hypjudy/sparkles

Sparkles: Unlocking Chats Across Multiple Images for Multimodal Instruction-Following Models

sparkles's Issues

Slow Download Speed

Dear author, thank you very much for your outstanding work. I would like to train our model using the Sparkles dataset, but the file download speed from OneDrive is very slow because the files you uploaded are scattered. If you could upload a .zip compressed file, our download speed would be much faster.

How to finetune LLaVA with your data

Thank you for your great work on allowing multiple images in multimodal model!

In your paper, you mentioned about finetuning LLaVA v1.5 with SparklesDialog. Could you share how to reoganize the input of LLaVA for multiple images? As the origin LLaVA implementation requires to move the image to the very front. Thank you!

Error on arxiv

The label for the BISON - Statement demo on page 18 should be 2

Questions about the evaluation of SparklesChat versus Minigpt-4

To the best of my knowledge, MiniGPT-4 can only process a single image at a time. Could you please explain how you managed to get it to handle multiple images simultaneously?

The Minigpt-4 url: https://huggingface.co/spaces/Vision-CAIR/minigpt4