When I tried to train the model with a batch size larger than 1, the UNET would raise bugs for shape issues due to the input. Based on my debug process, I found that this is because of encoder_hidden_states and sigmas shape issues. After modifying these two variables, the bug doesn't exist.
When I tried to train the model with a batch size larger than 1, the UNET would raise bugs for shape issues due to the input. Based on my debug process, I found that this is because of encoder_hidden_states and sigmas shape issues. After modifying these two variables, the bug doesn't exist.