def input_fn()

in code/train_deploy.py [0:0]

• 23 lines of code
• 10 McCabe index (conditional complexity)

def input_fn(request_body, request_content_type):
    """An input_fn that loads a pickled tensor"""
    if request_content_type == "application/json":
        data = json.loads(request_body)
        print("================ input sentences ===============")
        print(data)
        
        if isinstance(data, str):
            data = [data]
        elif isinstance(data, list) and len(data) > 0 and isinstance(data[0], str):
            pass
        else:
            raise ValueError("Unsupported input type. Input type can be a string or an non-empty list. \
                             I got {}".format(data))
                       
        #encoded = [tokenizer.encode(x, add_special_tokens=True) for x in data]
        #encoded = tokenizer(data, add_special_tokens=True) 
        
        # for backward compatibility use the following way to encode 
        # https://github.com/huggingface/transformers/issues/5580
        input_ids = [tokenizer.encode(x, add_special_tokens=True) for x in data]
        
        print("================ encoded sentences ==============")
        print(input_ids)

        # pad shorter sentence
        padded =  torch.zeros(len(input_ids), MAX_LEN) 
        for i, p in enumerate(input_ids):
            padded[i, :len(p)] = torch.tensor(p)
     
        # create mask
        mask = (padded != 0)
        
        print("================= padded input and attention mask ================")
        print(padded, '\n', mask)

        return padded.long(), mask.long()
    raise ValueError("Unsupported content type: {}".format(request_content_type))