metadata

language:
  - zh
license: apache-2.0

Mengzi-oscar-base (Chinese Multi-modal pre-training model)

Mengzi-oscar is trained based on the Multi-modal pre-training model Oscar, and is initialized using Mengzi-Bert-Base. 3.7M pairs of images and texts were used, including 0.7M Chinese image-caption pairs, 3M Chinese image-question pairs, a total of 0.22M different images.

Usage

Installation

Check INSTALL.md for installation instructions.

Pretrain & fine-tune

See the Mengzi-Oscar.md for details.

Citation

If you find the technical report or resource is useful, please cite the following technical report in your paper.

example