Efficient way to create block diagonal with different block size

deployment
1

I want to create a mask matrix through torch.block_diag() method. However, each diagonal block has a different shape.

torch.block_diag(torch.ones((2,2)), torch.ones((3,3)), torch.ones((2,2)))

Please can I know if there is an efficient way to achieve this?

2

Your code works fine so I’m unsure I understand the issue correctly. Are you seeing any errors?