Torch.bmm Doubt

1

I am trying to multiply a 2 dimensional matrix with a 4 dimensional tensor, but torch.bmm doesn’t seem to allow 4-dim tensors.
This is the code I am using:

batch_size = 40
max_len = 96

logits = torch.randn(batch_size, 3)
outs = torch.randn(batch_size, 3, max_len, 2)

# Shape of desired output: (batch_size, max_len, 2)
result = torch.bmm(logits, outs)

The error is:

RuntimeError: Expected 3-dimensional tensor, but got 4-dimensional tensor for argument #1 'batch1' (while checking arguments for bmm)

Is this possible any way?

2

No It is not possible .Refer documentation

It is clearly documented there.

4

Yeah I read it, but is there any kind of workaround?