处理图文数学题的完整工作流程如下:
- Data preparation
创建JSONL格式文件,包含:image_urls
字段指向本地图片路径(如file:///home/user/question.jpg
)conversations
字段存储问题描述(如’求图中阴影部分面积’)
- model-based reasoning
execute a commandpython scripts/inference.py --model MM-Eureka-8B --data problem.jsonl
- 结果解析
Example output:<think>识别出图形由1个矩形和2个半圆组成...</think>
<answer>38.5cm²</answer>
caveat:当处理三维几何等复杂图形时,建议使用38B模型并启用--enable_reflection
参数激活反思功能。
This answer comes from the articleMM-EUREKA: A Multimodal Reinforcement Learning Tool for Exploring Visual ReasoningThe