Open double-fire-0 opened 5 months ago
Hello, thanks for your attention and suggestions. We did not load the optimizer during the resuming process, as we usually find that it only takes a few optimization steps to generate normally. You can also implement it with some simple modifications.
The training is break during my training process. Does this code support resume training?
I notice that the code can resume model at https://github.com/ali-vilab/VGen/blob/7ad0f25df50b2c35d7eb95cbabdf772c5b9761c8/tools/train/train_t2v_enterance.py#L154 but it seems that the code doesn't resume the optimizer and scheduler ?