admin管理员组

文章数量:1023782

Hi I have two tensors:

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)

I want the result has a shape of (125, 8, 64)

My first observation is: last dimension of a match the first dimension of b then I do:

result = torch.matmul(a,b)

It gave me the error:

Expected size for first two dimensions of batch2 tensor to be: [128, 128] but got: [128, 8].

How can I do this.

Edit: I also dont' want to reshape into 2D and then reshape the result into 3D again.

Hi I have two tensors:

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)

I want the result has a shape of (125, 8, 64)

My first observation is: last dimension of a match the first dimension of b then I do:

result = torch.matmul(a,b)

It gave me the error:

Expected size for first two dimensions of batch2 tensor to be: [128, 128] but got: [128, 8].

How can I do this.

Edit: I also dont' want to reshape into 2D and then reshape the result into 3D again.

  • pytorch
  • matmul
Share Improve this question edited Nov 19, 2024 at 5:42 Dinosaur asked Nov 19, 2024 at 5:37 DinosaurDinosaur 254 bronze badges
Add a comment  | 

1 Answer 1

Reset to default 1

You can use an einsum

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)
c = torch.einsum('ij,jkl->ikl', a, b)
print(c.shape)
> torch.Size([125, 8, 64])

Hi I have two tensors:

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)

I want the result has a shape of (125, 8, 64)

My first observation is: last dimension of a match the first dimension of b then I do:

result = torch.matmul(a,b)

It gave me the error:

Expected size for first two dimensions of batch2 tensor to be: [128, 128] but got: [128, 8].

How can I do this.

Edit: I also dont' want to reshape into 2D and then reshape the result into 3D again.

Hi I have two tensors:

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)

I want the result has a shape of (125, 8, 64)

My first observation is: last dimension of a match the first dimension of b then I do:

result = torch.matmul(a,b)

It gave me the error:

Expected size for first two dimensions of batch2 tensor to be: [128, 128] but got: [128, 8].

How can I do this.

Edit: I also dont' want to reshape into 2D and then reshape the result into 3D again.

  • pytorch
  • matmul
Share Improve this question edited Nov 19, 2024 at 5:42 Dinosaur asked Nov 19, 2024 at 5:37 DinosaurDinosaur 254 bronze badges
Add a comment  | 

1 Answer 1

Reset to default 1

You can use an einsum

a = torch.randn(125, 128)    # Shape: (125, 128)
b = torch.randn(128, 8, 64)  # Shape: (128, 8, 64)
c = torch.einsum('ij,jkl->ikl', a, b)
print(c.shape)
> torch.Size([125, 8, 64])

本文标签： pytorchtorch matmul between 2D and 3D tensorStack Overflow

版权声明：本文标题：pytorch - torch matmul between 2D and 3D tensor - Stack Overflow 内容由热心网友自发贡献，该文观点仅代表作者本人， 转载请联系作者并注明出处：http://it.en369.cn/questions/1745580239a2157265.html， 本站仅提供信息存储空间服务，不拥有所有权，不承担相关法律责任。如发现本站有涉嫌抄袭侵权/违法违规的内容，一经查实，本站将立刻删除。