List your company

Inference Efficiency on LLaMA 8B 32K context length 3.1

1,115Theoretical Compute (TFLOPs)SpecKV
920.52971.011,021.51,071.99Mar 11, 2026

Evaluation results

MethodLinks
SpecKVCache Budget (C)=1282026.031,1151062,156402.82,263503
LAQCache Budget (C)=1282026.039304511,993239.262,314554
LOOKAHEADKVCache Budget (C)=1282026.03929131,7551.741,79838
Forward Pass OnlyCache Budget (C)=1282026.03928131,754—1,760—
SnapKVCache Budget (C)=1282026.03928131,7540.011,83878