List your company

Memory Footprint Analysis on Qwen2.5-3B-Instruct K3V3 deployment

155.7KV Cache Size (MB)TQ-MSE
-76.6441,491.6783,0604,628.322Jun 23, 2026

Evaluation results

MethodLinks
TQ-MSET (Context Length)=16K2026.06155.73.887.941.77
Block-GTQT (Context Length)=16K2026.06186.43.247.971.8
FP16 FA-2T (Context Length)=16K2026.06604112.536.36
TQ-MSET (Context Length)=64K2026.06622.93.8812.946.77
Block-GTQT (Context Length)=64K2026.06745.53.2413.066.89
TQ-MSET (Context Length)=128K2026.061,245.73.8819.613.43
Block-GTQT (Context Length)=128K2026.061,491.13.2419.8513.67
FP16 FA-2T (Context Length)=64K2026.062,415.9131.2925.12
TQ-MSET (Context Length)=256K2026.062,491.43.8832.9326.76
Block-GTQT (Context Length)=256K2026.062,982.23.2433.4227.25
FP16 FA-2T (Context Length)=128K2026.064,831.8156.3150.13
TQ-MSET (Context Length)=512K2026.064,982.83.8859.5853.41
Block-GTQT (Context Length)=512K2026.065,964.33.2460.5654.39