Huggingface BERT SequenceClassification - ValueError: too many values to unpack (expected 2)

Question

The class labels for the two class model is 0, 1, 0, 0, etc. There is only one label per input sequence. The labels are set in a python list and converted to torch.Tensor. (reading from a csv file - straight forward).

its value: tensor(0) its shape is shown as: torch.Size([])

when i create the batch, etc. and set the parameters in the model (huggingface sequenceclassification model)

for each batch iteration,

   b_input_ids = batch[0].to(device)
   b_input_mask = batch[1].to(device)
   b_token_type_ids = batch[2].to(device)
   b_labels = batch[3].to(device)

and then set to the model (as part of training)

loss, logits = model(b_input_ids, 
                     attention_mask=b_input_mask, 
                     token_type_ids=b_token_type_ids,
                     labels=b_labels)

The error message is as follows:

    ValueError                                
    
        Traceback (most recent call last)
        
        <ipython-input-17-b3a36df2f659> in <module>()
             89                              attention_mask=b_input_mask,
             90                              token_type_ids=b_token_type_ids,
        ---> 91                              labels=b_labels)
             92 
             93         # Accumulate the training loss over all of the batches so that we can
        
ValueError: too many values to unpack (expected 2)

This was code I was able to use and train a few days ago. (I have a saved model). The error message (I assume ) probably has something to do with the size etc.

Has anyone hit a similar problem? thanks.