ExecuTorch MLX Delegate Speeds Qwen3 4.52x on M1 Max

ExecuTorch's experimental MLX delegate, released in May 2026, enables PyTorch models to run on Apple Silicon GPUs. Testing Qwen3-0.6B on an M1 Max showed decode throughput of 188.9 tokens/s with INT4 quantization, 4.52x faster than PyTorch MPS BF16. However, INT4 altered output in two of three prompts, highlighting a trade-off between speed and accuracy.