app.py

import gradio as gr
import torch
import warnings
from transformers import AutoTokenizer, AutoModelForCausalLM
import random
import numpy as np

warnings.filterwarnings('ignore')

# 设置可复现的随机种子
def setup_seed(seed):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)
    torch.backends.cudnn.deterministic = True
    torch.backends.cudnn.benchmark = False

# 加载模型和分词器
model_path = "cmz1024/minimind-zero"  # 替换为你的模型路径
tokenizer = AutoTokenizer.from_pretrained(model_path)
model = AutoModelForCausalLM.from_pretrained(model_path, trust_remote_code=True)

# 将模型移至GPU（如果可用）
device = "cuda" if torch.cuda.is_available() else "cpu"
model = model.to(device).eval()

print(f'MiniMind模型参数量: {sum(p.numel() for p in model.parameters() if p.requires_grad) / 1e6:.2f}M(illion)')

# 生成文本函数
def generate_text(prompt, max_length=512, temperature=0.85, top_p=0.85, history_cnt=0):
    # 如果输入为空，则返回提示信息
    if not prompt.strip():
        return "请输入您的问题或指令..."
    
    # 设置随机种子
    setup_seed(random.randint(0, 2048))
    
    # 处理历史对话
    messages = []
    if history_cnt > 0 and 'chat_history' in globals():
        messages = chat_history[-history_cnt:] if len(chat_history) > 0 else []
    
    # 添加当前用户输入
    messages.append({"role": "user", "content": prompt})
    
    # 应用聊天模板
    new_prompt = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    
    # 对输入进行编码
    inputs = tokenizer(new_prompt, return_tensors="pt").to(device)
    input_length = inputs["input_ids"].shape[1]
    
    # 生成文本
    with torch.no_grad():
        outputs = model.generate(
            inputs["input_ids"],
            max_new_tokens=max_length,
            temperature=temperature,
            top_p=top_p,
            pad_token_id=tokenizer.pad_token_id,
            eos_token_id=tokenizer.eos_token_id
        )
    
    # 只解码新生成的部分
    generated_text = tokenizer.decode(outputs[0][input_length:], skip_special_tokens=True)
    
    # 更新对话历史
    if 'chat_history' in globals():
        chat_history.append({"role": "user", "content": prompt})
        chat_history.append({"role": "assistant", "content": generated_text})
    
    return generated_text

# 初始化全局对话历史
chat_history = []

# 清除对话历史的函数
def clear_history():
    global chat_history
    chat_history = []
    return "✅ 对话历史已清除"

# 自定义CSS样式
custom_css = """
:root {
    --primary-color: #4F46E5;
    --secondary-color: #6366F1;
    --accent-color: #818CF8;
    --background-color: #FFFFFF;
    --text-color: #1F2937;
    --card-bg: #F9FAFB;
    --border-radius: 12px;
    --shadow: 0 4px 6px -1px rgba(0, 0, 0, 0.1), 0 2px 4px -1px rgba(0, 0, 0, 0.06);
}

body {
    background-color: var(--background-color);
    color: var(--text-color);
    font-family: 'Inter', -apple-system, BlinkMacSystemFont, sans-serif;
}

.container {
    max-width: 1200px;
    margin: 0 auto;
}

.header-container {
    background: #FFFFFF;
    border-bottom: 2px solid #E5E7EB;
    border-radius: var(--border-radius);
    padding: 20px;
    margin-bottom: 24px;
    box-shadow: var(--shadow);
    color: var(--text-color);
    text-align: center;
}

.header-container h1 {
    font-size: 2.5rem;
    margin: 0;
    font-weight: 700;
    color: var(--primary-color);
}

.header-container p {
    opacity: 0.9;
    margin-top: 8px;
    font-size: 1.1rem;
}

.logo-pulse {
    display: inline-block;
    margin-right: 12px;
    animation: pulse 2s infinite;
}

@keyframes pulse {
    0% { transform: scale(1); }
    50% { transform: scale(1.1); }
    100% { transform: scale(1); }
}

.gradio-container {
    margin-top: 0 !important;
}

.main-card {
    background-color: var(--card-bg);
    border-radius: var(--border-radius);
    box-shadow: var(--shadow);
    padding: 24px;
    margin-bottom: 24px;
}

.input-area textarea, .output-area textarea {
    border-radius: 8px !important;
    border: 1px solid #E5E7EB !important;
    padding: 12px !important;
    font-size: 1rem !important;
    transition: all 0.3s ease;
}

.input-area textarea:focus, .output-area textarea:focus {
    border-color: var(--primary-color) !important;
    box-shadow: 0 0 0 2px rgba(79, 70, 229, 0.2) !important;
}

.button-primary {
    background: linear-gradient(135deg, var(--primary-color), var(--secondary-color)) !important;
    border: none !important;
    color: white !important;
    padding: 10px 20px !important;
    border-radius: 8px !important;
    font-weight: 600 !important;
    transition: all 0.3s ease !important;
    box-shadow: 0 2px 4px rgba(79, 70, 229, 0.3) !important;
}

.button-primary:hover {
    transform: translateY(-2px) !important;
    box-shadow: 0 4px 8px rgba(79, 70, 229, 0.4) !important;
}

