---
license: mit
---

Dataset proposed by paper: https://arxiv.org/abs/2402.16641. Converted to LLaVA format.
We will organize more details on how to train with this special dataset (with multi-image instruction-response pairs) soon.