List your company

Inference Latency on Qwen 30B on M2 Ultra

0.572Wall Time (s)NPUMoE
0.549120.703560.8581.01244Apr 20, 2026

Evaluation results

MethodLinks
NPUMoEMode=ours2026.040.5720.4640.0861
coremlMode=coreml2026.041.1390.9780.0951.99
anemllMode=anemll2026.041.1440.9880.092