app.py

import streamlit as st

from transformers import MarkupLMProcessor, MarkupLMForQuestionAnswering

import requests
from bs4 import BeautifulSoup

import numpy as np

import torch
import torch.nn.functional as F

# Prediction Parameters
MAX_LEN = 512
STRIDE = 100

# Answer filtering parameters
MAX_ANSWER_LEN = 30
MIN_CONFIDENCE = 0.9

# Model name
MODEL_STR = "microsoft/markuplm-base-finetuned-websrc"

# Load markuplm model
processor = MarkupLMProcessor.from_pretrained(MODEL_STR)
model = MarkupLMForQuestionAnswering.from_pretrained(MODEL_STR)

headers = {
    'user-agent': 'Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/86.0.4240.111 Safari/537.36',
}

# User Input
input_url = st.text_input(
    label="Enter url of page to scrape",
    value="https://finance.yahoo.com/quote/META/",
    key="url",
)

input_question = st.text_input(
    label="Enter Question",
    value="What is the market capitalization?",
    key="question",
)

st.write("Getting html page ...")

# Request page
page = requests.get(input_url, headers=headers)

# Parse page with beautifulsoup
soup = BeautifulSoup(page.content, "html.parser")

# Extract page body
body = soup.find('body')

html_string = str(body)
len(html_string)

# Process input string
encoding = processor(html_string, questions=input_question, return_tensors="pt", truncation="only_second", 
                     stride=STRIDE, max_length=MAX_LEN, return_overflowing_tokens=True, padding=True)

# Postprocess encoding
del encoding['overflow_to_sample_mapping']
encoding['token_type_ids'] = encoding['token_type_ids'].fill_(0)

# Keep index of question for future use
n_segments = encoding['input_ids'].shape[0]
question_index = encoding[0].tokens.index('</s>')

# Run model
with torch.no_grad():
    outputs = model(**encoding)

# Get start and end probabilities
start_probs = F.softmax(outputs.start_logits, dim=1).numpy()
end_probs = F.softmax(outputs.end_logits, dim=1).numpy()

# Extract and filter answers for each window
answers = []

for i in range(n_segments):
    
    start_index = np.argmax(start_probs[i])
    end_index = np.argmax(end_probs[i])
    confidence = max(start_probs[i]) * max(end_probs[i])

    if end_index > start_index and end_index - start_index <= MAX_ANSWER_LEN and start_index > question_index and end_index > question_index and confidence > MIN_CONFIDENCE:

        predict_answer_tokens = encoding.input_ids[0, start_index : end_index + 1]
        answer = processor.decode(predict_answer_tokens, skip_special_tokens=True)

        answers.append({"answer": answer, "confidence": confidence})

# Print answers
for answer in answers:
    st.write(answer)

st.write("Done!")