List your company

End-to-end attention latency measurement on Apple M2

0.87Latency (ms)IntAttention
-20.2712122.4319265.135407.8381Nov 26, 2025

Evaluation results

MethodLinks
IntAttentionSequence Length=1K2025.110.87
Quant-OnlySequence Length=1K2025.111.61
FP16Sequence Length=1K2025.112.08
IntAttentionSequence Length=2K2025.112.62
FP32Sequence Length=1K2025.112.81
Quant-OnlySequence Length=2K2025.115.85
FP16Sequence Length=2K2025.116.72
FP32Sequence Length=2K2025.119.04
IntAttentionSequence Length=4K2025.1110.64
Quant-OnlySequence Length=4K2025.1122.41
FP16Sequence Length=4K2025.1125.42
FP32Sequence Length=4K2025.1132.64
IntAttentionSequence Length=8K2025.1138.86
Quant-OnlySequence Length=8K2025.1184.6
FP16Sequence Length=8K2025.11100.36
FP32Sequence Length=8K2025.11125.13
IntAttentionSequence Length=16K2025.11142.5
Quant-OnlySequence Length=16K2025.11341.7
FP16Sequence Length=16K2025.11398.39
FP32Sequence Length=16K2025.11529.4