Commentary Generation on Fighting
81AlignmentQwen2.5-VL-7B-Instruct
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Qwen2.5-VL-7B-InstructDecoding Strategy=Feedback2026.03 | 81 | 13 | 2.3 | |
| Qwen2.5-VL-7B-InstructDecoding Strategy=stateless2026.03 | 78 | 13 | 0.18 | |
| LLaVA-NeXT-Video-7B-hfDecoding Strategy=Feedback (ICL)2026.03 | 30 | 23 | 0.7 | |
| GPT-4.1Decoding Strategy=Realtime2026.03 | 27 | 31 | 0.4 | |
| Qwen2.5-VL-7B-InstructDecoding Strategy=Feedback (ICL)2026.03 | 24 | 28 | 1.02 | |
| Qwen2.5-VL-7B-InstructDecoding Strategy=Realtime2026.03 | 22 | 22 | 0.98 | |
| LLaVA-NeXT-Video-7B-hfDecoding Strategy=Feedback2026.03 | 21 | 30 | 1 | |
| GPT-4.1Decoding Strategy=Feedback2026.03 | 20 | 28 | 3.2 | |
| GPT-4.1Decoding Strategy=Feedback (ICL)2026.03 | 20 | 30 | 3.8 | |
| LLaVA-NeXT-Video-7B-hfDecoding Strategy=stateless2026.03 | 18 | 28 | 4.9 | |
| LLaVA-NeXT-Video-7B-hfDecoding Strategy=Realtime2026.03 | 18 | 30 | 0.6 | |
| GPT-4.1Decoding Strategy=stateless2026.03 | 18 | 29 | 6.3 |