Create app.py

app.py

@@ -0,0 +1,79 @@
+
+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
+from PIL import Image
+import io
+
+# Configure Gemini API
+genai.configure(api_key=os.getenv("GOOGLE_API_KEY"))
+
+# 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)
+
+def process_pdf(file_path, question):
+    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"])
+    
+    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)
+    return stuff_answer['output_text']
+
+def process_image(image, question):
+    model = genai.GenerativeModel('gemini-pro-vision')
+    response = model.generate_content([image, question])
+    return response.text
+
+def generate_mistral_followup(answer):
+    mistral_prompt = f"Based on this answer: {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)
+    return mistral_output
+
+def process_input(file, image, question):
+    try:
+        if file is not None:
+            gemini_answer = process_pdf(file.name, question)
+        elif image is not None:
+            gemini_answer = process_image(image, question)
+        else:
+            return "Please upload a PDF file or an image."
+
+        mistral_followup = generate_mistral_followup(gemini_answer)
+        combined_output = f"Gemini Answer: {gemini_answer}\n\nMistral Follow-up: {mistral_followup}"
+        return combined_output
+    except Exception as e:
+        return f"An error occurred: {str(e)}"
+
+# Define Gradio Interface
+with gr.Blocks() as demo:
+    gr.Markdown("# Multi-modal RAG Knowledge Retrieval using Gemini API and Mistral Model")
+    
+    with gr.Row():
+        with gr.Column():
+            input_file = gr.File(label="Upload PDF File")
+            input_image = gr.Image(type="pil", label="Upload Image")
+        input_question = gr.Textbox(label="Ask about the document or image")
+    
+    output_text = gr.Textbox(label="Answer - Combined Gemini and Mistral")
+    
+    submit_button = gr.Button("Submit")
+    submit_button.click(fn=process_input, inputs=[input_file, input_image, input_question], outputs=output_text)
+
+demo.launch()