abs_pe size mismatch when loading LARP_AR from hywang66/LARP-L-long-AR for video generation

[CCdydy]

Hi, thank you for releasing this great work!

I'm trying to run video generation with the model hywang66/LARP-L-long-AR using the official sample.py script. However, when loading the model from HuggingFace Hub, I encountered the following error:

RuntimeError: Error(s) in loading state_dict for LARP_AR: size mismatch for abs_pe: copying a param with shape torch.Size([1, 1024, 1280]) from checkpoint, the shape in current model is torch.Size([1, 1025, 1280]). It seems the positional embedding abs_pe in the checkpoint does not match the model definition. I'm using the default command provided in your README, and haven't changed any model structure.

Is there a recommended way to safely load the model for video generation?

Should we manually remove or adjust abs_pe when loading?

Thanks in advance!

[ThisisBillhe]

Hi, thanks for your interest. Please make sure you download the correct checkpoint from here rather than hywang66/LARP-L-long-AR.

[CCdydy]

Thank you for your prompt and detailed response. Your explanation helped me successfully resolve the issue!