KaraKaraWitch/wii

Quality of Youtube 1 & 2, baked in a portable format for vision, audio and textual learning.

"But WHY?"

The answer is a "Why not".

Plus this could be a good benchmark for Google/OpenAI/etc to try and beat it.

Good luck!

Rough Idea of using this on benchmarking

1. Do not use the dataset for training
2. The model must be able to accept visual and audio input
3. When asking to describe the video, it should not produce any "made up" or "wrong / incorrect" facts.

License

N/A

Citation Infomation

Putting it because I know someone is gonna ask eventually.

@dataset{wiidataset,
  title = {The Wii Dataset},
  year = {2024},
  publisher = {KaraKaraWitch},
  url = {<https://huggingface.co/datasets/KaraKaraWitch/wii>}
}