Long-term Chatting Evaluation on MTBench+ (test)
81.56Retrospection ScoreMemRetrieval-ChatGPT + FragRel
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| MemRetrieval-ChatGPT + FragRelTemporary context length=2K, w_rel=0.8, alpha=0.52024.06 | 81.56 | 77.89 | 82.17 | 80.54 | |
| MemRetrieval-ChatGPTTemporary context length=2K2024.06 | 81.17 | 74.56 | 69.39 | 75.04 | |
| MemoChat-ChatGPTTemporary context length=2K2024.06 | 66.28 | 73.5 | 72.5 | 70.76 | |
| MPC-ChatGPTTemporary context length=2K2024.06 | 53 | 61.22 | 49.33 | 54.52 | |
| ChatGPT-2KTemporary context length=2K2024.06 | 52.11 | 55.33 | 48.22 | 51.89 | |
| MemoryBank-ChatGPTTemporary context length=2K2024.06 | 23.39 | 55.28 | 48.67 | 42.44 |