Summarization on TL;DR (Win Rate variants)
83.4Win Rate (0% Flip Rate)α-URM
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| α-URMModels=Llama-3.2-3B2026.07 | 83.4 | 81.9 | 76.1 | 79.9 | |
| α-URMModels=Qwen-3-1.7B2026.07 | 82.9 | 79.2 | 68.9 | 69.6 | |
| URMModels=Llama-3.2-3B2026.07 | 82.8 | 83.2 | 77.5 | 79.4 | |
| URMModels=Qwen-3-1.7B2026.07 | 81.5 | 80 | 71.6 | 70.1 | |
| RMModels=Qwen-3-1.7B2026.07 | 81.3 | 76.2 | 63.3 | 63.9 | |
| cRMModels=Llama-3.2-3B2026.07 | 80.8 | 76.5 | 69.9 | 75.7 | |
| cRMModels=Qwen-3-1.7B2026.07 | 80.4 | 76.6 | 66.4 | 65 | |
| RMModels=Llama-3.2-3B2026.07 | 79.4 | 76.4 | 69.1 | 75.5 | |
| rRMModels=Llama-3.2-3B2026.07 | 77.3 | 77.7 | 68.6 | 75.7 | |
| rRMModels=Qwen-3-1.7B2026.07 | 76.8 | 73.6 | 64.7 | 65.9 |