liuliu warns: claimed 4x-10x speedups over MLX or llama.cpp on Apple hardware are noise

teortaxesTex · x · 2026-09-23

Researcher liuliu followed up on his earlier take: be suspicious of any claims of 4x, 8x, or 10x speedups over MLX or llama.cpp from "custom / model-specific inference engines" on Apple hardware — such numbers are mostly noise.

Original post →

More from Infra

Infra channel →