Relevance Assessment on T2Ranking (test)
52.1F1-Score (0)R³A
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| R³ATraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-7B, Parameter Update=w/ parameter update2025.08 | 52.1 | 34.3 | 66.6 | 67.5 | 67.8 | 53.9 | |
| Qwen2.5-7B + pretrainedTraining Strategy=Supervised Fine-Tuning, Parameter Update=w/ parameter update2025.08 | 51.2 | 31.6 | 65.3 | 66.9 | 67.5 | 51.4 | |
| R1Training Strategy=Reinforcement Learning, Backbone=Qwen2.5-1.5B, Parameter Update=w/ parameter update2025.08 | 51.1 | 25.4 | 63.6 | 65.6 | 64.8 | 47.7 | |
| GPT-4oMethod Family=UMbrela, Training Strategy=Prompting, Parameter Update=w/o parameter update2025.08 | 50.9 | 33.4 | 66.2 | 66.3 | 67.1 | 54.2 | |
| R1Training Strategy=Reinforcement Learning, Backbone=Qwen2.5-7B, Parameter Update=w/ parameter update2025.08 | 50.8 | 27.3 | 57.7 | 66.4 | 65.6 | 51.4 | |
| R³A-ZeroTraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-7B, Parameter Update=w/ parameter update2025.08 | 50.8 | 33.2 | 63.5 | 67.3 | 64.8 | 51.2 | |
| Qwen2.5-1.5B + pretrainedTraining Strategy=Supervised Fine-Tuning, Parameter Update=w/ parameter update2025.08 | 50.7 | 33.3 | 58.2 | 69.1 | 65.4 | 48.3 | |
| R³A-Distill-1.5BTraining Strategy=Distilling, Parameter Update=Online Serving2025.08 | 50.4 | 29.8 | 64.9 | 66.3 | 67.2 | 51.3 | |
| DeepSeek-V3Method Family=UMbrela, Training Strategy=Prompting, Parameter Update=w/o parameter update2025.08 | 49.8 | 32.7 | 65.8 | 66.5 | 67.6 | 52.3 | |
| R³ATraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-1.5B, Parameter Update=w/ parameter update2025.08 | 49.8 | 30.3 | 62.7 | 66.1 | 66.4 | 49 | |
| Qwen2.5-7BTraining Strategy=Supervised Fine-Tuning, Parameter Update=w/ parameter update2025.08 | 49.7 | 32.2 | 62.8 | 66.4 | 67.5 | 50.7 | |
| Qwen2.5-1.5BTraining Strategy=Supervised Fine-Tuning, Parameter Update=w/ parameter update2025.08 | 48.3 | 36 | 50.4 | 67.6 | 62.9 | 46.3 | |
| QwQ-32BMethod Family=UMbrela, Training Strategy=Prompting, Parameter Update=w/o parameter update2025.08 | 48.2 | 33.8 | 67.1 | 65.5 | 66.9 | 54 | |
| R³A-ZeroTraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-1.5B, Parameter Update=w/ parameter update2025.08 | 48 | 24.8 | 61.8 | 64.4 | 65 | 47.9 | |
| R1-ZeroTraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-7B, Parameter Update=w/ parameter update2025.08 | 48 | 24.8 | 61.8 | 64.4 | 65 | 49.9 | |
| R1-ZeroTraining Strategy=Reinforcement Learning, Backbone=Qwen2.5-1.5B, Parameter Update=w/ parameter update2025.08 | 47.9 | 26.8 | 62.3 | 65 | 64.1 | 46.2 | |
| Qwen2.5-7BMethod Family=UMbrela, Training Strategy=Prompting, Parameter Update=w/o parameter update2025.08 | 47.7 | 30.1 | 60 | 65.2 | 65.9 | 48.9 | |
| DeepSeek-R1Method Family=UMbrela, Training Strategy=Prompting, Parameter Update=w/o parameter update2025.08 | 46.5 | 31.9 | 66.9 | 64.3 | 63.2 | 53.6 |