Precompute embeddings
isamu-isozaki opened this issue · 2 comments
isamu-isozaki commented
Precompute embeddings
pcuenca commented
Is the idea here to pre-compute the encodings of the dataset with the VQGAN encoder? If so, I've done similar things in the past, could probably help.
isamu-isozaki commented
@pcuenca Yup exactly! We are thinking of doing this with t5/VQGAN in the training/train_muse.py. Any help will be awesome!