app.py

"""Run codes."""
# pylint: disable=line-too-long, broad-exception-caught, invalid-name, missing-function-docstring, too-many-instance-attributes, missing-class-docstring
# ruff: noqa: E501
import os
import time
from dataclasses import asdict, dataclass
from pathlib import Path
from types import SimpleNamespace

import gradio as gr
from about_time import about_time

# from ctransformers import AutoConfig, AutoModelForCausalLM
from ctransformers import AutoModelForCausalLM
from huggingface_hub import hf_hub_download
from loguru import logger

os.environ["TZ"] = "Asia/Shanghai"
try:
    time.tzset()  # type: ignore # pylint: disable=no-member
except Exception:
    # Windows
    logger.warning("Windows, cant run time.tzset()")

ns = SimpleNamespace(
    response="",
    generator=[],
)

default_system_prompt = "A conversation between a user and an LLM-based AI assistant named Local Assistant. Local Assistant gives helpful and honest answers."

user_prefix = "[user]: "
assistant_prefix = "[assistant]: "


def predict_str(prompt, bot):  # bot is in fact bot_history
    # logger.debug(f"{prompt=}, {bot=}, {timeout=}")

    if bot is None:
        bot = []

    logger.debug(f"{prompt=}, {bot=}")

    try:
        # user_prompt = prompt
        generator = generate(
            LLM,
            GENERATION_CONFIG,
            system_prompt=default_system_prompt,
            user_prompt=prompt.strip(),
        )

        ns.generator = generator  # for .then

    except Exception as exc:
        logger.error(exc)

    # bot.append([prompt, f"{response} {_}"])
    # return prompt, bot

    return prompt, bot + [[prompt, None]]


def bot_str(bot):
    if bot:
        bot[-1][1] = ""
    else:
        bot = [["Something is wrong", ""]]

    print(assistant_prefix, end=" ", flush=True)

    response = ""

    flag = 1
    then = time.time()
    for word in ns.generator:
        # record first response time
        if flag:
            logger.debug(f"\t {time.time() - then:.1f}s")
            flag = 0
        print(word, end="", flush=True)
        # print(word, flush=True)  # vertical stream
        response += word
        bot[-1][1] = response
        yield bot


def predict(prompt, bot):
    # logger.debug(f"{prompt=}, {bot=}, {timeout=}")
    logger.debug(f"{prompt=}, {bot=}")

    ns.response = ""
    then = time.time()
    with about_time() as atime:  # type: ignore
        try:
            # user_prompt = prompt
            generator = generate(
                LLM,
                GENERATION_CONFIG,
                system_prompt=default_system_prompt,
                user_prompt=prompt.strip(),
            )

            ns.generator = generator  # for .then

            print(assistant_prefix, end=" ", flush=True)

            response = ""
            buff.update(value="diggin...")

            flag = 1
            for word in generator:
                # record first response time
                if flag:
                    logger.debug(f"\t {time.time() - then:.1f}s")
                    flag = 0
                # print(word, end="", flush=True)
                print(word, flush=True)  # vertical stream
                response += word
                ns.response = response
                buff.update(value=response)
            print("")
            logger.debug(f"{response=}")
        except Exception as exc:
            logger.error(exc)
            response = f"{exc=}"

    # bot = {"inputs": [response]}
    _ = (
        f"(time elapsed: {atime.duration_human}, "  # type: ignore
        f"{atime.duration/(len(prompt) + len(response)):.1f}s/char)"  # type: ignore
    )

    bot.append([prompt, f"{response} {_}"])

    return prompt, bot


def predict_api(prompt):
    logger.debug(f"{prompt=}")
    ns.response = ""
    try:
        # user_prompt = prompt
        _ = GenerationConfig(
            temperature=0.2,
            top_k=0,
            top_p=0.9,
            repetition_penalty=1.0,
            max_new_tokens=512,  # adjust as needed
            seed=42,
            reset=False,  # reset history (cache)
            stream=True,  # TODO stream=False and generator
            threads=os.cpu_count() // 2,  # type: ignore  # adjust for your CPU
            stop=["<|im_end|>", "|<"],
        )

