---
license: apache-2.0
datasets:
- HuggingFaceM4/the_cauldron
---

Fine-tune of https://huggingface.co/vikhyatk/moondream2 on a subset of the Cauldron, designed to improve visual question answering and reading of text off of natural images. 

This is a WIP, and the model versions available may change with commits. Still figuring out what the best subset is to make this as useful as possible for real world scenarios.

This small model is able to be hosted on smaller hardware, such as a Raspberry Pi.

More context on the model training can be found on the WandB logs and Git repo. 

https://wandb.ai/noahpunintended/moondream-ft-picorder?nw=nwusernoahpunintended

https://github.com/nkasmanoff/pi-card