MM-REACT/app.py

import re
import io
import os
from typing import Optional, Tuple
import datetime
import sys
import gradio as gr
import requests
import json
from threading import Lock
from langchain import ConversationChain, LLMChain
from langchain.agents import load_tools, initialize_agent, Tool
from langchain.tools.bing_search.tool import BingSearchRun, BingSearchAPIWrapper
from langchain.chains.conversation.memory import ConversationBufferMemory
from langchain.llms import OpenAI
from langchain.chains import PALChain
from langchain.llms import AzureOpenAI
from langchain.utilities import ImunAPIWrapper, ImunMultiAPIWrapper
from langchain.utils import get_url_path
from openai.error import AuthenticationError, InvalidRequestError, RateLimitError
import argparse
import logging
from opencensus.ext.azure.log_exporter import AzureLogHandler
import uuid

logger = None


OPENAI_API_KEY = os.environ.get("OPENAI_API_KEY")
BUG_FOUND_MSG = "Some Functionalities not supported yet. Please refresh and hit 'Click to wake up MM-REACT'"
AUTH_ERR_MSG = "OpenAI key needed"
REFRESH_MSG = "Please refresh and hit 'Click to wake up MM-REACT'"
MAX_TOKENS = 512


def get_logger():
    global logger
    if logger is None:
        logger = logging.getLogger(__name__)
        logger.addHandler(AzureLogHandler())
    return logger


# load chain
def load_chain(history, log_state):
    global ARGS

    if ARGS.openAIModel == 'openAIGPT35':
        # openAI GPT 3.5
        llm = OpenAI(temperature=0, max_tokens=MAX_TOKENS)
    elif ARGS.openAIModel == 'azureChatGPT':
        # for Azure OpenAI ChatGPT
        llm = AzureOpenAI(deployment_name="text-chat-davinci-002", model_name="text-chat-davinci-002", temperature=0, max_tokens=MAX_TOKENS)
    elif ARGS.openAIModel == 'azureGPT35turbo':
        # for Azure OpenAI gpt3.5 turbo
        llm = AzureOpenAI(deployment_name="gpt-35-turbo-version-0301", model_name="gpt-35-turbo (version 0301)", temperature=0, max_tokens=MAX_TOKENS)
    elif ARGS.openAIModel == 'azureTextDavinci003':
        # for Azure OpenAI text davinci
        llm = AzureOpenAI(deployment_name="text-davinci-003", model_name="text-davinci-003", temperature=0, max_tokens=MAX_TOKENS)
    elif ARGS.openAIModel == 'azureGPT4':
        # for Azure GPT4 private preview
        llm = AzureOpenAI(deployment_name="gpt-4-32k-0613", temperature=0, chat_completion=True, max_tokens=MAX_TOKENS, openai_api_version="2023-03-15-preview")

    memory = ConversationBufferMemory(memory_key="chat_history")


    #############################
    # loading all tools

    imun_dense = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_URL2"),
        params=os.environ.get("IMUN_PARAMS2"),
        imun_subscription_key=os.environ.get("IMUN_SUBSCRIPTION_KEY2"))

    imun = ImunAPIWrapper()
    imun = ImunMultiAPIWrapper(imuns=[imun, imun_dense])

    imun_celeb = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_CELEB_URL"),
        params="")

    imun_read = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_OCR_READ_URL"),
        params=os.environ.get("IMUN_OCR_PARAMS"),
        imun_subscription_key=os.environ.get("IMUN_OCR_SUBSCRIPTION_KEY"))

    imun_receipt = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_OCR_RECEIPT_URL"),
        params=os.environ.get("IMUN_OCR_PARAMS"),
        imun_subscription_key=os.environ.get("IMUN_OCR_SUBSCRIPTION_KEY"))

    imun_businesscard = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_OCR_BC_URL"),
        params=os.environ.get("IMUN_OCR_PARAMS"),
        imun_subscription_key=os.environ.get("IMUN_OCR_SUBSCRIPTION_KEY"))

