miemiebabu/PhysCoT
收藏Hugging Face2026-03-23 更新2026-03-29 收录
下载链接：
https://hf-mirror.com/datasets/miemiebabu/PhysCoT
下载链接
链接失效反馈官方服务：
资源简介：
# PhysCoT: Physics Chain-of-Thought Reasoning Dataset
Large-scale dataset for training VLMs on physics reasoning with step-by-step solutions.
## Current Size
- Train: 1,887 entries (v0.1 — ScienceQA + MMMU physics)
- Val: 209 entries
- Scaling to 100K with GPT-4o distillation
## Sources
- ScienceQA physics (elementary level, with images)
- MMMU Physics (university level, with images)
- More sources coming: SciBench, OlympiadBench, GPT-4o generated
## Baseline Results (PhyX testmini)
- Qwen2.5-VL-7B (zero-shot): **37.0%**
- GPT-4o (reported): 32.5%
- GPT-5 (reported): 65.2%
## Format
JSONL with fields:
- : path to image file
- : [{from: human, value: question}, {from: gpt, value: CoT solution}]
- : {source, difficulty, domain}
## License
Apache 2.0
提供机构：
miemiebabu
