main.py

#!/usr/bin/python3
# -*- coding: utf-8 -*-
"""
https://huggingface.co/spaces/fffiloni/langchain-chat-with-pdf-openai
"""
import argparse
import httpx
import importlib
import json
import logging
import os
import platform
import shutil
import time
from typing import List, Tuple

logging.basicConfig(
    level=logging.INFO if platform.system() == "Windows" else logging.INFO,
    format="%(asctime)s %(levelname)s %(message)s",
    datefmt="%Y-%m-%d %H:%M:%S",
)

import gradio as gr
import openai
from openai import OpenAI
from PIL import Image

import project_settings as settings
from project_settings import project_path

logger = logging.getLogger(__name__)


def get_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--examples_json_file",
        default="examples.json",
        type=str
    )
    parser.add_argument(
        "--history_json_file",
        default="history.json",
        type=str
    )
    parser.add_argument(
        "--description_md_file",
        default="description.md",
        type=str
    )
    parser.add_argument(
        "--openai_api_key",
        default=settings.environment.get("openai_api_key", default=None, dtype=str),
        type=str
    )
    args = parser.parse_args()
    return args


def dynamic_import_function(package_name: str, function_name: str):
    try:
        lib = importlib.import_module("functions.{}".format(package_name))
    except ModuleNotFoundError as e:
        raise e

    function = getattr(lib, function_name)

    # del lib
    return function


def click_create_assistant(openai_api_key: str,
                           name: str,
                           instructions: str,
                           description: str,
                           tools: str,
                           files: List[str],
                           file_ids: str,
                           model: str,
                           ):
    logger.info("click create assistant, name: {}".format(name))
    client = OpenAI(
        api_key=openai_api_key,
    )

    # tools
    tools = str(tools).strip()
    if tools is not None and len(tools) != 0:
        tools = tools.split("\n")
        tools = [json.loads(tool) for tool in tools if len(tool.strip()) != 0]
    else:
        tools = list()

    # files
    if files is not None and len(files) != 0:
        files = [
            client.files.create(
                file=open(file, "rb"),
                purpose='assistants'
            ) for file in files
        ]
    else:
        files = list()

    # file_ids
    file_ids = str(file_ids).strip()
    if file_ids is not None and len(file_ids) != 0:
        file_ids = file_ids.split("\n")
        file_ids = [file_id.strip() for file_id in file_ids if len(file_id.strip()) != 0]
    else:
        file_ids = list()

    # assistant
    assistant = client.beta.assistants.create(
        name=name,
        instructions=instructions,
        description=description,
        tools=tools,
        file_ids=file_ids + [file.id for file in files],
        model=model,
    )
    assistant_id = assistant.id

    return assistant_id, None


def click_list_assistant(openai_api_key: str) -> str:
    client = OpenAI(
        api_key=openai_api_key,
    )
    assistant_list = client.beta.assistants.list()
    assistant_list = assistant_list.model_dump(mode="json")

    result = ""
    for a in assistant_list["data"]:
        assis = "id: \n{}\nname: \n{}\ndescription: \n{}\n\n".format(a["id"], a["name"], a["description"])
        result += assis

    return result


def click_delete_assistant(openai_api_key: str,
                           assistant_id: str) -> str:
    assistant_id = assistant_id.strip()

    logger.info("click delete assistant, assistant_id: {}".format(assistant_id))

    client = OpenAI(
        api_key=openai_api_key,
    )

    try:
        assistant_deleted = client.beta.assistants.delete(assistant_id=assistant_id)
        result = "success" if assistant_deleted.deleted else "failed"
    except openai.NotFoundError as e:
        result = e.message
    return result


def click_delete_all_assistant(openai_api_key: str):
    client = OpenAI(
        api_key=openai_api_key,
    )
    assistant_list = client.beta.assistants.list()

    for a in assistant_list.data:
        client.beta.assistants.delete(a.id)
    return None


def click_list_file(openai_api_key: str):
    client = OpenAI(
        api_key=openai_api_key,
    )
    file_list = client.files.list()
    file_list = file_list.model_dump(mode="json")

    result = ""
    for f in file_list["data"]:
        file = "id: \n{}\nfilename: \n{}\nbytes: \n{}\nstatus: \n{}\n\n".format(
            f["id"], f["filename"], f["bytes"], f["status"]
        )
        result += file

    return result


def click_delete_file(openai_api_key: str,
                      file_id: str) -> str:
    file_id = file_id.strip()

    logger.info("click delete file, file_id: {}".format(file_id))

    client = OpenAI(
        api_key=openai_api_key,
    )

