The resulting dataset contains 8000 samples of the openbmb/RLAIF-V-Dataset.
The original RLAIF-V-Dataset is a visual preference learning dataset containing images paired with a question, a chosen answer, and a rejected answer. This split of an even smaller subset is provided for very fast experimentation and evaluation of models when computational resources are highly limited or for quick prototyping.
The dataset is provided as a… See the full description on the dataset page:
https://huggingface.co/datasets/Vishva007/RLAIF-V-Dataset-8k.
