gokaygokay

/

Florence-2-SD3-Captioner

like 27

https://colab.research.google.com/drive/1hKDrJ5AH_o7I95PtZ9__VlCTNAo1Gjpf?usp=sharing

i used this notebook when fine-tuning. i just changed data accordingly. i used "DESCRIPTION" tag instead of "DocVQA" in this notebook.

And i mixed data from google/imageinwords, google/docci and ProGamerGov/synthetic-dataset-1m-dalle3-high-quality-captions

I selected 400-500 data from each of them. I also restricted character count to be between 600-1000 (not token count).

Incredible, thank you so much for your quick response!!

Thank you, really helpful.