Semantic Caching on SemCacheLMArena
70Static Origin Served FractionLD1 (Logistic regression to compute threshold)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LD1 (Logistic regression to compute threshold)Threshold/Delta=N/A, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 70 | — | 2.6 | |
| LD2 (Logistic regression sigmoid fit)Threshold/Delta=N/A, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 68 | — | 2.1 | |
| GS (GPTCache)Threshold/Delta=0.97, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 67 | — | 5.2 | |
| LD3 (vCache)Threshold/Delta=0.05, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 54 | — | 2 | |
| GS (GPTCache)Threshold/Delta=0.98, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 53 | — | 4.1 | |
| LD3 (vCache)Threshold/Delta=0.03, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 46 | — | 1.1 | |
| GD (vCache with global threshold)Threshold/Delta=0.05, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 45 | — | 4.3 | |
| LD3 (vCache)Threshold/Delta=0.02, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 41 | — | 0.5 | |
| GS (GPTCache)Threshold/Delta=0.99, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 37 | — | 2.5 | |
| GD (vCache with global threshold)Threshold/Delta=0.03, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 26 | — | 2.5 | |
| KritesPolicy=Krites, Mechanism=Asynchronous VerifyAndPromote loop, Threshold=Pareto optimal threshold t*2026.02 | 19.4 | 136.5 | — | |
| GD (vCache with global threshold)Threshold/Delta=0.02, Embedding model=E5-large-v2, LLM=GPT-4o mini2025.02 | 14 | — | 1.3 | |
| BaselinePolicy=GPTCache-style, Threshold=Pareto optimal threshold t*2026.02 | 8.2 | — | — |