Pytorch expects each tensor to be equal size

user_18121469 0
Viewed 6686

When running this code: embedding_matrix = torch.stack(embeddings)

I got this error:

RuntimeError: stack expects each tensor to be equal size, but got [7, 768] at entry 0 and [8, 768] at entry 1

I'm trying to get embedding using BERT via:

    split_sent = sent.split()
    tokens_embedding = []
    j = 0
    for full_token in split_sent:
        curr_token = ''
        x = 0
        for i,_ in enumerate(tokenized_sent[1:]): 
            token = tokenized_sent[i+j]
            piece_embedding = bert_embedding[i+j]
            if token == full_token and curr_token == '' :
               tokens_embedding.append(piece_embedding)
               j += 1
               break                                     
    sent_embedding = torch.stack(tokens_embedding)
    embeddings.append(sent_embedding)
embedding_matrix = torch.stack(embeddings)

Does anyone know how I can fix this?

bert-language-model tensor pytorch python
1 Answers

As per PyTorch Docs about torch.stack() function, it needs the input tensors in the same shape to stack. I don't know how will you be using the embedding_matrix but either you can add padding to your tensors (which will be a list of zeros at the end till a certain user-defined length and is recommended if you will train with this stacked tensor, refer this tutorial) to make them equidimensional or you can simply use something like torch.cat(data,dim=0).

user_98067170
Related