List your company
ResearchFormerly WizWand

LLM Decoding Throughput on Apple M4 Pro tg128

465Throughput (tokens/s)BaseRT
52.12159.31266.5373.69Jul 1, 2026
Updated 4mo ago

Evaluation results

MethodLinks
BaseRTModel=Qwen3 0.6B, Quantization=4-bit2026.074651.17
uzuModel=Qwen3 0.6B, Quantization=4-bit2026.07398—
BaseRTModel=Qwen3 0.6B, Quantization=8-bit2026.073211.19
BaseRTModel=Llama 3.2 1B, Quantization=4-bit2026.072951.02
uzuModel=Llama 3.2 1B, Quantization=4-bit2026.07290—
uzuModel=Qwen3 0.6B, Quantization=8-bit2026.07271—
BaseRTModel=Llama 3.2 1B, Quantization=8-bit2026.071841.06
uzuModel=Llama 3.2 1B, Quantization=8-bit2026.07173—
uzuModel=Llama 3.2 3B, Quantization=4-bit2026.07119—
BaseRTModel=Llama 3.2 3B, Quantization=4-bit2026.071170.99
BaseRTModel=Llama 3.2 3B, Quantization=8-bit2026.07711.04
uzuModel=Llama 3.2 3B, Quantization=8-bit2026.0768—