Update app.py

app.py

@@ -1,43 +1,41 @@
-import os
-os.environ["HF_HOME"] = "./transformers"  # Set HF_HOME to a local directory to avoid TensorFlow dependency
-
 import streamlit as st
 import requests
 from transformers import pipeline, BertTokenizer
 
 # Function to generate answers using the BERT model
-def generate_answers(questions, paper_link):
-    # Download the research paper
-    response = requests.get(paper_link)
-    paper_text = response.text
-
+def generate_answers(chunks, question):
     # Initialize the BERT tokenizer
     tokenizer = BertTokenizer.from_pretrained("bert-large-uncased-whole-word-masking-finetuned-squad")
 
     # Initialize the question-answering pipeline
     model = pipeline("question-answering", model="bert-large-uncased-whole-word-masking-finetuned-squad")
 
-    # Generate answers for each question
-    answers = []
-    for question in questions.split(","):
-        inputs = tokenizer(question.strip(), paper_text, return_tensors="pt")
-        inputs = {k: v.tolist()[0] for k, v in inputs.items()}  # Convert tensors to lists
-        answer = model(**inputs)
-        answers.append(answer['answer'])
+    # Concatenate chunks into a single text
+    paper_text = ' '.join(chunks)
 
-    return '\n\n'.join(answers)
+    # Generate answers for the question based on the entire context
+    answer = model(question, paper_text)
+    return answer['answer']
 
 # Streamlit app
 st.title("Research Paper Question Answering")
 
-questions = st.text_input("Enter comma-separated questions:")
 paper_link = st.text_input("Enter the link to the research paper (Arxiv link):")
+question = st.text_input("Enter your question:")
 
-if st.button("Generate Answers"):
-    if not (questions and paper_link):
-        st.warning("Please provide both questions and the paper link.")
+if st.button("Generate Answer"):
+    if not (paper_link and question):
+        st.warning("Please provide both the paper link and the question.")
     else:
-        with st.spinner("Generating answers..."):
-            answers = generate_answers(questions, paper_link)
-            st.success("Answers generated successfully!")
-            st.text_area("Generated Answers", answers)
+        # Download the research paper
+        response = requests.get(paper_link)
+        paper_text = response.text
+
+        # Split the paper text into chunks of 512 words
+        paper_chunks = [paper_text[i:i+512] for i in range(0, len(paper_text), 512)]
+
+        # Generate answer based on chunks
+        answer = generate_answers(paper_chunks, question)
+        st.success("Answer generated successfully!")
+        st.text("Generated Answer:")
+        st.write(answer)