the structure or an example of the data used for training BAGEL on World Navigation

Thanks for sharing your brilliant work!

**Could you please show the structure or an example of the data used for training BAGEL on World Navigation?**

I'm currently planning to fine-tune BAGEL's thinking module with my own dataset. My data is similar to the World Navigation task presented in your paper, except that it includes Chain-of-Thought (CoT) annotations. Specifically, in my dataset, each sample consists of a single input image and an instruction, paired with multiple output images and a corresponding CoT reasoning process.