Question Answering on HotPotQA (test) (Custom Metrics: Acc, Prec, Abs(0), Delta)
25.52AccuracyAFH (Absolute)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| AFH (Absolute)Backbone=Qwen2.5-7B-Base2025.10 | 25.52 | 36.73 | 40.3 | 27.93 | |
| DPOBackbone=Qwen2.5-7B-Base2025.10 | 24.35 | 48.25 | 72.8 | 60.4 | |
| 5-shot PromptingBackbone=Qwen2.5-7B-Base2025.10 | 23.06 | 34.17 | 39.45 | 24.18 | |
| MASH w/ OTC-STBackbone=Qwen2.5-7B-Base2025.10 | 20.98 | 60.67 | 87.14 | 64.51 | |
| MASH w/ OTC-STBackbone=Qwen3-4B-Base2025.10 | 20.96 | 53.59 | 81.63 | 67.07 | |
| AFH (Multisample)Backbone=Qwen2.5-7B-Base2025.10 | 17.68 | 51.36 | 82 | 53.22 | |
| 5-shot PromptingBackbone=Qwen3-4B-Base2025.10 | 17.58 | 34.37 | 59.76 | 43.72 | |
| DPOBackbone=Qwen3-4B-Base2025.10 | 16.72 | 60.84 | 92.38 | 75.87 | |
| AFH (Absolute)Backbone=Qwen3-4B-Base2025.10 | 13.44 | 47.97 | 82.81 | 41.5 | |
| AFH (Multisample)Backbone=Qwen3-4B-Base2025.10 | 7.25 | 66.54 | 96.54 | 36.51 | |
| OTCBackbone=Qwen2.5-7B-Base2025.10 | 0 | — | 100 | 0 | |
| OTCBackbone=Qwen3-4B-Base2025.10 | 0 | — | 100 | 0 |