    imun_layout = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_OCR_LAYOUT_URL"),
        params=os.environ.get("IMUN_OCR_PARAMS"),
        imun_subscription_key=os.environ.get("IMUN_OCR_SUBSCRIPTION_KEY"))

    imun_invoice = ImunAPIWrapper(
        imun_url=os.environ.get("IMUN_OCR_INVOICE_URL"),
        params=os.environ.get("IMUN_OCR_PARAMS"),
        imun_subscription_key=os.environ.get("IMUN_OCR_SUBSCRIPTION_KEY"))

    bing = BingSearchAPIWrapper(k=2)

    def edit_photo(query: str) -> str:
        endpoint = os.environ.get("PHOTO_EDIT_ENDPOINT_URL")
        query = query.strip()
        url_idx, img_url = get_url_path(query)
        if not img_url.startswith(("http://", "https://")):
            return "Invalid image URL"
        img_url = img_url.replace("0.0.0.0", os.environ.get("PHOTO_EDIT_ENDPOINT_URL_SHORT"))
        instruction = query[:url_idx]
        # This should be some internal IP to wherever the server runs
        job = {"image_path": img_url, "instruction": instruction}
        response = requests.post(endpoint, json=job)
        if response.status_code != 200:
            return "Could not finish the task try again later!"
        return "Here is the edited image " + endpoint + response.json()["edited_image"]

    # these tools should not step on each other's toes
    tools = [
        Tool(
            name="PAL-MATH",
            func=PALChain.from_math_prompt(llm).run,
            description=(
            "A wrapper around calculator. "
            "A language model that is really good at solving complex word math problems."
            "Input should be a fully worded hard word math problem."
            )
        ),
        Tool(
            name = "Image Understanding",
            func=imun.run,
            description=(
            "A wrapper around Image Understanding. "
            "Useful for when you need to understand what is inside an image (objects, texts, people)."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "OCR Understanding",
            func=imun_read.run,
            description=(
            "A wrapper around OCR Understanding (Optical Character Recognition). "
            "Useful after Image Understanding tool has found text or handwriting is present in the image tags."
            "This tool can find the actual text, written name, or product name in the image."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "Receipt Understanding",
            func=imun_receipt.run,
            description=(
            "A wrapper receipt understanding. "
            "Useful after Image Understanding tool has recognized a receipt in the image tags."
            "This tool can find the actual receipt text, prices and detailed items."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "Business Card Understanding",
            func=imun_businesscard.run,
            description=(
            "A wrapper around business card understanding. "
            "Useful after Image Understanding tool has recognized businesscard in the image tags."
            "This tool can find the actual business card text, name, address, email, website on the card."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "Layout Understanding",
            func=imun_layout.run,
            description=(
            "A wrapper around layout and table understanding. "
            "Useful after Image Understanding tool has recognized businesscard in the image tags."
            "This tool can find the actual business card text, name, address, email, website on the card."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "Invoice Understanding",
            func=imun_invoice.run,
            description=(
            "A wrapper around invoice understanding. "
            "Useful after Image Understanding tool has recognized businesscard in the image tags."
            "This tool can find the actual business card text, name, address, email, website on the card."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        Tool(
            name = "Celebrity Understanding",
            func=imun_celeb.run,
            description=(
            "A wrapper around celebrity understanding. "
            "Useful after Image Understanding tool has recognized people in the image tags that could be celebrities."
            "This tool can find the name of celebrities in the image."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
        BingSearchRun(api_wrapper=bing),
        Tool(
            name = "Photo Editing",
            func=edit_photo,
            description=(
            "A wrapper around photo editing. "
            "Useful to edit an image with a given instruction."
            "Input should be an image url, or path to an image file (e.g. .jpg, .png)."
            )
        ),
    ]

    chain = initialize_agent(tools, llm, agent="conversational-assistant", verbose=True, memory=memory, return_intermediate_steps=True, max_iterations=4)
    log_state = log_state or ""
    print ("log_state {}".format(log_state))
    log_state = str(uuid.uuid1())
    print("langchain reloaded")
    # eproperties = {'custom_dimensions': {'key_1': 'value_1', 'key_2': 'value_2'}} 
    properties = {'custom_dimensions': {'session': log_state}}
    get_logger().warning("langchain reloaded", extra=properties)    
    history = []
    history.append(("Show me what you got!", "Hi Human, Please upload an image to get started!"))    
    
