app.py

import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# 加载模型
model_name = "deepseek-ai/deepseek-coder-1.3b-base"  # 可替换为 "deepseek-ai/deepseek-coder-1.3b-instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.float16,  # 使用 FP16 减少内存
    device_map="cpu",          # 强制 CPU
    trust_remote_code=True,
    low_cpu_mem_usage=True
)

def respond(

    message,

    history: list[tuple[str, str]],

    system_message,

    max_tokens,

    temperature,

    top_p,

):
    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

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

    # 使用聊天模板格式化输入（base 模型可能无模板，需调整）
    try:
        input_text = tokenizer.apply_chat_template(messages, tokenize=False)
    except:
        # 如果 base 模型无聊天模板，直接拼接
        input_text = "\n".join([f"{msg['role']}: {msg['content']}" for msg in messages])

    inputs = tokenizer(input_text, return_tensors="pt").to("cpu")

    # 生成响应
    outputs = model.generate(
        **inputs,
        max_new_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    yield response

# Gradio 界面
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly coding assistant.", label="System message"),
        gr.Slider(minimum=1, maximum=2048, value=256, step=1, label="Max new tokens"),  # 降低以加快 CPU 推理
        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)",
        ),
    ],
)

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