license: apache-2.0
task_categories:
- text-generation
language:
- en
- zh
size_categories:
- 1M<n<10M
Dataset Card for Oryx-SFT-Data
This dataset is used for the training of the Oryx model. We only allow the use of this dataset for academic research and education purpose.
For more details, please refer to our Github Repo
Dataset Details
A mixture of 1.2M image/video data.
For more statistics of the dataset, please refer to our paper (coming soon)
Source Data
We preprocess all the data into multiple files.
You can use our preprocessed data according to our Github Repo
Or you can download raw data alternatively.
The links are as below:
VideoChatGPT-Plus: https://huggingface.co/datasets/MBZUAI/VCG-plus_112K
LLaVA-Hound: https://huggingface.co/ShareGPTVideo/LLaVA-Hound-SFT
ShareGPT4Video: https://huggingface.co/datasets/ShareGPT4Video/ShareGPT4Video
Cinepile: https://huggingface.co/datasets/tomg-group-umd/cinepile
NExT-QA: https://github.com/doc-doc/NExT-QA
Perception Test: https://github.com/google-deepmind/perception_test
ScanQA: https://github.com/ATR-DBI/ScanQA