Multi-Objective Task on Multi-Objective Tasks
67.2Accuracy (2-obj)MemAct (SFT+RL) w/ Fixed Update
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| MemAct (SFT+RL) w/ Fixed UpdateBackbone=Qwen2.5-14B-Instruct, Training Protocol (SFT/RL)=SFT + RL, Memory Management Strategy=Static memory schedule, Update Interval Schedule=Fixed (5 turns)2025.10 | 67.2 | 58.1 | 55.7 | 51.6 | 58.2 | 7 | |
| MemAct (SFT+RL)Backbone=Qwen2.5-14B-Instruct, Training Protocol (SFT/RL)=SFT + RL, Memory Management Strategy=MemAct2025.10 | 66 | 59.1 | 57 | 54.3 | 59.1 | 8.2 | |
| MemAct (SFT)Backbone=Qwen2.5-14B-Instruct, Training Protocol (SFT/RL)=SFT, Memory Management Strategy=MemAct2025.10 | 65.2 | 54.7 | 52 | 49.3 | 55.3 | 9.4 | |
| Search-R1 (SFT + RL)Backbone=Qwen2.5-14B-Instruct, Training Protocol (SFT/RL)=SFT + RL2025.10 | 62.5 | 54.6 | 49.3 | 39.3 | 51.4 | 19.3 | |
| Qwen3-235BBackbone=Qwen3-235B2025.10 | 58 | 53.1 | 52.3 | 48.9 | 53.1 | 16.7 | |
| SummarizationBackbone=Qwen2.5-14B-Instruct, Memory Management Strategy=Summarization2025.10 | 54 | 49.8 | 49.5 | 47.1 | 50.1 | 4.1 | |
| Sliding WindowBackbone=Qwen2.5-14B-Instruct, Memory Management Strategy=Sliding Window2025.10 | 51.7 | 51.3 | 47.4 | 47.5 | 49.5 | 3.8 | |
| Qwen2.5-14BBackbone=Qwen2.5-14B-Instruct2025.10 | 51 | 46.6 | 42.2 | 37.8 | 44.4 | 6.5 | |
| MEM1 (RL)Backbone=Qwen2.5-14B-Instruct, Training Protocol (SFT/RL)=RL2025.10 | 49.4 | 43.5 | 35.7 | 29.3 | 43.5 | 5 | |
| A-MEMBackbone=Qwen2.5-14B-Instruct, Memory Management Strategy=A-MEM2025.10 | 43 | 38.5 | 36.4 | 33.9 | 39.9 | 3.9 |