ByungKwanLee / Meteor

Official PyTorch implementation code for realizing the technical part of Mamba-based traversal of rationale (Meteor) to improve performance of numerous vision language performances for diverse capabilities. (Under Review)
MIT License
97 stars 4 forks source link

Accessing curated dataset with GPT4v filter #4

Open RifleZhang opened 3 months ago

RifleZhang commented 3 months ago

How do we access the data in the section 3 after curating rationale? The current Huggingface data doesn't seem to have rationales generated,

RifleZhang commented 3 months ago

Actually, I found your 1 M dataset at https://huggingface.co/datasets/BK-Lee/Meteor?row=4.

In the training data section on project page, do you mean you only need the image from the sources, but the instructions are replaced with ones at your huggingface repo?

Thanks for clarification!

ByungKwanLee commented 3 months ago

image

I am confused with your question!

There are rationales..