Swift now supports inference, training, and deployment of the Florence models.
#14
by
hu5enpai
- opened
https://github.com/modelscope/swift
For more information, please refer to our document
@hu5enpai Hi, does it support continuous training on new multi-langual data such as WuKong specific for Chinese?
Due to the limitations of the Florence model, it is necessary to use a dataset that includes images for training. If you have the corresponding
dataset, you can refer to our documentation to train with a custom dataset.
What I mean, it's actually images data.. Wukong is a multi-modal dataset. The question is, how to avoid it forgeting old knowledge since we do not have microsoft 126M data