app.py

import subprocess

# Installing flash_attn
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"},
               shell=True)

from threading import Thread
import spaces
import gradio as gr
import torch
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    StoppingCriteria,
    StoppingCriteriaList,
    TextIteratorStreamer
)

model = AutoModelForCausalLM.from_pretrained("THUDM/longwriter-glm4-9b", trust_remote_code=True, device_map='auto')
tokenizer = AutoTokenizer.from_pretrained("THUDM/longwriter-glm4-9b", trust_remote_code=True)


class StopOnTokens(StoppingCriteria):
    def __call__(self, input_ids: torch.LongTensor, scores: torch.FloatTensor, **kwargs) -> bool:
        stop_ids = model.config.eos_token_id
        for stop_id in stop_ids:
            if input_ids[0][-1] == stop_id:
                return True
        return False


@spaces.GPU(duration=280)
def predict(history, prompt, max_length, top_p, temperature):
    stop = StopOnTokens()
    messages = []
    if prompt:
        messages.append({"role": "system", "content": prompt})
    for idx, (user_msg, model_msg) in enumerate(history):
        if prompt and idx == 0:
            continue
        if idx == len(history) - 1 and not model_msg:
            query = user_msg
            break
        if user_msg:
            messages.append({"role": "user", "content": user_msg})
        if model_msg:
            messages.append({"role": "assistant", "content": model_msg})

    model_inputs = tokenizer.build_chat_input(query, history=messages, role='user').input_ids.to(
        next(model.parameters()).device)
    streamer = TextIteratorStreamer(tokenizer, timeout=600, skip_prompt=True)
    eos_token_id = [tokenizer.eos_token_id, tokenizer.get_command("<|user|>"),
                    tokenizer.get_command("<|observation|>")]
    generate_kwargs = {
        "input_ids": model_inputs,
        "streamer": streamer,
        "max_new_tokens": max_length,
        "do_sample": True,
        "top_p": top_p,
        "temperature": temperature,
        "stopping_criteria": StoppingCriteriaList([stop]),
        "repetition_penalty": 1,
        "eos_token_id": eos_token_id,
    }
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()
    for new_token in streamer:
        if new_token and '<|user|>' not in new_token:
            history[-1][1] += new_token
        yield history


with gr.Blocks() as demo:
    gr.Markdown(
        """
        <div style="text-align: center; font-size: 32px; font-weight: bold; margin-bottom: 20px;">
            longwriter-glm4-9b Huggingface Space🤗
        </div>
        <div style="text-align: center;">
            <a href="https://huggingface.co/THUDM/LongWriter-glm4-9b">🤗 Model Hub</a> |
            <a href="https://github.com/THUDM/LongWriter">🌐 Github</a> |
            <a href="https://arxiv.org/pdf/2408.07055">📜 arxiv </a>
        </div>
        <div style="text-align: center; font-size: 15px; font-weight: bold; margin-bottom: 20px; line-height: 1.5;">
    <div style="color: black;">
        ⚠️ Due to the limitations of Huggingface ZERO GPUs, in order to output 5K characters in one go,
        we need to request a 4-5 minute quota each time.
        This will result in you only being able to use it once every 4 hours.
    </div>
    <br>
    <div style="color: red;">
        ⚠️ After 4-5 minutes, it will result in a timeout error, regardless of whether the output is complete. 
        This is not caused by the model.<br>
        If you plan to use it long-term, please consider deploying the model or forking this space yourself.
    </div>
</div>
        """
    )
    chatbot = gr.Chatbot()

    with gr.Row():
        with gr.Column(scale=3):
            with gr.Column(scale=12):
                user_input = gr.Textbox(show_label=False, placeholder="Input...(Example: Write a 10000-word China travel guide)", lines=10, container=False)
            with gr.Column(min_width=32, scale=1):
                submitBtn = gr.Button("Submit")
        with gr.Column(scale=1):
            prompt_input = gr.Textbox(show_label=False, placeholder="Prompt", lines=10, container=False)
            pBtn = gr.Button("Set Prompt")
        with gr.Column(scale=1):
            emptyBtn = gr.Button("Clear History")
            max_length = gr.Slider(0, 128000, value=10240, step=1.0, label="Maximum length(Input + Output)",
                                   interactive=True)
            top_p = gr.Slider(0, 1, value=0.8, step=0.01, label="Top P", interactive=True)
            temperature = gr.Slider(0.01, 1, value=0.6, step=0.01, label="Temperature", interactive=True)


    def user(query, history):
        return "", history + [[query, ""]]


    def set_prompt(prompt_text):
        return [[prompt_text, "Set prompt successfully"]]


    pBtn.click(set_prompt, inputs=[prompt_input], outputs=chatbot)

    submitBtn.click(user, [user_input, chatbot], [user_input, chatbot], queue=False).then(
        predict, [chatbot, prompt_input, max_length, top_p, temperature], chatbot
    )
    emptyBtn.click(lambda: (None, None), None, [chatbot, prompt_input], queue=False)

demo.queue()
demo.launch()