app.py

from __future__ import annotations
import spaces
import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import whoami
import os

os.removedirs('/data-nvme/zerogpu-offload/')

# 定义系统提示语
system_prompt = """你是 Skywork-o1，Skywork AI 开发的思维模型，擅长通过深度思考解决涉及数学、编码和逻辑推理的复杂问题。面对用户请求时，你首先会进行一段漫长而深入的思考过程，探索问题的可能解决方案。完成思考后，你会在回复中详细解释解决过程。"""

# 初始化模型和分词器
model_name = "Skywork/Skywork-o1-Open-Llama-3.1-8B"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# 定义生成回复的函数
@spaces.GPU
def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # 构造对话历史
    conversation = [{"role": "system", "content": system_message}]
    for user_msg, assistant_msg in history:
        if user_msg:
            conversation.append({"role": "user", "content": user_msg})
        if assistant_msg:
            conversation.append({"role": "assistant", "content": assistant_msg})

    conversation.append({"role": "user", "content": message})

    # 构造输入
    input_ids = tokenizer.apply_chat_template(
        conversation,
        tokenize=True,
        add_generation_prompt=True,
        return_tensors="pt"
    ).to(model.device)

    # 模型生成
    generation = model.generate(
        input_ids=input_ids,
        max_new_tokens=max_tokens,
        do_sample=True,
        temperature=temperature,
        top_p=top_p,
        pad_token_id=tokenizer.pad_token_id,
    )

    # 解码生成内容
    completion = tokenizer.decode(
        generation[0][len(input_ids[0]):],
        skip_special_tokens=True,
        clean_up_tokenization_spaces=True
    )
    return completion

# 定义Gradio界面
demo = gr.ChatInterface(
    fn=respond,
    additional_inputs=[
        gr.Textbox(value=system_prompt, label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(
            minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"
        ),
    ],
    # chatbot_style="default"
)

def hello(profile: gr.OAuthProfile | None) -> str:
    if profile is None:
        return "I don't know you."
    return f"Hello {profile.name}"

def list_organizations(oauth_token: gr.OAuthToken | None) -> str:
    if oauth_token is None:
        return "Please deploy this on Spaces and log in to list organizations."
    org_names = [org["name"] for org in whoami(oauth_token.token)["orgs"]]
    return f"You belong to {', '.join(org_names)}."

with gr.Blocks() as demo:
    gr.LoginButton()
    m1 = gr.Markdown()
    m2 = gr.Markdown()
    demo.load(hello, inputs=None, outputs=m1)
    demo.load(list_organizations, inputs=None, outputs=m2)

if __name__ == "__main__":
    demo.launch()