.button-secondary {
    background-color: #F3F4F6 !important;
    border: 1px solid #E5E7EB !important;
    color: var(--text-color) !important;
    padding: 10px 20px !important;
    border-radius: 8px !important;
    font-weight: 600 !important;
    transition: all 0.3s ease !important;
}

.button-secondary:hover {
    background-color: #E5E7EB !important;
    transform: translateY(-2px) !important;
}

.footer {
    text-align: center;
    margin-top: 40px;
    color: #6B7280;
    font-size: 0.9rem;
}

.slider-container label span {
    font-weight: 600 !important;
    color: var(--text-color) !important;
}

.accordion {
    border: 1px solid #E5E7EB !important;
    border-radius: var(--border-radius) !important;
    overflow: hidden !important;
}

.accordion-header {
    background-color: #F9FAFB !important;
    padding: 12px 16px !important;
    font-weight: 600 !important;
}

.status-box {
    background-color: #F3F4F6;
    border-radius: 8px;
    padding: 12px;
    font-size: 0.9rem;
    color: #4B5563;
}

.status-box.success {
    background-color: #ECFDF5;
    color: #065F46;
}

/* 响应式调整 */
@media (max-width: 768px) {
    .header-container h1 {
        font-size: 2rem;
    }
    
    .main-card {
        padding: 16px;
    }
}
"""

# 创建Gradio界面
with gr.Blocks(css=custom_css) as demo:
    with gr.Column(elem_classes="container"):
        with gr.Column(elem_classes="header-container"):
            gr.HTML("""
            <div>
                <span class="logo-pulse">🧠</span>
                <h1>MiniMind AI</h1>
                <p>小参数模型</p>
            </div>
            """)
        
        with gr.Column(elem_classes="main-card"):
            with gr.Row():
                with gr.Column(scale=1, elem_classes="input-area"):
                    input_text = gr.Textbox(
                        label="您的问题",
                        placeholder="请在此输入您的问题或指令...",
                        lines=5,
                        elem_id="input-box"
                    )
                    
                    with gr.Row():
                        submit_btn = gr.Button("🚀 生成回答", elem_classes="button-primary")
                        clear_btn = gr.Button("🗑️ 清除历史", elem_classes="button-secondary")
                    
                    with gr.Accordion("⚙️ 高级参数设置", open=False, elem_classes="accordion"):
                        max_length = gr.Slider(
                            minimum=10, maximum=2048, value=512, step=1,
                            label="最大生成长度",
                            info="控制生成文本的最大长度",
                            elem_classes="slider-container"
                        )
                        temperature = gr.Slider(
                            minimum=0.1, maximum=1.5, value=0.85, step=0.01,
                            label="温度系数",
                            info="较高的值会使输出更加随机，较低的值使输出更加确定",
                            elem_classes="slider-container"
                        )
                        top_p = gr.Slider(
                            minimum=0.1, maximum=1.0, value=0.85, step=0.01,
                            label="Top-p 采样",
                            info="控制词汇选择的多样性",
                            elem_classes="slider-container"
                        )
                        history_cnt = gr.Slider(
                            minimum=0, maximum=10, value=0, step=2,
                            label="历史对话轮数",
                            info="考虑多少轮历史对话作为上下文",
                            elem_classes="slider-container"
                        )
                
                with gr.Column(scale=1, elem_classes="output-area"):
                    output_text = gr.Textbox(
                        label="AI 回答",
                        lines=25,
                        elem_id="output-box"
                    )
                    clear_output = gr.Textbox(
                        label="状态",
                        value="系统就绪，等待您的输入...",
                        elem_classes="status-box"
                    )
        
        with gr.Column(elem_classes="footer"):
            gr.HTML("""
            <div>
                <p>MiniMind 模型演示 | 基于先进的自然语言处理技术</p>
                <p>运行环境：<span id="device-info"></span></p>
            </div>
            <script>
                document.getElementById('device-info').innerText = 
                    navigator.userAgent.includes('Mobile') ? '移动设备' : '桌面设备';
                
                // 添加输入动画效果
                document.addEventListener('DOMContentLoaded', function() {
                    const inputBox = document.getElementById('input-box');
                    const outputBox = document.getElementById('output-box');
                    
                    if(inputBox) {
                        inputBox.addEventListener('focus', function() {
                            this.style.transform = 'translateY(-2px)';
                            this.style.boxShadow = '0 4px 8px rgba(0, 0, 0, 0.1)';
                        });
                        
                        inputBox.addEventListener('blur', function() {
                            this.style.transform = 'translateY(0)';
                            this.style.boxShadow = '0 1px 3px rgba(0, 0, 0, 0.1)';
                        });
                    }
                });
            </script>
            """)
    
    # 设置事件
    submit_btn.click(
        fn=generate_text,
        inputs=[input_text, max_length, temperature, top_p, history_cnt],
        outputs=output_text
    )
    
    # 添加回车键触发生成回答的功能
    input_text.submit(
        fn=generate_text,
        inputs=[input_text, max_length, temperature, top_p, history_cnt],
        outputs=output_text
    )
    
    clear_btn.click(
        fn=clear_history,
        inputs=[],
        outputs=clear_output
    )

# 启动应用
demo.launch()