Original post
Remi Cadene#1727
Steeve Morin@steeve
.@zml_ai's Metal backend is now as fast as MLX in tok/s (llama 3.1 8B)
1:16 AM · Jun 6, 2026 · 2.4K Views
.@zml_ai's Metal backend is now as fast as MLX in tok/s (llama 3.1 8B)

@steeve @zml_ai Maybe you find this useful: https://github.com/pedronahum/MetalHLO

@steeve @zml_ai Were y'all watching that? Automatic delivery