app.py

import os
import wget
apex_file = wget.download(os.getenv('apex'))
os.system('tar zxvf {}'.format(apex_file))

os.system('ls -l; cd apex;' 
    'pip install -v --no-cache-dir --global-option="--cpp_ext" --global-option="--cuda_ext" .; cd ..')

# os.system('pip install gradio --upgrade')

resources = os.getenv('resources')
resources_tokenizer = os.getenv('resources_tokenizer')

_ = wget.download(resources_tokenizer)
resources_filename = wget.download(resources)

os.system('tar zxvf {}'.format(resources_filename))

# if not os.path.exists('./model_optim_rng.pth'):
#    resources_weight = os.getenv('resources_weight')
#    _ = wget.download(resources_weight)

os.system('ls -l')


import time
import argparse
import datetime
import json
import re
import os
import gradio as gr
import requests
from utils import (
    mplug_owl, load_demo_refresh_model_list, vote_last_response, 
    upvote_last_response, downvote_last_response, flag_last_response, regenerate, 
    add_text, after_process_image, get_inputs, init,
    headers, no_change_btn, enable_btn, disable_btn, get_window_url_params
)
from gradio_css import code_highlight_css
from gradio_patch import Chatbot as grChatbot
from conversation import default_conversation

SHARED_UI_WARNING = f'''### [NOTE] 
You can duplicate and use it with a paid private GPU.
<a class="duplicate-button" style="display:inline-block" target="_blank" href="https://huggingface.co./spaces/MAGAer13/mPLUG-Owl?duplicate=true"><img style="margin-top:0;margin-bottom:0" src="https://huggingface.co./datasets/huggingface/badges/raw/main/duplicate-this-space-xl-dark.svg" alt="Duplicate Space"></a>
Alternatively, you can also use the Colab demo on our project page.
<a style="display:inline-block" href="https://https://github.com/X-PLUG/mPLUG-Owl"><img style="margin-top:0;margin-bottom:0" src="https://img.shields.io/badge/Project%20Page-online-brightgreen"></a>
'''

io = None
init()
model = mplug_owl(device="cuda")
log_dir = ""

def load_demo(url_params, request: gr.Request):

    dropdown_update = gr.Dropdown.update(visible=True)
    if "model" in url_params:
        model = url_params["model"]
        if model in models:
            dropdown_update = gr.Dropdown.update(
                value=model, visible=True)

    state = default_conversation.copy()

    return (state,
            dropdown_update,
            gr.Chatbot.update(visible=True),
            gr.Textbox.update(visible=True),
            gr.Button.update(visible=True),
            gr.Row.update(visible=True),
            gr.Accordion.update(visible=True))

def contains_chinese(string):
    pattern = re.compile(r'[一-龥]')
    match = pattern.search(string)
    return match is not None

def clear_history(request: gr.Request):
    state = default_conversation.copy()

    return (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5

def http_bot(state, topk, max_new_tokens, random_seed, request: gr.Request):
    prompt = after_process_image(state.get_prompt())
    images = state.get_images()
    state.messages[-1][-1] = "▌"
    yield (state, state.to_gradio_chatbot()) + (disable_btn,) * 5

    # if contains_chinese(prompt):
    #     state.messages[-1][-1] = "**CURRENTLY WE ONLY SUPPORT ENGLISH. PLEASE REFRESH THIS PAGE TO RESTART.**"
    #     yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
    #     return

    try:
        data = get_inputs(prompt, images, topk, max_new_tokens, random_seed)
        output = model.prediction(data, log_dir)
        print(output)
        # output = output.replace("```", "")

        state.messages[-1][-1] = output + "▌"
        yield (state, state.to_gradio_chatbot()) + (disable_btn,) * 5
        time.sleep(0.03)

    except requests.exceptions.RequestException as e:
        state.messages[-1][-1] = "**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**"
        yield (state, state.to_gradio_chatbot()) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
        return

    state.messages[-1][-1] = state.messages[-1][-1][:-1]
    yield (state, state.to_gradio_chatbot()) + (enable_btn,) * 5


def add_text_http_bot(state, text, image, video, topk, max_new_tokens, random_seed, request: gr.Request):
    if len(text) <= 0 and (image is None or video is None):
        state.skip_next = True
        return (state, state.to_gradio_chatbot(), "", None, None) + (no_change_btn,) * 5

    if image is not None:
        multimodal_msg = None
        if '<image>' not in text:
            text = text + '\n<image>'

        if multimodal_msg is not None:
            return (state, state.to_gradio_chatbot(), multimodal_msg, None, None) + (
                no_change_btn,) * 5
        text = (text, image)
    
    if video is not None:
        num_frames = 4
        if '<image>' not in text:
            text = text + '\n<image>' * num_frames
        text = (text, video)

    state.append_message(state.roles[0], text)
    state.append_message(state.roles[1], None)
    state.skip_next = False

    yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5
    
    prompt = after_process_image(state.get_prompt())
    images = state.get_images()
    state.messages[-1][-1] = "▌"
    yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5

