app.py

#!/usr/bin/env python

import gradio as gr
import os
import re
from PIL import Image
import base64
import time

DESCRIPTION = '''# <a href="https://github.com/THUDM/VisualGLM">VisualGLM</a>'''

MAINTENANCE_NOTICE='Hint 1: If the app report "Something went wrong, connection error out", please turn off your proxy and retry.\nHint 2: If you upload a large size of image like 10MB, it may take some time to upload and process. Please be patient and wait.'

NOTES = 'This app is adapted from <a href="https://github.com/THUDM/VisualGLM">https://github.com/THUDM/VisualGLM</a>. It would be recommended to check out the repo if you want to see the detail of our model and training process.'

import json
import requests
import base64


URL = os.environ.get("URL")

def process_image(image_prompt):
    image = Image.open(image_prompt)
    print(f"height:{image.height}, width:{image.width}")
    resized_image = image.resize((224, 224), )
    timestamp = int(time.time())
    file_ext = os.path.splitext(image_prompt)[1]
    filename = f"examples/{timestamp}{file_ext}"
    resized_image.save(filename)
    print(f"temporal filename {filename}")
    with open(filename, "rb") as image_file:
        encoded_img = str(base64.b64encode(image_file.read()), encoding='utf-8')
        os.remove(filename)
        return encoded_img


def is_chinese(text):
    zh_pattern = re.compile(u'[\u4e00-\u9fa5]+')
    return zh_pattern.search(text)


def post(
        input_text,
        temperature,
        top_p,
        image_prompt,
        result_previous        
        ):
    result_text = [(ele[0], ele[1]) for ele in result_previous]
    for i in range(len(result_text)-1, -1, -1):
        if result_text[i][0] == "":
            del result_text[i]
    print(f"history {result_text}")

    is_zh = is_chinese(input_text)

    if image_prompt is None:
        print("Image empty")
        if is_zh:
            result_text.append((input_text, '图片为空！请上传图片并重试。'))
        else:
            result_text.append((input_text, 'Image empty! Please upload a image and retry.'))
        return input_text, result_text
    elif input_text == "":
        print("Text empty")
        result_text.append((input_text, 'Text empty! Please enter text and retry.'))
        return "", result_text                

    headers = {
            "Content-Type": "application/json; charset=UTF-8",
            "User-Agent": "Mozilla/5.0 (Windows NT 6.1; WOW64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/67.0.3396.87 Safari/537.36",
        }
    if image_prompt:
        encoded_img = process_image(image_prompt)
    else:
        encoded_img = None

    print('开始请求...')
    data = json.dumps({
        'text': input_text,
        'image_prompt': encoded_img,
        'temperature': temperature,
        'top_p': top_p,
        'history': result_text
    })
    try:
        response = requests.request("POST", URL, headers=headers, data=data, timeout=(60, 100)).json()
    except Exception as e:
        print("error message", e)
        if is_zh:
            result_text.append((input_text, '超时！请稍等几分钟再重试。'))
        else:
            result_text.append((input_text, 'Timeout! Please wait a few minutes and retry.'))
        return "", result_text
    print('请求完毕...')

    answer = str(response['result'])
    result_text.append((input_text, answer))
    print(result_text)
    print('finished')
    return "", result_text


def clear_fn(value):
    return "", [("", "Hi, What do you want to know about this image?")], None

def clear_fn2(value):
    return [("", "Hi, What do you want to know about this image?")]


def main():
    gr.close_all()
    examples = []
    with open("./examples/example_inputs.jsonl") as f:
        for line in f:
            data = json.loads(line)
            examples.append(data)


    with gr.Blocks(css='style.css') as demo:
        
        gr.Markdown(DESCRIPTION)
        gr.Markdown(MAINTENANCE_NOTICE)


        with gr.Row():
            with gr.Column():
                with gr.Group():
                    input_text = gr.Textbox(label='Input Text', placeholder='Please enter text prompt below and press ENTER.')
                    with gr.Row():
                        run_button = gr.Button('Generate')
                        clear_button = gr.Button('Clear')

                    image_prompt = gr.Image(type="filepath", label="Image Prompt", value=None)
                with gr.Row():
                    temperature = gr.Slider(maximum=1, value=0.95, minimum=0, label='Temperature')
                    top_p = gr.Slider(maximum=1, value=0.7, minimum=0, label='Top P')

            result_text = gr.components.Chatbot(label='Multi-round conversation History', value=[("", "Hi, What do you want to know about this image?")])


        gr_examples = gr.Examples(examples=[[example["text"], example["image"]] for example in examples], 
                                  inputs=[input_text, image_prompt],
                                  label="Example Inputs (Click to insert an examplet into the input box)",
                                  examples_per_page=3)

        gr.Markdown(NOTES)

        print(gr.__version__)
        run_button.click(fn=post,inputs=[input_text, temperature, top_p, image_prompt, result_text],
                         outputs=[input_text, result_text])
        input_text.submit(fn=post,inputs=[input_text, temperature, top_p, image_prompt, result_text],
                         outputs=[input_text, result_text])
        clear_button.click(fn=clear_fn, inputs=clear_button, outputs=[input_text, result_text, image_prompt])
        image_prompt.change(fn=clear_fn2, inputs=clear_button, outputs=[result_text])

        print(gr.__version__)


    demo.queue(concurrency_count=10)
    demo.launch()


if __name__ == '__main__':
    main()