app.py

from fastapi import FastAPI, Request
import json
from transformers import BertForQuestionAnswering, BertTokenizer
import torch

app = FastAPI()

# Load the model and tokenizer once, globally
bert_model = BertForQuestionAnswering.from_pretrained(
    'bert-large-uncased-whole-word-masking-finetuned-squad',
    ignore_mismatched_sizes=True
)
bert_tokenizer = BertTokenizer.from_pretrained('bert-large-uncased-whole-word-masking-finetuned-squad')

def get_answer_using_bert(question, reference_text):
    # Tokenize input text
    input_ids = bert_tokenizer.encode(question, reference_text)
    input_tokens = bert_tokenizer.convert_ids_to_tokens(input_ids)

    # Locate [SEP] token
    sep_location = input_ids.index(bert_tokenizer.sep_token_id)
    first_seg_len, second_seg_len = sep_location + 1, len(input_ids) - (sep_location + 1)
    seg_embedding = [0] * first_seg_len + [1] * second_seg_len

    # Pass through the model
    model_scores = bert_model(torch.tensor([input_ids]), token_type_ids=torch.tensor([seg_embedding]))
    ans_start_loc, ans_end_loc = torch.argmax(model_scores[0]), torch.argmax(model_scores[1])
    result = ' '.join(input_tokens[ans_start_loc:ans_end_loc + 1])

    # Clean up result
    result = result.replace(' ##', '')
    return result

@app.post("/questionAnswering")
async def questionAnswering(request: Request):
    try:
        json_data = await request.json()
        query = json_data['query']
        context_list = json_data['context_list']
        result = []

        # Loop over contexts
        for val in context_list:
            context = val['context'].replace("\n", " ")
            answer_json_final = {
                'answer': get_answer_using_bert(query, context),
                'id': val['id'],
                'question': query
            }
            result.append(answer_json_final)

        return {"results": result}

    except Exception as e:
        return {"Error": str(e)}