Retrieval on LongMemEval-S session granularity v1
98.7R@5LETHE v1
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| LETHE v1Question type=knowledge-update, N=78, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 98.7 | — | — | |
| LETHE v1Question type=single-session-assistant, N=56, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 96.4 | — | — | |
| LETHE v1Question type=multi-session, N=133, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 94 | — | — | |
| LETHE v1Question type=Overall (micro), N=500, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 93.8 | 79.6 | 98.2 | |
| LETHE v1Question type=temporal-reasoning, N=133, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 92.5 | — | — | |
| LETHE v1Question type=single-session-preference, N=30, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 90 | — | — | |
| LETHE v1Question type=single-session-user, N=70, Granularity=session, Retrieval Pipeline=BM25 (FTS5) + dense ANN (sqlite-vec), Embeddings=MiniLM-L6-v2, Fusion=reciprocal rank fusion, Fine-tuning=none2026.06 | 90 | — | — |