CODEWITHSUNDEEP
pytorchsumtensortorch
User
User

Reputation: 23

how can i sum the size of this tensors?

I have different sizes or shapes for each tensor like

torch.Size([1, 12, 1000])
torch.Size([1, 12, 1000])
torch.Size([1, 10, 1000])
torch.Size([1, 11, 1000])
torch.Size([1, 11, 1000])
torch.Size([1, 15, 1000])
torch.Size([1, 10, 1000])
....

and need to be like torch.Size((12+12+10+11+11+15+ .... ),1000)

my code is

def extract():
     for y in sentences:
            tokenized_text = tokenizer.tokenize(y)
            indexed_tokens = tokenizer.convert_tokens_to_ids(tokenized_text)        
            tokens_tensor = torch.tensor([indexed_tokens])
            model.eval()
            outputs = model(tokens_tensor)
            hidden_states = outputs[2]
            my_tensor = torch.cat([hidden_states[i] for i in [-1,-2,-3,-4]], dim=-1)
     return my_tensor

Upvotes: 0

Views: 852

Answers (1)

Neb
Neb

Reputation: 2280

Concatenate them:

tensors = [t1, t2, t3, ...]
result = torch.cat(tensors, dim=1)

# result.size(): torch.Size([1, 12+12+10+..., 1000])

If you also want to remove the first dimension, as it has size 1:

result = result.squeeze()
# result.size(): torch.Size([12+12+10+..., 1000])

Upvotes: 1

Related Questions