Multi-hop Question Answering on HotpotQA (Size-Cut Evaluation)
79.69Score (7k)Qwen3-4B-Instruct + MemTrain + MemAgent
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Qwen3-4B-Instruct + MemTrain + MemAgentBackbone=Qwen3-4B-Instruct, Training Stage=MemTrain + MemAgent2026.06 | 79.69 | 73.44 | 75.78 | 73.44 | 68.75 | 67.19 | 61.72 | 62.5 | 70.31 | |
| Qwen2.5-7B-Instruct + MemTrain + MemAgentBackbone=Qwen2.5-7B-Instruct, Training Stage=MemTrain + MemAgent2026.06 | 76.56 | 79.69 | — | 75 | 70.31 | 75.78 | 64.84 | 68.75 | 73.53 | |
| Qwen3-4B-Instruct + MemAgentBackbone=Qwen3-4B-Instruct, Training Stage=MemAgent2026.06 | 70.31 | 64.06 | 71.88 | 62.5 | 64.84 | 66.41 | 64.06 | 57.03 | 65.14 | |
| Qwen2.5-7B-Instruct + MemAgentBackbone=Qwen2.5-7B-Instruct, Training Stage=MemAgent2026.06 | 64.06 | 67.19 | 62.5 | 59.38 | 55.47 | 50 | 46.88 | 41.41 | 55.86 | |
| Qwen3-4B-Instruct + MemTrainBackbone=Qwen3-4B-Instruct, Training Stage=MemTrain2026.06 | 63.28 | 60.16 | 60.16 | 57.03 | 60.94 | 58.59 | 48.44 | 40.62 | 56.15 | |
| Qwen2.5-7B-Instruct + MemTrainBackbone=Qwen2.5-7B-Instruct, Training Stage=MemTrain2026.06 | 59.38 | 55.47 | 48.44 | 46.09 | 42.19 | 38.28 | 39.84 | 33.59 | 45.41 | |
| Qwen3-4B-InstructBackbone=Qwen3-4B-Instruct2026.06 | 57.81 | 51.56 | 34.38 | 10.94 | 8.59 | 4.69 | 3.91 | 3.91 | 21.97 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2026.06 | 53.12 | 51.56 | 35.16 | 13.28 | 10.16 | 1.56 | 1.56 | 0 | 20.8 |