Closed dailingx closed 6 months ago
You can check the issue above, in addition, you can select the parameters you want to train to reduce memory usage. By default, the entire unet is trainable.
# for name, para in unet.named_parameters():
# if 'temporal_transformer_block' in name and 'down_blocks' in name:
# parameters_list.append(para)
# para.requires_grad = True
# else:
# para.requires_grad = False
# optimizer = optimizer_cls(
# parameters_list,
# lr=args.learning_rate,
# betas=(args.adam_beta1, args.adam_beta2),
# weight_decay=args.adam_weight_decay,
# eps=args.adam_epsilon,
# )
First of all, thank you very much for sharing the code. At the same time, I would like to ask how much memory is needed when training SVD, and what is the minimum video memory configuration required?