Update app.py

app.py

@@ -1,50 +1,71 @@
-import streamlit as st
+import os
+import gradio as gr
+from langchain_core.prompts import PromptTemplate
+from langchain_community.document_loaders import PyPDFLoader
+from langchain_google_genai import ChatGoogleGenerativeAI
 import google.generativeai as genai
+from langchain.chains.question_answering import load_qa_chain
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
 
 # Configure Gemini API
-genai.configure(api_key='AIzaSyCgfXYWyJjK8YrApUjO6433P3WAdpwQi0Y')
-model = genai.GenerativeModel('gemini-pro')
+genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
 
-def generate_response(user_message):
-    try:
-        # Use Gemini API to generate a response
-        response = model.generate_content(user_message)
+# Load Mistral model
+model_path = "nvidia/Mistral-NeMo-Minitron-8B-Base"
+mistral_tokenizer = AutoTokenizer.from_pretrained(model_path)
+device = 'cuda' if torch.cuda.is_available() else 'cpu'
+dtype = torch.bfloat16
+mistral_model = AutoModelForCausalLM.from_pretrained(model_path, torch_dtype=dtype, device_map=device)
 
-        # Check if the response has text
-        if response.text:
-            return response.text
+def initialize(file_path, question):
+    try:
+        model = ChatGoogleGenerativeAI(model="gemini-pro", temperature=0.3)
+        prompt_template = """Answer the question as precise as possible using the provided context. If the answer is
+                              not contained in the context, say "answer not available in context" \n\n
+                              Context: \n {context}?\n
+                              Question: \n {question} \n
+                              Answer:
+                            """
+        prompt = PromptTemplate(template=prompt_template, input_variables=["context", "question"])
+        
+        if os.path.exists(file_path):
+            pdf_loader = PyPDFLoader(file_path)
+            pages = pdf_loader.load_and_split()
+            context = "\n".join(str(page.page_content) for page in pages[:30])
+            stuff_chain = load_qa_chain(model, chain_type="stuff", prompt=prompt)
+            stuff_answer = stuff_chain({"input_documents": pages, "question": question, "context": context}, return_only_outputs=True)
+            gemini_answer = stuff_answer['output_text']
+            
+            # Use Mistral model for additional text generation
+            mistral_prompt = f"Based on this answer: {gemini_answer}\nGenerate a follow-up question:"
+            mistral_inputs = mistral_tokenizer.encode(mistral_prompt, return_tensors='pt').to(device)
+            with torch.no_grad():
+                mistral_outputs = mistral_model.generate(mistral_inputs, max_length=50)
+            mistral_output = mistral_tokenizer.decode(mistral_outputs[0], skip_special_tokens=True)
+            
+            combined_output = f"Gemini Answer: {gemini_answer}\n\nMistral Follow-up: {mistral_output}"
+            return combined_output
         else:
-            return "Error: The model generated an empty response."
+            return "Error: Unable to process the document. Please ensure the PDF file is valid."
     except Exception as e:
         return f"An error occurred: {str(e)}"
 
-# Initialize session state for conversation history
-if 'conversation' not in st.session_state:
-    st.session_state.conversation = []
-
-st.title("🦷🦷Gemini QA System🦷🦷")
-st.write("Ask a question and get an answer from Gemini AI.")
-
-# User input
-user_message = st.text_input("Enter your question here...")
-
-if st.button("Submit"):
-    if user_message:
-        response = generate_response(user_message)
-        st.session_state.conversation.append((user_message, response))
+# Define Gradio Interface
+input_file = gr.File(label="Upload PDF File")
+input_question = gr.Textbox(label="Ask about the document")
+output_text = gr.Textbox(label="Answer - Combined Gemini and Mistral")
 
-# Display conversation history
-if st.session_state.conversation:
-    st.write("### Conversation History")
-    for i, (question, answer) in enumerate(st.session_state.conversation):
-        st.write(f"**Q{i+1}:** {question}")
-        st.write(f"**A{i+1}:** {answer}")
+def pdf_qa(file, question):
+    if file is None:
+        return "Please upload a PDF file first."
+    return initialize(file.name, question)
 
-# Examples in the sidebar
-st.sidebar.title("Examples")
-st.sidebar.write("Click on an example to use it:")
-examples = ["What is the capital of France?", "Explain quantum computing in simple terms."]
-for example in examples:
-    if st.sidebar.button(example):
-        response = generate_response(example)
-        st.session_state.conversation.append((example, response))
+# Create Gradio Interface
+gr.Interface(
+    fn=pdf_qa,
+    inputs=[input_file, input_question],
+    outputs=output_text,
+    title="RAG Knowledge Retrieval using Gemini API and Mistral Model",
+    description="Upload a PDF file and ask questions about the content."
+).launch()