List your company

Question Answering on NarrativeQA LongBench (32K Context)

17.2F1 ScoreFull cache
0.1444.572913.428May 18, 2026

Evaluation results

MethodLinks
Full cacheModel=Qwen2.5-3B, Condition=Full cache, C (Cache capacity)=32768, Retention=100%2026.0517.2100—
LRU + protModel=Qwen2.5-3B, Condition=LRU + prot, C (Cache capacity)=4096, Retention=12.5%2026.0510.862.8—
Random + protModel=Qwen2.5-3B, Condition=Random + prot, C (Cache capacity)=4096, Retention=12.5%2026.0510.359.90.873
LRU + protModel=Qwen2.5-7B, Condition=LRU + prot, C (Cache capacity)=4096, Retention=12.5%2026.058.6112.1—
H2O + protModel=Qwen2.5-7B, Condition=H2O + prot, C (Cache capacity)=4096, Retention=12.5%2026.058.6111.90.893
SnapKV + protModel=Qwen2.5-7B, Condition=SnapKV + prot, C (Cache capacity)=4096, Retention=12.5%2026.058.6111.90.893
Random + protModel=Qwen2.5-7B, Condition=Random + prot, C (Cache capacity)=4096, Retention=12.5%2026.058.3107.20.99
Full cacheModel=Qwen2.5-7B, Condition=Full cache, C (Cache capacity)=32768, Retention=100%2026.057.7100—
Random (no prot)Model=Qwen2.5-7B, Condition=Random (no prot), C (Cache capacity)=4096, Retention=12.5%2026.051.722.40.001
H2O (no prot)Model=Qwen2.5-7B, Condition=H2O (no prot), C (Cache capacity)=4096, Retention=12.5%2026.051.721.90.001
LRU (no prot)Model=Qwen2.5-7B, Condition=LRU (no prot), C (Cache capacity)=4096, Retention=12.5%2026.051.317.50.001
LRU (no prot)Model=Qwen2.5-3B, Condition=LRU (no prot), C (Cache capacity)=4096, Retention=12.5%2026.050.84.60.001