    return history, history, chain, log_state, \
        gr.Textbox.update(visible=True), \
        gr.Button.update(visible=True), \
        gr.UploadButton.update(visible=True), \
        gr.Row.update(visible=True), \
        gr.HTML.update(visible=True), \
        gr.Button.update(variant="secondary")


# executes input typed by human
def run_chain(chain, inp):
    # global chain
    
    output = ""
    try:
        output = chain.conversation(input=inp, keep_short=ARGS.noIntermediateConv)
        # output = chain.run(input=inp)
    except AuthenticationError as ae:
        output = AUTH_ERR_MSG + str(datetime.datetime.now()) + ". " + str(ae)
        print("output", output)
    except RateLimitError as rle:
        output = "\n\nRateLimitError: " + str(rle)
    except ValueError as ve:
        output = "\n\nValueError: " + str(ve)
    except InvalidRequestError as ire:
        output = "\n\nInvalidRequestError: " + str(ire)
    except Exception as e:
        output = "\n\n" + BUG_FOUND_MSG + ":\n\n" + str(e)

    return output

# simple chat function wrapper 
class ChatWrapper:

    def __init__(self):
        self.lock = Lock()

    def __call__(
            self, inp: str, history: Optional[Tuple[str, str]], chain: Optional[ConversationChain], log_state
    ):
        
        """Execute the chat functionality."""
        self.lock.acquire()
        try:
            print("\n==== date/time: " + str(datetime.datetime.now()) + " ====")
            print("inp: " + inp)

            properties = {'custom_dimensions': {'session': log_state}}
            get_logger().warning("inp: " + inp, extra=properties)


            history = history or []
            # If chain is None, that is because no API key was provided.
            output = "Please paste your OpenAI key from openai.com to use this app. " + str(datetime.datetime.now())
            
            ########################
            # multi line 
            outputs = run_chain(chain, inp)

            outputs = process_chain_output(outputs)

            print (" len(outputs) {}".format(len(outputs)))
            for i, output in enumerate(outputs):
                if i==0:
                    history.append((inp, output))    
                else:
                    history.append((None, output))
            

        except Exception as e:
            raise e
        finally:
            self.lock.release()

        print (history)
        properties = {'custom_dimensions': {'session': log_state}}
        if outputs is None:
            outputs = ""
        get_logger().warning(str(json.dumps(outputs)), extra=properties)

        return history, history, ""
   
def add_image_with_path(state, chain, imagepath, log_state):
    global ARGS
    state = state or []

    url_input_for_chain = "http://0.0.0.0:{}/file={}".format(ARGS.port, imagepath)

    outputs = run_chain(chain, url_input_for_chain)

    ########################
    # multi line response handling
    outputs = process_chain_output(outputs)

    for i, output in enumerate(outputs):
        if i==0:
            # state.append((f"![](/file={imagepath})", output))    
            state.append(((imagepath,), output))
        else:
            state.append((None, output))
    

    print (state)
    properties = {'custom_dimensions': {'session': log_state}}
    get_logger().warning("url_input_for_chain: " + url_input_for_chain, extra=properties)
    if outputs is None:
        outputs = ""
    get_logger().warning(str(json.dumps(outputs)), extra=properties)
    return state, state


# upload image 
def add_image(state, chain, image, log_state):
    global ARGS
    state = state or []

    # handling spaces in image path
    imagepath = image.name.replace(" ", "%20")

    url_input_for_chain = "http://0.0.0.0:{}/file={}".format(ARGS.port, imagepath)
    
    outputs = run_chain(chain, url_input_for_chain)

