List your company

Inference Efficiency on LLaMA 8B 8K context length 3.1

159Theoretical Compute (TFLOPs)SpecKV
135.08141.29147.5153.71Mar 11, 2026

Evaluation results

MethodLinks
SpecKVCache Budget (C)=1282026.031598133779.53411121
SpecKV2026.031598133779.53411120.51
LOOKAHEADKVCache Budget (C)=1282026.03137132581.0330211
LAQCache Budget (C)=1282026.03137445492234.59800509
LOOKAHEADKV2026.03137132581.0330210.88
LAQ2026.03137445492234.59800509.38
Forward Pass OnlyCache Budget (C)=1282026.0313613257—291—
SnapKVCache Budget (C)=1282026.03136132570.0131120
Forward Pass Only2026.0313613257—291—
SnapKV2026.03136132570.0131120.17