    # if contains_chinese(prompt):
    #     state.messages[-1][-1] = "**CURRENTLY WE ONLY SUPPORT ENGLISH. PLEASE REFRESH THIS PAGE TO RESTART OR CLEAR HISTORY.**"
    #     yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
    #     return

    try:
        data = get_inputs(prompt, images, topk, max_new_tokens, random_seed)
        output = model.prediction(data, log_dir)
        print(output)
        # output = output.replace("```", "")

        state.messages[-1][-1] = output + "▌"
        yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5
        time.sleep(0.03)

    except requests.exceptions.RequestException as e:
        state.messages[-1][-1] = "**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**"
        yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
        return

    state.messages[-1][-1] = state.messages[-1][-1][:-1]
    yield (state, state.to_gradio_chatbot(), "", None, None) + (enable_btn,) * 5


def regenerate_http_bot(state, topk, max_new_tokens, random_seed, request: gr.Request):
    state.messages[-1][-1] = None
    state.skip_next = False
    yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5
    
    prompt = after_process_image(state.get_prompt())
    images = state.get_images()
    state.messages[-1][-1] = " "
    yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5

    # if contains_chinese(prompt):
    #     state.messages[-1][-1] = "**CURRENTLY WE ONLY SUPPORT ENGLISH. PLEASE REFRESH THIS PAGE TO RESTART OR CLEAR HISTORY.**"
    #     yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
    #     return

    try:
        data = get_inputs(prompt, images, topk, max_new_tokens, random_seed)
        output = model.prediction(data, log_dir)
        print(">>>> output:", output)
        # output = output.replace("```", "")

        state.messages[-1][-1] = output + " "
        yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn,) * 5
        time.sleep(0.03)

    except requests.exceptions.RequestException as e:
        print(e)
        state.messages[-1][-1] = "**NETWORK ERROR DUE TO HIGH TRAFFIC. PLEASE REGENERATE OR REFRESH THIS PAGE.**"
        yield (state, state.to_gradio_chatbot(), "", None, None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
        return

    state.messages[-1][-1] = state.messages[-1][-1][:-1]
    yield (state, state.to_gradio_chatbot(), "", None, None) + (enable_btn,) * 5

title_markdown = ("""
# mPLUG-Owl🦉 (GitHub Repo: https://github.com/X-PLUG/mPLUG-Owl)
If you like our project, please give us a star on Github for latest update.
""")

tos_markdown = ("""
**Notice:** The output is generated by top-k sampling scheme and may involve some randomness. For multiple image and video, we cannot ensure it's performance since only image-text pairs are used during training.

### Terms of use
By using this service, users are required to agree to the following terms:
The service is a research preview intended for non-commercial use only. It only provides limited safety measures and may generate offensive content. It must not be used for any illegal, harmful, violent, racist, or sexual purposes. The service may collect user dialogue data for future research.
Please click the "Flag" button if you get any inappropriate answer! We will collect those to keep improving our moderator.
For an optimal experience, please use desktop computers for this demo, as mobile devices may compromise its quality.

Copyright 2023 Alibaba DAMO Academy.
""")

learn_more_markdown = ("""
### License
The service is a research preview intended for non-commercial use only, subject to the model [License](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) of LLaMA, [Terms of Use](https://openai.com/policies/terms-of-use) of the data generated by OpenAI, and [Privacy Practices](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb) of ShareGPT. Please contact us if you find any potential violation.
""")

css = code_highlight_css + """
version 1.0
"""

def build_demo():
    # with gr.Blocks(title="mPLUG-Owl🦉", theme=gr.themes.Base(), css=css) as demo:
    with gr.Blocks(title="mPLUG-Owl🦉", css=css) as demo:
        state = gr.State()

        gr.Markdown(title_markdown)

        with gr.Row():
            with gr.Column(scale=3):

                imagebox = gr.Image(type="pil")

                videobox = gr.Video()

                with gr.Accordion("Parameters", open=True, visible=False) as parameter_row:
                    topk = gr.Slider(minimum=1, maximum=5, value=5, step=1, interactive=True, label="Top K",)
                    max_output_tokens = gr.Slider(minimum=0, maximum=1024, value=512, step=64, interactive=True, label="Max output tokens",)
                    temperature = gr.Slider(minimum=0, maximum=10, value=1, step=0.1, interactive=True, label="Temperature",)
                gr.Markdown(tos_markdown)

