ValueError due to Mismatch in Tensor Shapes when Loading Model

#3
by Nikuson - opened

Dear Sa2VA Developers,

I am encountering a ValueError when attempting to load the model from the following path: ByteDance/Sa2VA-1B.

Here is the error message:
ValueError: Trying to set a tensor of shape torch.Size([151679, 896]) in "weight" (which has shape torch.Size([151658, 896])), this looks incorrect.

It seems that the model weights have a shape mismatch, causing the error during the loading process. I have verified the model and weights, and this issue appears to be related to an inconsistency between the expected and actual tensor shapes.

I would greatly appreciate it if you could provide any guidance on how to resolve this issue.

I am also getting the same error specifically on 1B but not the other models

same error here

Hi @Nikuson @DenC123 @chux0519 ,

Thanks for your reporting. We have fixed this bug.

Sign up or log in to comment