    try:
        assistant_deleted = client.files.delete(file_id=file_id)
        result = "success" if assistant_deleted.deleted else "failed"
    except openai.NotFoundError as e:
        result = e.message
    except httpx.InvalidURL as e:
        result = str(e)
    return result


def click_upload_files(openai_api_key: str,
                       files: List[str],
                       ):
    logger.info("click upload files, files: {}".format(files))

    client = OpenAI(
        api_key=openai_api_key,
    )

    result = list()
    if files is not None and len(files) != 0:
        files = [
            client.files.create(
                file=open(file, "rb"),
                purpose='assistants'
            ) for file in files
        ]
        file_ids = [file.id for file in files]
        result.extend(file_ids)
    return result


def click_list_function_python_script():
    function_script_dir = project_path / "functions"
    result = ""
    for script in function_script_dir.glob("*.py"):
        if script.name == "__init__.py":
            continue
        result += script.name
        result += "\n"
    return result


def click_upload_function_python_script(files: List[str]):
    tgt = project_path / "functions"
    if files is None:
        return None
    for file in files:
        shutil.copy(file, tgt.as_posix())
    return None


def click_delete_function_python_script(filename: str):
    function_script_dir = project_path / "functions"
    filename = function_script_dir / filename.strip()
    filename = filename.as_posix()
    try:
        os.remove(filename)
        result = "success"
    except FileNotFoundError as e:
        result = str(e)
    except Exception as e:
        result = str(e)
    return result


def click_download_function_python_script(name: str):
    function_script_dir = project_path / "functions"
    filename = function_script_dir / name.strip()
    if not filename.exists():
        files = None
        flag = "File Not Found: {}".format(name.strip())
    else:
        files = [filename.as_posix()]
        flag = "You can download it on `upload_python_script_files` now."
    return files, flag


def convert_message_list_to_conversation(message_list: List[dict]) -> List[Tuple[str, str]]:
    conversation = list()
    for message in message_list:
        role = message["role"]
        content = message["content"]
        for c in content:
            c_type = c["type"]
            if c_type != "text":
                continue
            text: dict = c["text"]

            if c_type == "text":
                text_value = text["value"]
                text_annotations = text["annotations"]

                msg = text_value
                for text_annotation in text_annotations:
                    a_type = text_annotation["type"]
                    if a_type == "file_citation":
                        msg += "\n\n"
                        msg += "\nquote: \n{}\nfile_id: \n{}".format(
                            text_annotation["file_citation"]["quote"],
                            text_annotation["file_citation"]["file_id"],
                        )
            else:
                raise NotImplementedError

            if role == "assistant":
                msg = [None, msg]
            else:
                msg = [msg, None]

            conversation.append(msg)

    return conversation


def refresh(openai_api_key: str,
            thread_id: str,
            ):
    client = OpenAI(
        api_key=openai_api_key,
    )

    message_list = client.beta.threads.messages.list(
        thread_id=thread_id
    )
    message_list = message_list.model_dump(mode="json")
    message_list = message_list["data"]
    message_list = list(sorted(message_list, key=lambda x: x["created_at"]))

    logger.debug("message_list: {}".format(message_list))
    conversation = convert_message_list_to_conversation(message_list)
    return conversation


def add_and_run(openai_api_key: str,
                assistant_id: str,
                thread_id: str,
                name: str,
                instructions: str,
                description: str,
                tools: str,
                files: List[str],
                file_ids: str,
                model: str,
                query: str,
                ):
    chatting_image = None

    client = OpenAI(
        api_key=openai_api_key,
    )
    if assistant_id is None or len(assistant_id.strip()) == 0:
        assistant_id, _ = click_create_assistant(
            openai_api_key,
            name, instructions, description, tools, files, file_ids, model
        )

    if thread_id is None or len(thread_id.strip()) == 0:
        thread = client.beta.threads.create()
        thread_id = thread.id

    logger.info(f"assistant_id: {assistant_id}, thread_id: {thread_id}, query: {query}")

    message = client.beta.threads.messages.create(
        thread_id=thread_id,
        role="user",
        content=query
    )
    run = client.beta.threads.runs.create(
        thread_id=thread_id,
        assistant_id=assistant_id,
    )

    delta_time = 0.1
    last_conversation = None
    no_updates_count = 0
    max_no_updates_count = 5
    while True:
        time.sleep(delta_time)
        run = client.beta.threads.runs.retrieve(
            thread_id=thread_id,
            run_id=run.id,
        )