        # TODO: stream does not make sense in api?
        generator = generate(
            LLM, _, system_prompt=default_system_prompt, user_prompt=prompt.strip()
        )
        print(assistant_prefix, end=" ", flush=True)

        response = ""
        buff.update(value="diggin...")
        for word in generator:
            print(word, end="", flush=True)
            response += word
            ns.response = response
            buff.update(value=response)
        print("")
        logger.debug(f"{response=}")
    except Exception as exc:
        logger.error(exc)
        response = f"{exc=}"
    # bot = {"inputs": [response]}
    # bot = [(prompt, response)]

    return response


def download_quant(destination_folder: str, repo_id: str, model_filename: str):
    local_path = os.path.abspath(destination_folder)
    return hf_hub_download(
        repo_id=repo_id,
        filename=model_filename,
        local_dir=local_path,
        local_dir_use_symlinks=True,
    )


@dataclass
class GenerationConfig:
    temperature: float
    top_k: int
    top_p: float
    repetition_penalty: float
    max_new_tokens: int
    seed: int
    reset: bool
    stream: bool
    threads: int
    stop: list[str]


def format_prompt(system_prompt: str, user_prompt: str):
    """Format prompt based on: https://huggingface.co/spaces/m osaicml/mpt-30b-chat/blob/main/app.py."""
    # TODO im_start/im_end possible fix for WizardCoder

    system_prompt = f"<|im_start|>system\n{system_prompt}<|im_end|>\n"
    user_prompt = f"<|im_start|>user\n{user_prompt}<|im_end|>\n"
    assistant_prompt = "<|im_start|>assistant\n"

    return f"{system_prompt}{user_prompt}{assistant_prompt}"


def generate(
    llm: AutoModelForCausalLM,
    generation_config: GenerationConfig,
    system_prompt: str = default_system_prompt,
    user_prompt: str = "",
):
    """Run model inference, will return a Generator if streaming is true."""
    # if not user_prompt.strip():
    return llm(
        format_prompt(
            system_prompt,
            user_prompt,
        ),
        **asdict(generation_config),
    )


_ = """full url: https://huggingface.co/The Bloke/mpt-30B-chat-GGML/blob/main/mpt-30b-chat.ggmlv0.q4_1.bin"""

# https://huggingface.co/The Bloke/mpt-30B-chat-GGML
_ = """
mpt-30b-chat.ggmlv0.q4_0.bin 	q4_0 	4 	16.85 GB 	19.35 GB 	4-bit.
mpt-30b-chat.ggmlv0.q4_1.bin 	q4_1 	4 	18.73 GB 	21.23 GB 	4-bit. Higher accuracy than q4_0 but not as high as q5_0. However has quicker inference than q5 models.
mpt-30b-chat.ggmlv0.q5_0.bin 	q5_0 	5 	20.60 GB 	23.10 GB
mpt-30b-chat.ggmlv0.q5_1.bin 	q5_1 	5 	22.47 GB 	24.97 GB
mpt-30b-chat.ggmlv0.q8_0.bin 	q8_0 	8 	31.83 GB 	34.33 GB
"""
MODEL_FILENAME = "mpt-30b-chat.ggmlv0.q4_1.bin"
MODEL_FILENAME = "WizardCoder-15B-1.0.ggmlv3.q4_0.bin"  # 10.7G
MODEL_FILENAME = "WizardCoder-15B-1.0.ggmlv3.q4_1.bin"  # 11.9G
MODEL_FILENAME = "WizardCoder-15B-1.0.ggmlv3.q4_1.bin"  # 11.9G

# https://huggingface.co/TheBloke/WizardLM-13B-V1.0-Uncensored-GGML
MODEL_FILENAME = "wizardlm-13b-v1.0-uncensored.ggmlv3.q4_1.bin"  # 8.4G

DESTINATION_FOLDER = "models"

REPO_ID = "The Bloke/mpt-30B-chat-GGML"
if "WizardCoder" in MODEL_FILENAME:
    REPO_ID = "The Bloke/WizardCoder-15B-1.0-GGML"

if "uncensored" in MODEL_FILENAME.lower():
    REPO_ID = "TheBloke/WizardLM-13B-V1.0-Uncensored-GGML"

logger.info(f"start dl, {REPO_ID=}, {MODEL_FILENAME=}, {DESTINATION_FOLDER=}")
download_quant(DESTINATION_FOLDER, REPO_ID, MODEL_FILENAME)
logger.info("done dl")

# if "mpt" in model_filename:
#     config = AutoConfig.from_pretrained("m osaicml/mpt-30b-cha t", context_length=8192)
#     llm = AutoModelForCausalLM.from_pretrained(
#         os.path.abspath(f"models/{model_filename}"),
#         model_type="mpt",
#         config=config,
#     )

