Skip to content

Conversation

@Kin-Zhang
Copy link

No error info will pop up on the previous version.

Need specific the len in the dataset otherwise it will out of index.

@Kin-Zhang
Copy link
Author

btw, another point for this sentence in README:
You may achieve better model performance by adjusting hyper-parameters such as the learning rate and increasing the batch size.

I think we need be careful about the batch size since the diffsynth only support bz=1 right now, some discussion can be found Wan2.1 finetuning script seems to only support bs = 1

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant