed. Will need to update my review soon!