        # required action
        if run.required_action is not None:
            if run.required_action.type == "submit_tool_outputs":
                tool_outputs = list()
                for tool_call in run.required_action.submit_tool_outputs.tool_calls:
                    function_name = tool_call.function.name
                    function_to_call = dynamic_import_function(function_name, function_name)
                    kwargs_required: List[str] = dynamic_import_function(function_name, "kwargs")()
                    try:
                        function_args = json.loads(tool_call.function.arguments)
                    except json.decoder.JSONDecodeError as e:
                        print(tool_call.function.arguments)
                        raise e
                    kwargs = {k: function_args.get(k) for k in kwargs_required}
                    function_response = function_to_call(**kwargs)
                    tool_outputs.append({
                        "tool_call_id": tool_call.id,
                        "output": function_response["text"],
                    })

                    # chatting image
                    image_filename: str = function_response.get("image", None)
                    if image_filename is not None:
                        # with open(image_filename, "rb") as f:
                        chatting_image = Image.open(open(image_filename, "rb"))

                run = client.beta.threads.runs.submit_tool_outputs(
                    thread_id=thread_id,
                    run_id=run.id,
                    tool_outputs=tool_outputs
                )

        # get message
        conversation = refresh(openai_api_key, thread_id)
        if conversation == last_conversation:
            if any([run.completed_at is not None,
                    run.cancelled_at is not None,
                    run.failed_at is not None,
                    run.expires_at is not None]):
                no_updates_count += 1

        if no_updates_count >= max_no_updates_count:
            break
        last_conversation = conversation

        result = [
            assistant_id, thread_id,
            conversation,

            # query and chatting image
            None, chatting_image
        ]
        yield result


def main():
    args = get_args()

    brief_description = """
    ## OpenAI Assistant

    基于 [OpenAI platform](https://platform.openai.com/docs/introduction) 开发的 assistant 界面及示例。使用方法等详细介绍在最下面。
    """

    with open(args.description_md_file, "r", encoding="utf-8") as f:
        md_description = f.read()

    # example json
    with open(args.examples_json_file, "r", encoding="utf-8") as f:
        examples = json.load(f)
    for example in examples:
        files: List[str] = example[4]
        if files is None:
            continue
        files = [(project_path / file).as_posix() for file in files]
        example[4] = files

    # history
    with open(args.history_json_file, "r", encoding="utf-8") as f:
        history = json.load(f)

    # ui
    with gr.Blocks() as blocks:
        gr.Markdown(value=brief_description)

        with gr.Row():
            # settings
            with gr.Column(scale=3):
                with gr.Tabs():
                    with gr.TabItem("create assistant"):
                        openai_api_key = gr.Text(
                            value=args.openai_api_key,
                            label="openai_api_key",
                            placeholder="Fill with your `openai_api_key`"
                        )

                        name = gr.Textbox(label="name")
                        instructions = gr.Textbox(label="instructions")
                        description = gr.Textbox(label="description")

                        model = gr.Dropdown(["gpt-4-1106-preview"], value="gpt-4-1106-preview", label="model")

                        # functions
                        tools = gr.TextArea(label="functions")

                        # upload files
                        retrieval_files = gr.Files(label="retrieval_files")
                        retrieval_file_ids = gr.TextArea(label="retrieval_file_ids")

                        # create assistant
                        create_assistant_button = gr.Button("create assistant", variant="secondary")

                    with gr.TabItem("assistants"):
                        list_assistant_button = gr.Button("list assistant")
                        assistant_list = gr.TextArea(label="assistant_list")

                        delete_assistant_id = gr.Textbox(max_lines=1, label="delete_assistant_id")
                        delete_assistant_button = gr.Button("delete assistant")

                        delete_all_assistant_button = gr.Button("delete all assistant")

                    with gr.TabItem("files"):
                        list_file_button = gr.Button("list file")
                        file_list = gr.TextArea(label="file_list")

                        upload_files = gr.Files(label="upload_files")
                        upload_files_button = gr.Button("upload file")

                        delete_file_id = gr.Textbox(max_lines=1, label="delete_file_id")
                        delete_file_button = gr.Button("delete file")

                    with gr.TabItem("function script"):
                        list_function_python_script_button = gr.Button("list python script")
                        list_function_python_script_list = gr.TextArea(label="python_script_list")

                        upload_function_python_script_files = gr.Files(label="upload_python_script_files")
                        upload_function_python_script_button = gr.Button("upload python script")

                        function_python_script_file = gr.Textbox(max_lines=1, label="python_script_file")
                        delete_function_python_script_button = gr.Button("delete python script")
                        download_function_python_script_button = gr.Button("download python script")