# https://huggingface.co/spaces/matthoffner/wizardcoder-ggml/blob/main/main.py
_ = """
llm = AutoModelForCausalLM.from_pretrained(
    "The Bloke/WizardCoder-15B-1.0-GGML",
    model_file="WizardCoder-15B-1.0.ggmlv3.q4_0.bin",
    model_type="starcoder",
    threads=8
)
# """

logger.debug(f"{os.cpu_count()=}")
logger.info("load llm")

_ = Path("models", MODEL_FILENAME).absolute().as_posix()
logger.debug(f"model_file: {_}, exists: {Path(_).exists()}")
LLM = AutoModelForCausalLM.from_pretrained(
    # "The Bloke/WizardCoder-15B-1.0-GGML",
    REPO_ID,  # DESTINATION_FOLDER,  # model_path_or_repo_id: str required
    model_file=_,
    model_type="llama",   # "starcoder",  AutoConfig.from_pretrained("TheBloke/WizardLM-13B-V1.0-Uncensored-GGML")
    threads=os.cpu_count() // 2,  # type: ignore
)

logger.info("done load llm")

cpu_count = os.cpu_count() // 2  # type: ignore
logger.debug(f"{cpu_count=}")

GENERATION_CONFIG = GenerationConfig(
    temperature=0.2,
    top_k=0,
    top_p=0.9,
    repetition_penalty=1.0,
    max_new_tokens=512,  # adjust as needed
    seed=42,
    reset=False,  # reset history (cache)
    stream=True,  # streaming per word/token
    threads=cpu_count,
    stop=["<|im_end|>", "|<"],  # TODO possible fix of stop
)

css = """
    .importantButton {
        background: linear-gradient(45deg, #7e0570,#5d1c99, #6e00ff) !important;
        border: none !important;
    }
    .importantButton:hover {
        background: linear-gradient(45deg, #ff00e0,#8500ff, #6e00ff) !important;
        border: none !important;
    }
    .disclaimer {font-variant-caps: all-small-caps; font-size: xx-small;}
    .xsmall {font-size: x-small;}
"""
etext = """In America, where cars are an important part of the national psyche, a decade ago people had suddenly started to drive less, which had not happened since the oil shocks of the 1970s. """
examples = [
    ["How to pick a lock? Provide detailed steps."],
    ["Explain the plot of Cinderella in a sentence."],
    [
        "How long does it take to become proficient in French, and what are the best methods for retaining information?"
    ],
    ["What are some common mistakes to avoid when writing code?"],
    ["Build a prompt to generate a beautiful portrait of a horse"],
    ["Suggest four metaphors to describe the benefits of AI"],
    ["Write a pop song about leaving home for the sandy beaches."],
    ["Write a summary demonstrating my ability to tame lions"],
    ["鲁迅和周树人什么关系 说中文"],
    ["鲁迅和周树人什么关系"],
    ["鲁迅和周树人什么关系 用英文回答"],
    ["从前有一头牛，这头牛后面有什么？"],
    ["正无穷大加一大于正无穷大吗？"],
    ["正无穷大加正无穷大大于正无穷大吗？"],
    ["-2的平方根等于什么"],
    ["树上有5只鸟，猎人开枪打死了一只。树上还有几只鸟？"],
    ["树上有11只鸟，猎人开枪打死了一只。树上还有几只鸟？提示：需考虑鸟可能受惊吓飞走。"],
    ["以红楼梦的行文风格写一张委婉的请假条。不少于320字。"],
    [f"{etext} 翻成中文，列出3个版本"],
    [f"{etext} \n 翻成中文，保留原意，但使用文学性的语言。不要写解释。列出3个版本"],
    ["假定 1 + 2 = 4, 试求 7 + 8"],
    ["判断一个数是不是质数的 javascript 码"],
    ["实现python 里 range(10)的 javascript 码"],
    ["实现python 里 [*(range(10)]的 javascript 码"],
    ["Erkläre die Handlung von Cinderella in einem Satz."],
    ["Erkläre die Handlung von Cinderella in einem Satz. Auf Deutsch"],
]

with gr.Blocks(
    # title="mpt-30b-chat-ggml",
    title=f"{MODEL_FILENAME}",
    theme=gr.themes.Soft(text_size="sm", spacing_size="sm"),
    css=css,
) as block:
    with gr.Accordion("🎈 Info", open=False):
        # gr.HTML(
        #     """<center><a href="https://huggingface.co/spaces/mikeee/mpt-30b-chat?duplicate=true"><img src="https://bit.ly/3gLdBN6" alt="Duplicate"></a> and spin a CPU UPGRADE to avoid the queue</center>"""
        # )
        gr.Markdown(
            f"""<h4><center>{MODEL_FILENAME}</center></h4>
            It takes about 100 seconds for the initial reply
            message to appear. Average streaming rate ~1 sec/chat. The bot only speaks English.

