Can not use trained model because of assertion mistakes

[nva001]

Thank you so much for the wonderful model.
But there has been an issue related to using the model after fine-tuning.
After file-tuning, the model was saved by running torch.save(pipeline.state_dict(), /path/..). But when using it to do prediction tasks, it raised this error: AssertionError: Cell type annotation pipeline does not support zero shot setting. Please fine-tune the model on downstream datasets before inference.
When I read your source code, I see this mistake: the class was not initialized self.fitted attribute in the init method, the fitted attribute only would be created and assign True in the object derived from this class after running fit method. So when initializing a new object, the fitted attribute could not be created except running fit method. This led to the input of assert always being False. And, I can not bypass this assertion by assigning True to this attribute.



Could you please fix this issue?

[wehos]

Thanks for raising this issue! I'll work on it shortly and keep you posted.