Polarization detection on POLAR English (dev)
78.93F1 ScoreSFT + DPO
Evaluation results
| Method | Links | |
|---|---|---|
| SFT + DPOBackbone=Qwen2.5-7B-Instruct, Refinement=Direct Preference Optimization (DPO)2026.04 | 78.93 | |
| SFTBackbone=Qwen2.5-7B-Instruct, Fine-tuning=Structured LoRA2026.04 | 73.8 | |
| SLMs2026.04 | 71.49 | |
| Zero-shot baselineSetting=Zero-shot2026.04 | 71.05 |