Small bug in Showo.mmu_generate？

[CladernyJorn]

In line 221 of file .\models\modeling_showo.py, probably the shape of attention (L+1,L+1) instead of (1,1,L+1,L+1) mask is changed while iteration of generating new words. It can be fixed by adding codes like 'attention_mask = attention_mask_b.unsqueeze(0).unsqueeze(0)'.

[Sierkinhane]

Hi, I'm not sure if you have encountered errors when using the current version.

[CladernyJorn]

I encountered an error when using the current version for mmu: Attention mask should be of size (1, 1, 621, 621), but is torch.Size([621, 621]).