Science Question Answering on ScienceQA curated subset (324 questions) (test)
86.46AccuracyWeather-R1-7B
Evaluation results
| Method | Links | |
|---|---|---|
| Weather-R1-7BType=Fine-tuned VLM, Paradigm=LoCo-RFT, Base Model=Qwen2.5-VL-7B-Instruct2026.01 | 86.46 | |
| Qwen2.5-VL-7BType=Zero-shot VLM2026.01 | 81.48 | |
| RFT-7BType=Fine-tuned VLM2026.01 | 79.54 | |
| SFT-7BType=Fine-tuned VLM2026.01 | 78.22 |