Long-Term Conversational Memory Retrieval on LongMemEval
92.9R@5RRF
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| RRFLLM answerer=gpt-4o-mini, LLM judge=gpt-4o-mini, context_window=top-52026.05 | 92.9 | 97.7 | 46.8 | 48.6 | 62.5 | 66.1 | 24.8 | 31 | |
| agent_rrfrecency_bonus=additive, LLM answerer=gpt-4o-mini, LLM judge=gpt-4o-mini, context_window=top-52026.05 | 92.4 | 97.8 | 47.4 | 48.6 | 62.7 | 66 | 25.4 | 31.2 | |
| BM25LLM answerer=gpt-4o-mini, LLM judge=gpt-4o-mini, context_window=top-52026.05 | 90.9 | 94.6 | 46.6 | 48.4 | 62 | 65.3 | 24.6 | 30.6 | |
| Densebackbone=BGE-small, LLM answerer=gpt-4o-mini, LLM judge=gpt-4o-mini, context_window=top-52026.05 | 88.6 | 94.6 | 45.6 | 47.8 | 59.8 | 64.7 | 23.6 | 29.8 |