How is the flow model updated in M-step?

[tobylyf]

In iteration t (t>1), is the model fine-tuned based on C+T pre-trained model, or the "iteration t-1" model?

[coolbeam]

In iteration t (t>1), is the model fine-tuned based on C+T pre-trained model, or the "iteration t-1" model?

Hi, in iteration t (t>1), we finetune the model based on the 'iteration t-1' model.

[Eryo-iPython]

Using C+T to generate dataset. Should I load C+T as initialization to train the model using generated data?