app.py

import logging
import os

import gradio as gr
from langchain import OpenAI, HuggingFaceHub
from langchain.chains import ConversationalRetrievalChain
from langchain.document_loaders import PyPDFLoader
from langchain.embeddings import OpenAIEmbeddings, HuggingFaceEmbeddings
from langchain.text_splitter import CharacterTextSplitter
from langchain.vectorstores import Chroma

documents = []
qa = None


def get_file(file):
    try:
        global documents
        data = PyPDFLoader(file.name)
        documents = data.load_and_split(CharacterTextSplitter(chunk_size=2000, chunk_overlap=0))
    except Exception as e:
        logging.error(e, exc_info=True)
        return "Failed to upload."
    return "File Uploaded."


def model_configuration(model_name, api_key=None, hug_model=None, hug_token=None, temperature=0, max_length=512):
    try:
        embeddings, llm = None, None
        if not documents:
            return gr.update(value="Please upload correct PDF!", visible=True)
        global qa
        if model_name == "OpenAI":
            os.environ["OPENAI_API_KEY"] = api_key or os.getenv("OPENAI_API_KEY")
            embeddings = OpenAIEmbeddings()
            llm = OpenAI(temperature=temperature, max_tokens=max_length)
        elif model_name == "HuggingFace":
            os.environ["HUGGINGFACEHUB_API_TOKEN"] = hug_token or os.getenv("HUGGINGFACE_API_KEY")
            embeddings = HuggingFaceEmbeddings(model_name=hug_model, model_kwargs={'device': 'cpu'})
            llm = HuggingFaceHub(repo_id=hug_model, model_kwargs={"temperature": temperature, "max_length": max_length})

        if embeddings:
            db = Chroma.from_documents(documents, embeddings)
            retriever = db.as_retriever(search_type="similarity", search_kwargs={"k": 2})
            qa = ConversationalRetrievalChain.from_llm(llm, chain_type="map_reduce", retriever=retriever,
                                                       return_source_documents=True, verbose=False)
    except Exception as e:
        logging.error(e, exc_info=True)
        return gr.update(value="Error occurred!", visible=True)
    return gr.update(value="Model Built", visible=True)


def response(msg, chat_history):
    global qa
    result = qa({"question": msg, "chat_history": map(tuple, chat_history)})
    final_resp = result.get("answer", "").strip()
    chat_history.append((msg, final_resp))
    docs = result.get("source_documents")
    return "", chat_history


with gr.Blocks() as demo:
    with gr.Tab("PDF Ingestion") as pdf_input:
        file = None
        with gr.Column() as r1:
            file = gr.File(file_types=[".pdf"])
            op_txt = gr.Label(value="", label="")
            fl_btn = gr.Button("Upload & Ingest 🚀")
            fl_btn.click(get_file, inputs=[file], outputs=op_txt)

    with gr.Tab("Select Model") as model:
        model_name = gr.Dropdown(
            ["NA", "OpenAI", "HuggingFace"],
            show_label=True,
            label="Model Name",
            multiselect=False,
            value="NA"
        )
        with gr.Column(visible=False) as openai_config:
            api_key = gr.Textbox(value="", label="OPENAI API KEY", placeholder="sk-...", visible=True, interactive=True,
                                 type="password")

        with gr.Column(visible=False) as huggy_config:
            hug_model = gr.Dropdown(["google/flan-t5-xl"],
                                    value="google/flan-t5-xl", multiselect=False)
            hug_token = gr.Textbox(value="", placeholder="hf_...", interactive=True, type="password")

        with gr.Accordion("Advance Settings", open=False, visible=False) as advance_settings:
            temperature = gr.Slider(0, 1, label="Temperature")
            max_length = gr.components.Number(value=512, label="Max Token Length")


        def show_configuration(model_name):
            match model_name:
                case "OpenAI":
                    return {
                        openai_config: gr.update(visible=True),
                        huggy_config: gr.update(visible=False),
                        advance_settings: gr.update(visible=True)
                    }
                case "HuggingFace":
                    return {
                        openai_config: gr.update(visible=False),
                        huggy_config: gr.update(visible=True),
                        advance_settings: gr.update(visible=True)
                    }
                case _:
                    return {
                        openai_config: gr.update(visible=False),
                        huggy_config: gr.update(visible=False),
                        advance_settings: gr.update(visible=False)
                    }


        model_name.change(show_configuration, inputs=[model_name],
                          outputs=[openai_config, huggy_config, advance_settings])
        model_updated = gr.Label("", show_label=False, visible=True)
        btn = gr.Button("Configure Model 🤖")
        btn.click(model_configuration, inputs=[model_name, api_key, hug_model, hug_token, temperature, max_length],
                  outputs=model_updated)

    with gr.Tab("Q&A") as qna:
        with gr.Column() as r:
            chatbot = gr.Chatbot(show_label=True)
            msg = gr.Textbox(placeholder="Ask Something")
            clear = gr.Button("Clear")
            msg.submit(response, [msg, chatbot], [msg, chatbot])
            clear.click(lambda: None, None, chatbot, queue=False)

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