            Most examples are meant for another model. You probably should try to test
            some related prompts.

            Try to refresh the browser and try again when occasionally errors occur.

            It takes about >100 seconds to get a response. Restarting the space takes about 2 minutes if the space is asleep due to inactivity. If the space crashes for some reason, it will also take about 2 minutes to restart. You need to refresh the browser to reload the new space.
            """,
            elem_classes="xsmall",
        )

    # chatbot = gr.Chatbot().style(height=700)  # 500
    chatbot = gr.Chatbot(height=700)  # 500
    buff = gr.Textbox(show_label=False, visible=False)
    with gr.Row():
        with gr.Column(scale=5):
            msg = gr.Textbox(
                label="Chat Message Box",
                placeholder="Ask me anything (press Enter or click Submit to send)",
                show_label=False,
            ).style(container=False)
        with gr.Column(scale=1, min_width=80):
            with gr.Row():
                submit = gr.Button("Submit", elem_classes="xsmall")
                stop = gr.Button("Stop", visible=False)
                clear = gr.Button("Clear History", visible=True)
    with gr.Row(visible=False):
        with gr.Accordion("Advanced Options:", open=False):
            with gr.Row():
                with gr.Column(scale=2):
                    system = gr.Textbox(
                        label="System Prompt",
                        value=default_system_prompt,
                        show_label=False,
                    ).style(container=False)
                with gr.Column():
                    with gr.Row():
                        change = gr.Button("Change System Prompt")
                        reset = gr.Button("Reset System Prompt")

    with gr.Accordion("Example Inputs", open=True):
        examples = gr.Examples(
            examples=examples,
            inputs=[msg],
            examples_per_page=40,
        )

    # with gr.Row():
    with gr.Accordion("Disclaimer", open=False):
        _ = "-".join(MODEL_FILENAME.split("-")[:2])
        gr.Markdown(
            f"Disclaimer: {_} can produce factually incorrect output, and should not be relied on to produce "
            "factually accurate information. {_} was trained on various public datasets; while great efforts "
            "have been taken to clean the pretraining data, it is possible that this model could generate lewd, "
            "biased, or otherwise offensive outputs.",
            elem_classes=["disclaimer"],
        )
    _ = """
    msg.submit(
        # fn=conversation.user_turn,
        fn=predict,
        inputs=[msg, chatbot],
        outputs=[msg, chatbot],
        # queue=True,
        show_progress="full",
        api_name="predict",
    )
    submit.click(
        fn=lambda x, y: ("",) + predict(x, y)[1:],  # clear msg
        inputs=[msg, chatbot],
        outputs=[msg, chatbot],
        queue=True,
        show_progress="full",
    )
    # """
    msg.submit(
        # fn=conversation.user_turn,
        fn=predict_str,
        inputs=[msg, chatbot],
        outputs=[msg, chatbot],
        queue=True,
        show_progress="full",
        api_name="predict",
    ).then(bot_str, chatbot, chatbot)
    submit.click(
        fn=lambda x, y: ("",) + predict_str(x, y)[1:],  # clear msg
        inputs=[msg, chatbot],
        outputs=[msg, chatbot],
        queue=True,
        show_progress="full",
    ).then(bot_str, chatbot, chatbot)

    clear.click(lambda: None, None, chatbot, queue=False)

    # update buff Textbox, every: units in seconds)
    # https://huggingface.co/spaces/julien-c/nvidia-smi/discussions
    # does not work
    # AttributeError: 'Blocks' object has no attribute 'run_forever'
    # block.run_forever(lambda: ns.response, None, [buff], every=1)

    with gr.Accordion("For Chat/Translation API", open=False, visible=False):
        input_text = gr.Text()
        api_btn = gr.Button("Go", variant="primary")
        out_text = gr.Text()
    api_btn.click(
        predict_api,
        input_text,
        out_text,
        # show_progress="full",
        api_name="api",
    )

# concurrency_count=5, max_size=20
# max_size=36, concurrency_count=14
block.queue(concurrency_count=5, max_size=20).launch(debug=True)