Question about finetuning the model #122
Description
Hi, thank you for the outstanding work!
I have some question about finetuning the model. If I retrain a BitVAE, should I train the Infinity from scratch? Or is it possible to finetune the Infinity with a different BitVAE?
What's more, can you provide some detail information about finetuning, such as how many samples I need to prepare for finetuning the Infinity 2B model, how many GPUs resources, iterations and batchsize.
Thank you.