    ########################
    # multi line response handling
    outputs = process_chain_output(outputs)

    for i, output in enumerate(outputs):
        if i==0:
            state.append(((imagepath,), output))
        else:
            state.append((None, output))
    

    print (state)
    properties = {'custom_dimensions': {'session': log_state}}
    get_logger().warning("url_input_for_chain: " + url_input_for_chain, extra=properties)
    if outputs is None:
        outputs = ""
    get_logger().warning(str(json.dumps(outputs)), extra=properties)
    return state, state

# extract image url from response and process differently
def replace_with_image_markup(text):
    img_url = None
    text= text.strip()
    url_idx = text.rfind(" ")
    img_url = text[url_idx + 1:].strip()
    if img_url.endswith((".", "?")):
        img_url = img_url[:-1]

    # if img_url is not None:
    #     img_url = f"![](/file={img_url})"
    return img_url

# multi line response handling
def process_chain_output(outputs):
    global ARGS
    EMPTY_AI_REPLY = "AI:"
    # print("outputs {}".format(outputs))
    if isinstance(outputs, str): # single line output
        if outputs.strip() == EMPTY_AI_REPLY:
            outputs = REFRESH_MSG
        outputs = [outputs]
    elif isinstance(outputs, list): # multi line output
        if ARGS.noIntermediateConv: # remove the items with assistant in it. 
            cleanOutputs = []
            for output in outputs:
                if output.strip() == EMPTY_AI_REPLY:
                    output = REFRESH_MSG
                # found an edited image url to embed
                img_url = None
                # print ("type list: {}".format(output))
                if "assistant: here is the edited image " in output.lower():
                    img_url = replace_with_image_markup(output)
                    cleanOutputs.append("Assistant: Here is the edited image")
                    if img_url is not None:
                        cleanOutputs.append((img_url,))
                else:
                    cleanOutputs.append(output)
                # cleanOutputs = cleanOutputs + output+ "."
            outputs = cleanOutputs
    
    return outputs


def init_and_kick_off():
    global ARGS
    # initalize chatWrapper
    chat = ChatWrapper()

    exampleTitle = """<h3>Examples to start conversation..</h3>"""    
    comingSoon = """<center><b>MM-REACT: July 21th version with GPT4 and image understanding capabilities.</b> <i>(data will be collected for research purpose)</i></center>"""
    detailLinks = """    
    <center>
        <a href="https://multimodal-react.github.io/"> MM-ReAct Website</a>
        ·
        <a href="https://arxiv.org/abs/2303.11381">MM-ReAct Paper</a>
        ·
        <a href="https://github.com/microsoft/MM-REACT">MM-ReAct Code</a>
    </center>
    """

    with gr.Blocks(css="#tryButton {width: 120px;}") as block:  
        llm_state = gr.State()
        history_state = gr.State()
        chain_state = gr.State()      
        log_state = gr.State() 
        
        reset_btn = gr.Button(value="!!!CLICK to wake up MM-REACT!!!", variant="primary", elem_id="resetbtn").style(full_width=True)
        gr.HTML(detailLinks)
        gr.HTML(comingSoon)

