handler.py

from typing import Dict, List, Any
from transformers import BertForQuestionAnswering, BertTokenizer
import torch

# set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# def print_tokens_with_ids(tokenizer, input_ids):
#     # BERT only needs the token IDs, but for the purpose of inspecting the 
#     # tokenizer's behavior, let's also get the token strings and display them.
#     tokens = tokenizer.convert_ids_to_tokens(input_ids)
#     # For each token and its id...
#     for token, id in zip(tokens, input_ids):
#         # If this is the [SEP] token, add some space around it to make it stand out.
#         if id == tokenizer.sep_token_id:
#             print('')
#         # Print the token string and its ID in two columns.
#         print('{:<12} {:>6,}'.format(token, id))
#         if id == tokenizer.sep_token_id:
#             print('')

def get_segment_ids_aka_token_type_ids(tokenizer, input_ids):
    # Search the input_ids for the first instance of the `[SEP]` token.
    sep_index = input_ids.index(tokenizer.sep_token_id)
    # The number of segment A tokens includes the [SEP] token istelf.
    num_seg_a = sep_index + 1
    # The remainder are segment B.
    num_seg_b = len(input_ids) - num_seg_a
    # Construct the list of 0s and 1s.
    segment_ids = [0]*num_seg_a + [1]*num_seg_b
    # There should be a segment_id for every input token.
    assert len(segment_ids) == len(input_ids), \
        'There should be a segment_id for every input token.'
    return segment_ids

def to_model(
    model: BertForQuestionAnswering, 
    input_ids, 
    segment_ids
) -> tuple:
    # Run input through the model.
    output = model(
        torch.tensor([input_ids]), # The tokens representing our input text.
        token_type_ids=torch.tensor([segment_ids])
    )
    # print(output)
    # print(output.start_logits)
    # print(output.end_logits)
    # print(type(output))
    # The segment IDs to differentiate question from answer_text
    return output.start_logits, output.end_logits
    #output.hidden_states
    #output.attentions
    #output.loss

def get_answer(
    start_scores, 
    end_scores, 
    input_ids,
    tokenizer: BertTokenizer
) -> str:
    '''Side Note: 
    - It’s a little naive to pick the highest scores for start and end–what if it predicts an end word that’s before the start word?! 
    - The correct implementation is to pick the highest total score for which end >= start.
    '''
    # Find the tokens with the highest `start` and `end` scores.
    answer_start = torch.argmax(start_scores)
    answer_end = torch.argmax(end_scores)

    # Combine the tokens in the answer and print it out.
    # answer = ' '.join(tokens[answer_start:answer_end + 1])
    # Get the string versions of the input tokens.
    tokens = tokenizer.convert_ids_to_tokens(input_ids)
    # Start with the first token.
    answer = tokens[answer_start]
    # print('Answer: "' + answer + '"')
    # Select the remaining answer tokens and join them with whitespace.
    for i in range(answer_start + 1, answer_end + 1):
        # If it's a subword token, then recombine it with the previous token.
        if tokens[i][0:2] == '##':
            answer += tokens[i][2:]
        # Otherwise, add a space then the token.
        else:
            answer += ' ' + tokens[i]
    return answer


# def resonstruct_words(tokens, answer_start, answer_end):
#     '''reconstruct any words that got broken down into subwords.
#     '''
#     # Start with the first token.
#     answer = tokens[answer_start]
#     # Select the remaining answer tokens and join them with whitespace.
#     for i in range(answer_start + 1, answer_end + 1):
#         # If it's a subword token, then recombine it with the previous token.
#         if tokens[i][0:2] == '##':
#             answer += tokens[i][2:]
#         # Otherwise, add a space then the token.
#         else:
#             answer += ' ' + tokens[i]
#     print('Answer: "' + answer + '"')


class EndpointHandler:
    def __init__(self, path=""):
        self.model = BertForQuestionAnswering.from_pretrained(path).to(device)
        self.tokenizer = BertTokenizer.from_pretrained(path)

    def __call__(
        self, 
        data: Dict[str, str | bytes]
    ):
        """
        Args:
            data (:obj:):
                includes the deserialized image file as PIL.Image
        """
        question = data.pop("question", data)
        context = data.pop("context", data)

        input_ids = self.tokenizer.encode(question, context)
        # print('The input has a total of {:} tokens.'.format(len(input_ids)))

        segment_ids = get_segment_ids_aka_token_type_ids(
            self.tokenizer, 
            input_ids
        )
        # run prediction
        with torch.inference_mode():
            start_scores, end_scores = to_model(
                self.model, 
                input_ids, 
                segment_ids
            )
            answer = get_answer(
                start_scores, 
                end_scores, 
                input_ids,
                self.tokenizer
            )
        return answer