app.py

import gradio as gr
from llama_index.core.readers import SimpleDirectoryReader
from llama_index.core import VectorStoreIndex, Document
from llama_index.core.node_parser import SentenceSplitter
from llama_index.core import Settings
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.huggingface import HuggingFaceLLM
import csv
from docx import Document as DocxDocument
import fitz

# Define the list of LLMs with their names and models
lm_list = {
    "google/gemma-2-9b-it": "Google Gemma 2.9B IT",
    "mistralai/Mistral-7B-Instruct-v0.3": "Mistral 7B Instruct v0.3"
}

# Initialize the query engine globally
query_engine = None

def process_file(file):
    file_extension = file.name.split(".")[-1].lower()

    if file_extension == 'txt':
        with open(file.name, 'r', encoding='utf-8') as f:
            text = f.read()

    elif file_extension == 'csv':
        with open(file.name, 'r', encoding='utf-8') as f:
            reader = csv.reader(f)
            text = '\n'.join(','.join(row) for row in reader)

    elif file_extension == 'pdf':
        pdf_document = fitz.open(file.name, filetype=file_extension)
        text = ""
        for page_num in range(pdf_document.page_count):
            page = pdf_document.load_page(page_num)
            text += page.get_text("text")
        pdf_document.close()

    elif file_extension == 'docx':
        docx_document = DocxDocument(file.name)
        text = ""
        for paragraph in docx_document.paragraphs:
            text += paragraph.text + "\n"

    return [Document(text=text)]

def handle_file_upload(file, llm_name):
    global query_engine

    Settings.llm = HuggingFaceLLM(model_name=llm_name)

    documents = process_file(file)

    text_splitter = SentenceSplitter(chunk_size=512, chunk_overlap=10)
    Settings.embed_model = HuggingFaceEmbedding(model_name="nomic-embed-text:latest")
    Settings.text_splitter = text_splitter
    index = VectorStoreIndex.from_documents(
        documents, transformations=[text_splitter], embed_model=Settings.embed_model
    )

    return index.as_query_engine()

def document_qa(file_upload, llm_choice, question_input):
    query_engine = handle_file_upload(file_upload, llm_choice)
    result = query_engine.query(question_input)
    return str(result)


llm_choice = gr.Dropdown(choices=list(lm_list.values()), label="Choose LLM")
file_upload = gr.File(label="Upload Document")
question_input = gr.Textbox(label="Enter your question")

gr.Interface(
    fn=document_qa,
    inputs=[file_upload, llm_choice, question_input],
    outputs=gr.Textbox(label="Answer"),
    title="Document Question Answering",
    description="Upload a document and choose a language model to get answers.",
    allow_flagging=False
).launch()