Language Reasoning on DeepAccident-CCoT (val)
84.2AccuracyC-CoT
Evaluation Results
| Method | Links | |
|---|---|---|
| C-CoTBackbone=Qwen2.5-VL (7B), Evaluation Protocol=Fine-tuned (LoRA)2026.05 | 84.2 | |
| Qwen2.5-VLModel Size=7B, Evaluation Protocol=Zero-shot2026.05 | 75.6 | |
| DeepSeek-VLModel Size=7B, Evaluation Protocol=Zero-shot2026.05 | 74.1 | |
| InternVL-2.5Model Size=8B, Evaluation Protocol=Zero-shot2026.05 | 72.4 | |
| Llama-3.2-VisionModel Size=11B, Evaluation Protocol=Zero-shot2026.05 | 69.8 | |
| LLaVA-1.5Model Size=7B, Evaluation Protocol=Zero-shot2026.05 | 63.5 |