---
language:
- en
- zh
size_categories:
- n<1K
task_categories:
- visual-question-answering
tags:
- multimodal
- LLM
- MLLM
---

### Oasis: One Image is All You Need for Multimodal Instruction Data Synthesis
This dataset contains **Oasis-500k** dataset. 

<a href="https://huggingface.co/papers/2503.08741">[Read the Paper]</a> &nbsp; | &nbsp; <a href="https://github.com/Letian2003/MM_INF">[Github Repo]</a>

![image](assets/method-cropped.png)

All images come from Cambrian-10M. Instructions and responses are generated by MLLM.