List your company

KV Cache Compression on 5 sequences Qwen2.5-1.5B-Instruct

0.0358KL DivergenceKVSCULPT
-0.0627680.6025661.26791.933234Mar 29, 2026

Evaluation results

MethodLinks
KVSCULPTCompression Ratio (r)=0.7, Retain zone (m)=256, Optimizer=L-BFGS, Learning rate=0.52026.030.0358
KVSCULPTCompression Ratio (r)=0.5, Retain zone (m)=256, Optimizer=L-BFGS, Learning rate=0.52026.030.0463
KVSCULPTCompression Ratio (r)=0.3, Retain zone (m)=256, Optimizer=L-BFGS, Learning rate=0.52026.030.0575
Joint OptimizationCompression Ratio (r)=0.7, Retain zone (m)=2562026.030.115
Select+FitCompression Ratio (r)=0.7, Retain zone (m)=2562026.030.125
Attention ScoreCompression Ratio (r)=0.7, Retain zone (m)=2562026.030.139
Joint OptimizationCompression Ratio (r)=0.5, Retain zone (m)=2562026.030.18
Select+FitCompression Ratio (r)=0.5, Retain zone (m)=2562026.030.186
Attention ScoreCompression Ratio (r)=0.5, Retain zone (m)=2562026.030.204
Joint OptimizationCompression Ratio (r)=0.3, Retain zone (m)=2562026.030.224
Select+FitCompression Ratio (r)=0.3, Retain zone (m)=2562026.030.233
Attention ScoreCompression Ratio (r)=0.3, Retain zone (m)=2562026.030.254
RandomCompression Ratio (r)=0.7, Retain zone (m)=2562026.031.54
RandomCompression Ratio (r)=0.5, Retain zone (m)=2562026.031.8
RandomCompression Ratio (r)=0.3, Retain zone (m)=2562026.032.5