Question Answering on LoCoMo (Accuracy and Context Token Usage)
30Accuracynaive_rag
Evaluation results
| Method | Links | ||
|---|---|---|---|
| naive_ragRAG strategy=naive2026.06 | 30 | — | |
| advanced_ragRAG strategy=advanced2026.06 | 30 | — | |
| temporal_v6Model version=v6, Temporal processing=true2026.06 | 30 | — | |
| Full context†Policy=No eviction, Retention budget=None2026.06 | 23.3 | 18.8 | |
| LRE-supervisedRetention budget=20%, Evaluation protocol=token-F12026.06 | 17.7 | 6 | |
| LRE-self-supRetention budget=20%, Evaluation protocol=token-F12026.06 | 16 | 6.5 | |
| v6Model version=v62026.06 | 13 | — | |
| temporal_v6_lossyModel version=v6, Temporal processing=true, Compression status=lossy2026.06 | 13 | — | |
| RecencyRetention budget=20%, Evaluation protocol=token-F12026.06 | 10 | 3.6 | |
| Content salienceRetention budget=20%, Evaluation protocol=token-F12026.06 | 8.1 | 4 |