Persuasiveness classification on PVP (test)
76.6Balanced AccuracyReasoning-SFT
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Reasoning-SFTStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=Supervised Rationale2026.05 | 76.6 | 78.4 | 74.8 | 76.6 | |
| Reasoning-SFTStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=Supervised Rationale2026.05 | 74.6 | 74.6 | 74.5 | 74.6 | |
| SFTStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=No Rationale2026.05 | 73.1 | 72.6 | 76.6 | 74.5 | |
| SFTStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=No Rationale2026.05 | 72.1 | 69.9 | 80.4 | 74.8 | |
| GRPOStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 71.6 | 69.7 | 79.4 | 74.2 | |
| GRPOStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 66.7 | 71.8 | 57 | 63.5 | |
| BaseStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=No Rationale2026.05 | 65.7 | 63.6 | 78.5 | 70.3 | |
| GRPO-JointStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 65.1 | 62.1 | 84.1 | 71.4 | |
| Base-ReasoningStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 63.2 | 61 | 80.4 | 69.4 | |
| BaseStudent Model=Qwen2.5-VL-7B-Instruct, Rationale Category=No Rationale2026.05 | 62.7 | 60.3 | 81.5 | 69.3 | |
| GRPO-JointStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 60.1 | 57.5 | 89.7 | 70.1 | |
| Base-ReasoningStudent Model=Phi-3.5-Vision-Instruct, Rationale Category=Unsupervised Rationale2026.05 | 59.9 | 58.3 | 78.5 | 66.9 |