            with gr.Column(scale=6):
                chatbot = grChatbot(elem_id="chatbot", visible=False).style(height=800)
                with gr.Row():
                    with gr.Column(scale=8):
                        textbox = gr.Textbox(show_label=False,
                            placeholder="Enter text and press ENTER", visible=False).style(container=False)
                    with gr.Column(scale=1, min_width=60):
                        submit_btn = gr.Button(value="Submit", visible=False)
                with gr.Row(visible=False) as button_row:
                    upvote_btn = gr.Button(value="👍  Upvote", interactive=False)
                    downvote_btn = gr.Button(value="👎  Downvote", interactive=False)
                    flag_btn = gr.Button(value="⚠️  Flag", interactive=False)
                    regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
                    clear_btn = gr.Button(value="🗑️  Clear history", interactive=False)

        gr.Examples(examples=[
            [f"examples/monday.jpg", "Explain why this meme is funny."],
            [f'examples/rap.jpeg', 'Can you write me a master rap song that rhymes very well based on this image?'],
            [f'examples/titanic.jpeg', 'What happened at the end of this movie?'],
            [f'examples/vga.jpeg', 'What is funny about this image? Describe it panel by panel.'],
            [f'examples/mug_ad.jpeg', 'We design new mugs shown in the image. Can you help us write an advertisement?'],
            [f'examples/laundry.jpeg', 'Why this happens and how to fix it?'],
            [f'examples/ca.jpeg', "What do you think about the person's behavior?"],
            [f'examples/monalisa-fun.jpg', 'Do you know who drew this painting?​'],
            # [f"examples/Yao_Ming.jpeg", "What is the name of the man on the right?"],
        ], inputs=[imagebox, textbox])

        gr.Examples(examples=[
            [f"examples/surf.mp4", "What is the man doing?"],
            [f"examples/yoga.mp4", "What did the woman doing?"],
        ], inputs=[videobox, textbox])

        gr.Markdown(learn_more_markdown)
        url_params = gr.JSON(visible=False)

        btn_list = [upvote_btn, downvote_btn, flag_btn, regenerate_btn, clear_btn]
        upvote_btn.click(upvote_last_response,
            [state], [textbox, upvote_btn, downvote_btn, flag_btn])
        downvote_btn.click(downvote_last_response,
            [state], [textbox, upvote_btn, downvote_btn, flag_btn])
        flag_btn.click(flag_last_response,
            [state], [textbox, upvote_btn, downvote_btn, flag_btn])
        # regenerate_btn.click(regenerate, state,
        #     [state, chatbot, textbox, imagebox] + btn_list).then(
        #     http_bot, [state, topk, max_output_tokens, temperature],
        #     [state, chatbot] + btn_list)
        regenerate_btn.click(regenerate_http_bot, [state, topk, max_output_tokens, temperature],
            [state, chatbot, textbox, imagebox, videobox] + btn_list)
        clear_btn.click(clear_history, None, [state, chatbot, textbox, imagebox, videobox] + btn_list)

        # textbox.submit(add_text, [state, textbox, imagebox], [state, chatbot, textbox, imagebox] + btn_list
        #     ).then(http_bot, [state, topk, max_output_tokens, temperature],
        #            [state, chatbot] + btn_list)
        textbox.submit(add_text_http_bot, [state, textbox, imagebox, videobox, topk, max_output_tokens, temperature], [state, chatbot, textbox, imagebox, videobox] + btn_list)

        submit_btn.click(add_text_http_bot, [state, textbox, imagebox, videobox, topk, max_output_tokens, temperature], [state, chatbot, textbox, imagebox, videobox] + btn_list)

        demo.load(load_demo, [url_params], [state,
            chatbot, textbox, submit_btn, button_row, parameter_row],
            _js=get_window_url_params)

    return demo

if __name__ == "__main__":
    cur_dir = os.getenv("cur_dir")
    log_dir = cur_dir + "log"

    parser = argparse.ArgumentParser()
    parser.add_argument("--host", type=str, default="0.0.0.0")
    parser.add_argument("--debug", action="store_true", help="using debug mode")
    parser.add_argument("--port", type=int)
    parser.add_argument("--concurrency-count", type=int, default=4)
    args = parser.parse_args()
    demo = build_demo()
    demo.queue(concurrency_count=args.concurrency_count, status_update_rate=10, api_open=False).launch(server_name=args.host, debug=args.debug, server_port=args.port, share=False)