            # chat
            with gr.Column(scale=5):
                chat_bot = gr.Chatbot(label="conversation", height=600)
                with gr.Row():
                    with gr.Column(scale=3, min_width=100):
                        chatting_image = gr.Image(label="chatting_image", height=220, width=220, show_label=False)

                    with gr.Column(scale=7):
                        query = gr.Textbox(lines=1, label="query")
                        with gr.Row():
                            with gr.Column(scale=1, min_width=100):
                                add_and_run_button = gr.Button("Add and run", variant="primary")
                            with gr.Column(scale=1, min_width=100):
                                refresh_button = gr.Button("Refresh")

            # states
            with gr.Column(scale=2):
                assistant_id = gr.Textbox(value=None, label="assistant_id")
                thread_id = gr.Textbox(value=None, label="thread_id")

                tips = gr.TextArea(value=None, label="tips")

        # examples
        with gr.Row():
            gr.Examples(
                examples=examples,
                inputs=[
                    name, instructions, description, tools, retrieval_files, model,
                    query, tips
                ],
                examples_per_page=5
            )

        with gr.Row():
            with gr.Column(scale=1):
                gr.Markdown(value=md_description)
            with gr.Column(scale=1):
                gr.Markdown("会话记录")
                with gr.Tabs():
                    for h in history:
                        with gr.TabItem(h["title"]):
                            _ = gr.Chatbot(value=h["history"], label=h["title"], height=600)

        # create assistant
        create_assistant_button.click(
            click_create_assistant,
            inputs=[
                openai_api_key,
                name, instructions, description, tools, retrieval_files, retrieval_file_ids, model,
            ],
            outputs=[
                assistant_id, thread_id
            ]
        )

        # list assistant
        list_assistant_button.click(
            click_list_assistant,
            inputs=[
                openai_api_key
            ],
            outputs=[
                assistant_list
            ]
        )

        # delete assistant button
        delete_assistant_button.click(
            click_delete_assistant,
            inputs=[
                openai_api_key,
                delete_assistant_id
            ],
            outputs=[
                delete_assistant_id
            ]
        )

        # delete all assistant
        delete_all_assistant_button.click(
            click_delete_all_assistant,
            inputs=[
                openai_api_key
            ],
            outputs=[
                file_list
            ]
        )

        # list file
        list_file_button.click(
            click_list_file,
            inputs=[
                openai_api_key
            ],
            outputs=[
                file_list
            ],
        )

        # delete file
        delete_file_button.click(
            click_delete_file,
            inputs=[
                openai_api_key,
                delete_file_id
            ],
            outputs=[
                delete_file_id
            ]
        )

        # upload files
        upload_files_button.click(
            click_upload_files,
            inputs=[
                openai_api_key,
                upload_files
            ],
            outputs=[

            ]
        )

        # list python script
        list_function_python_script_button.click(
            click_list_function_python_script,
            inputs=[],
            outputs=[
                list_function_python_script_list
            ]
        )

        # upload function python script
        upload_function_python_script_button.click(
            click_upload_function_python_script,
            inputs=[
                upload_function_python_script_files
            ],
            outputs=[
                upload_function_python_script_files
            ]
        )

        # delete function python script
        delete_function_python_script_button.click(
            click_delete_function_python_script,
            inputs=[
                function_python_script_file
            ],
            outputs=[
                function_python_script_file
            ]
        )

        # download function python script
        download_function_python_script_button.click(
            click_download_function_python_script,
            inputs=[
                function_python_script_file
            ],
            outputs=[
                upload_function_python_script_files, function_python_script_file
            ]
        )

        # query submit
        query.submit(
            add_and_run,
            inputs=[
                openai_api_key,
                assistant_id, thread_id,
                name, instructions, description, tools, retrieval_files, retrieval_file_ids, model,
                query,
            ],
            outputs=[
                assistant_id, thread_id,
                chat_bot,
                query,
                chatting_image,
            ],
            api_name="query_submit",
        )

        # add and run
        add_and_run_button.click(
            add_and_run,
            inputs=[
                openai_api_key,
                assistant_id, thread_id,
                name, instructions, description, tools, retrieval_files, retrieval_file_ids, model,
                query,
            ],
            outputs=[
                assistant_id, thread_id,
                chat_bot,
                query,
                chatting_image
            ],
        )

        # refresh
        refresh_button.click(
            refresh,
            inputs=[
                openai_api_key,
                thread_id,
            ],
            outputs=[
                chat_bot
            ]
        )

    blocks.queue().launch(
        share=False if platform.system() == "Windows" else False,
        server_name="127.0.0.1" if platform.system() == "Windows" else "0.0.0.0",
        server_port=7860
    )

    return


if __name__ == '__main__':
    main()