app.py

from PyPDF2 import PdfReader
import gradio as gr
import openai
import os

 # The first line contains the OpenAI key, while the second line provides the OpenAI URL, which is useful when the OpenAI server is hidden behind a proxy server.
 # eg. first line "sk-xxxxxxxxxx", second line "http://PROXY-URL"
if os.path.isfile('config'):
    config = open("config").readlines()
else:
    config = ""
api_key_from_config = ""
if len(config) > 0 and len(config[0].strip()) > 0:
    api_key_from_config = config[0].strip()
if len(config) > 1 and len(config[1].strip()) > 0:
    openai.api_base = config[1].strip()

# config
DEBUG = True

'''
 gradio: [['first question', 'No'], ['second question', 'Yes']]
 openai: [{"role": "user", "content": "first question"}, {"role": "assistant", "content": "No"}
          {"role": "user", "content": "second question"}, {"role": "assistant", "content": "Yes"}]
'''
def gradio_messages_to_openai_messages(g):
    result = []
    for pair in g:
        result.append({"role": "user", "content": pair[0]})
        result.append({"role": "assistant", "content": pair[1]})
    return result

def respond(chat_history, message, system_message, key_txt, url_txt, model, temperature):
    messages = [
            {"role": "system", "content": system_message},
            *gradio_messages_to_openai_messages(chat_history),
            {"role": "user", "content": message}
    ] 
    openai.api_key = key_txt if key_txt else api_key_from_config
    if url_txt:
        openai.api_base = url_txt
    if DEBUG:
        print("messages:", messages)
        print("model:", model)
        print("temperature:", temperature)
    completion = openai.ChatCompletion.create(
        model=model,
        messages=messages,
        temperature=temperature,
    )
    if DEBUG:
        print("completion:", completion)
    response = completion['choices'][0]['message']['content']
    result = chat_history + [[message, response]]
    return result

def parse_pdf(prompt, pdfs, system_message, key_txt, url_txt, model, temperature):
    result = ""
    full_text = ""
    for pdf in pdfs:
        print("parse: ", pdf)
        text = ""
        reader = PdfReader(pdf.name)
        for page in reader.pages:
            text = text + page.extract_text()
        full_text = text + "\n----------\n"
    messages = [
            {"role": "system", "content": system_message},
            {"role": "user", "content": prompt + "\n\n###\n\n " + full_text}
    ]
    openai.api_key = key_txt if key_txt else api_key_from_config
    if url_txt:
        openai.api_base = url_txt
    if DEBUG:
        print("messages:", messages)
        print("model:", model)
        print("temperature:", temperature)
    completion = openai.ChatCompletion.create(
        model=model,
        messages=messages,
        temperature=temperature,
    )
    if DEBUG:
        print("completion:", completion)
    response = completion['choices'][0]['message']['content']

    return response

with gr.Blocks() as demo:
    with gr.Tab("Config"):
        with gr.Row():
            key_txt = gr.Textbox(label = "Openai Key", placeholder="Enter openai key 'sk-xxxx'%s" %
                    (", Leave empty to use value from config file" if api_key_from_config else ""))
            url_txt = gr.Textbox(label = "Openai API Base URL", placeholder="Enter openai base url 'https://xxx', Leave empty to use value '%s'" % openai.api_base)
        system_message = gr.Textbox(label = "System Message:", value = "You are an assistant who gives brief and concise answers.")
        model = gr.Dropdown(label="Model", choices=["gpt-3.5-turbo", "gpt-3.5-turbo-0301", "gpt-4"], multiselect=False, value="gpt-3.5-turbo", type="value")
        temperature = gr.Slider(0, 2, value=1, label="Temperature", step=0.1, info="What sampling temperature to use, between 0 and 2. Higher values like 0.8 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.")
    with gr.Tab("Chat"):
        gr.Markdown("## Chat with GPT")
        chatbot = gr.Chatbot()
        message = gr.Textbox(label = "Message:", placeholder="Enter text and press 'Send'")
        message.submit(
            respond,
            [chatbot, message, system_message, key_txt, url_txt, model, temperature],
            chatbot,
        )
        with gr.Row():
            clear = gr.Button("Clear")
            clear.click(lambda: None, None, chatbot)
            send = gr.Button("Send")
            send.click(
                respond,
                [chatbot, message, system_message, key_txt, url_txt, model, temperature],
                chatbot,
            )
    with gr.Tab("PDF"):
        gr.Markdown("## Parse PDF with GPT")
        prompt = gr.Text(label="Prompt")
        pdfs = gr.File(label="Upload PDF", file_count="multiple", file_types=[".pdf"])
        markdown = gr.Markdown(label="Output")
        with gr.Row():
            clear = gr.Button("Clear")
            clear.click(lambda: None, None, markdown)
            submit = gr.Button("Upload")
            submit.click(
                    parse_pdf,
                    [prompt, pdfs, system_message, key_txt, url_txt, model, temperature],
                    markdown
            )


if __name__ == "__main__":
    demo.launch()
else:
    # run with "gradio app.py"
    demo.launch(server_name="0.0.0.0")