app.py

import gradio as gr
from llama_index.core import StorageContext, load_index_from_storage, VectorStoreIndex, SimpleDirectoryReader, ChatPromptTemplate
from llama_index.llms.huggingface import HuggingFaceInferenceAPI
from dotenv import load_dotenv
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.core import Settings
import os
import tempfile

# Load environment variables
load_dotenv()

# Configure the Llama index settings
Settings.llm = HuggingFaceInferenceAPI(
    model_name="google/gemma-1.1-7b-it",
    tokenizer_name="google/gemma-1.1-7b-it",
    context_window=3000,
    token=os.getenv("HF_TOKEN"),
    max_new_tokens=512,
    generate_kwargs={"temperature": 0.1},
)
Settings.embed_model = HuggingFaceEmbedding(
    model_name="BAAI/bge-small-en-v1.5"
)

# Define the directory for persistent storage and data
PERSIST_DIR = "./db"
DATA_DIR = "data"

# Ensure data directory exists
os.makedirs(DATA_DIR, exist_ok=True)
os.makedirs(PERSIST_DIR, exist_ok=True)

def data_ingestion():
    documents = SimpleDirectoryReader(DATA_DIR).load_data()
    storage_context = StorageContext.from_defaults()
    index = VectorStoreIndex.from_documents(documents)
    index.storage_context.persist(persist_dir=PERSIST_DIR)

def handle_query(query):
    storage_context = StorageContext.from_defaults(persist_dir=PERSIST_DIR)
    index = load_index_from_storage(storage_context)
    chat_text_qa_msgs = [
    (
        "user",
        """You are a Q&A assistant named EazyPeazy, For all other inquiries, your main goal is to provide answers as accurately as possible, based on the instructions and context you have been given. If a question does not match the provided context or is outside the scope of the document, kindly advise the user to ask questions within the context of the document.
        Context:
        {context_str}
        Question:
        {query_str}
        """
    )
    ]
    text_qa_template = ChatPromptTemplate.from_messages(chat_text_qa_msgs)
    
    query_engine = index.as_query_engine(text_qa_template=text_qa_template)
    answer = query_engine.query(query)
    
    if hasattr(answer, 'response'):
        return answer.response
    elif isinstance(answer, dict) and 'response' in answer:
        return answer['response']
    else:
        return "Sorry, I couldn't find an answer."

def process_file(file):
    if file is None:
        return "Please upload a PDF file."
    
    temp_dir = tempfile.mkdtemp()
    temp_path = os.path.join(temp_dir, "uploaded.pdf")
    
    with open(temp_path, "wb") as f:
        f.write(file.read())
    
    # Copy the file to the DATA_DIR
    os.makedirs(DATA_DIR, exist_ok=True)
    dest_path = os.path.join(DATA_DIR, "saved_pdf.pdf")
    os.replace(temp_path, dest_path)
    
    # Process the uploaded PDF
    data_ingestion()
    
    return "PDF processed successfully. You can now ask questions about its content."

def chatbot(message, history):
    response = handle_query(message)
    history.append((message, response))
    return history, ""

# Gradio interface
with gr.Blocks() as demo:
    gr.Markdown("# (PDF) Information and Inference🗞️")
    gr.Markdown("Retrieval-Augmented Generation")
    
    with gr.Row():
        with gr.Column(scale=1):
            file_output = gr.Textbox(label="Upload Status")
            upload_button = gr.UploadButton("Upload PDF", file_types=[".pdf"])
            upload_button.upload(process_file, upload_button, file_output)
        
        with gr.Column(scale=2):
            chatbot = gr.Chatbot(
                [],
                elem_id="chatbot",
                bubble_full_width=False,
            )
            msg = gr.Textbox(label="Ask me anything about the content of the PDF:")
            clear = gr.Button("Clear")
    
    msg.submit(chatbot, [msg, chatbot], [chatbot, msg])
    clear.click(lambda: None, None, chatbot, queue=False)

if __name__ == "__main__":
    demo.launch()