Multi-hop Question Answering on NQ (Accuracy)
35.18AccuracyQwen2.5-7B-Instruct + MemTrain + MEM1
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-7B-Instruct + MemTrain + MEM1Backbone=Qwen2.5-7B-Instruct, Memory Tool=MEM1, Training=MemTrain2026.06 | 35.18 | |
| Qwen3-4B-Instruct-2507 + MemTrain + MEM1Backbone=Qwen3-4B-Instruct-2507, Memory Tool=MEM1, Training=MemTrain2026.06 | 32.24 | |
| Qwen2.5-7B-Instruct + MEM1Backbone=Qwen2.5-7B-Instruct, Memory Tool=MEM12026.06 | 24.21 | |
| Qwen3-4B-Instruct-2507 + MEM1Backbone=Qwen3-4B-Instruct-2507, Memory Tool=MEM12026.06 | 21.97 | |
| Qwen3-4B-Instruct-2507Backbone=Qwen3-4B-Instruct-25072026.06 | 19.92 | |
| Qwen2.5-7B-InstructBackbone=Qwen2.5-7B-Instruct2026.06 | 12.22 |