        example_image_size = 90
        col_min_width = 80
        button_variant = "primary"
        with gr.Row():
            with gr.Column(scale=1.0, min_width=100):
                chatbot = gr.Chatbot(elem_id="chatbot", label="MM-REACT Bot").style(height=620)
            with gr.Column(scale=0.20, min_width=200, visible=False) as exampleCol:
                with gr.Row():
                    grExampleTitle = gr.HTML(exampleTitle, visible=False)
                with gr.Row():
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example3Image = gr.Image("images/receipt.png", interactive=False).style(height=example_image_size, width=example_image_size)
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example3ImageButton = gr.Button(elem_id="tryButton", value="Try it!", variant=button_variant).style(full_width=True)
                        # dummy text field to hold the path
                        example3ImagePath = gr.Text("images/receipt.png", interactive=False, visible=False)
                with gr.Row():
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example1Image = gr.Image("images/money.png", interactive=False).style(height=example_image_size, width=example_image_size)
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example1ImageButton = gr.Button(elem_id="tryButton", value="Try it!", variant=button_variant).style(full_width=True)
                        # dummy text field to hold the path
                        example1ImagePath = gr.Text("images/money.png", interactive=False, visible=False)               
                with gr.Row():
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example2Image = gr.Image("images/cartoon.png", interactive=False).style(height=example_image_size, width=example_image_size)
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example2ImageButton = gr.Button(elem_id="tryButton", value="Try it!", variant=button_variant).style(full_width=True)
                        # dummy text field to hold the path
                        example2ImagePath = gr.Text("images/cartoon.png", interactive=False, visible=False)
                with gr.Row():
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example4Image = gr.Image("images/product.png", interactive=False).style(height=example_image_size, width=example_image_size)
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example4ImageButton = gr.Button(elem_id="tryButton", value="Try it!", variant=button_variant).style(full_width=True)
                        # dummy text field to hold the path
                        example4ImagePath = gr.Text("images/product.png", interactive=False, visible=False)
                with gr.Row():                        
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example5Image = gr.Image("images/celebrity.png", interactive=False).style(height=example_image_size, width=example_image_size)
                    with gr.Column(scale=0.50, min_width=col_min_width):
                        example5ImageButton = gr.Button(elem_id="tryButton", value="Try it!", variant=button_variant).style(full_width=True)
                        # dummy text field to hold the path
                        example5ImagePath = gr.Text("images/celebrity.png", interactive=False, visible=False)



        with gr.Row():
            with gr.Column(scale=0.75):
                message = gr.Textbox(label="Upload a pic and ask!",
                                        placeholder="Type your question about the uploaded image",
                                        lines=1, visible=False)
            with gr.Column(scale=0.15):
                submit = gr.Button(value="Send", variant="secondary", visible=False).style(full_width=True)
            with gr.Column(scale=0.10, min_width=0):
                btn = gr.UploadButton("🖼️", file_types=["image"], visible=False).style(full_width=True)


        message.submit(chat, inputs=[message, history_state, chain_state, log_state], outputs=[chatbot, history_state, message])

        submit.click(chat, inputs=[message, history_state, chain_state, log_state], outputs=[chatbot, history_state, message])

        btn.upload(add_image, inputs=[history_state, chain_state, btn, log_state], outputs=[history_state, chatbot])
        
        # load the chain    
        reset_btn.click(load_chain, inputs=[history_state, log_state], outputs=[chatbot, history_state, chain_state, log_state, message, submit, btn, exampleCol, grExampleTitle, reset_btn])

        # setup listener click for the examples
        example1ImageButton.click(add_image_with_path, inputs=[history_state, chain_state, example1ImagePath, log_state], outputs=[history_state, chatbot])
        example2ImageButton.click(add_image_with_path, inputs=[history_state, chain_state, example2ImagePath, log_state], outputs=[history_state, chatbot])
        example3ImageButton.click(add_image_with_path, inputs=[history_state, chain_state, example3ImagePath, log_state], outputs=[history_state, chatbot])
        example4ImageButton.click(add_image_with_path, inputs=[history_state, chain_state, example4ImagePath, log_state], outputs=[history_state, chatbot])
        example5ImageButton.click(add_image_with_path, inputs=[history_state, chain_state, example5ImagePath, log_state], outputs=[history_state, chatbot])


    # launch the app
    block.launch(server_name="0.0.0.0", server_port = ARGS.port)

if __name__ == '__main__':
    parser = argparse.ArgumentParser()

    parser.add_argument('--port', type=int, required=False, default=7860)
    parser.add_argument('--openAIModel', type=str, required=False, default='azureGPT4')
    parser.add_argument('--noIntermediateConv', default=True, action='store_true', help='if this flag is turned on no intermediate conversation should be shown')

    global ARGS
    ARGS = parser.parse_args()

    init_and_kick_off()