app.py

# -*- coding: utf-8 -*-
"""
UltraData-Math L3 Generator - Hugging Face Space Demo
"""

import os
import asyncio
import json
import gradio as gr

from openai import AsyncOpenAI

from qa_synthesis import QA_PROMPTS, get_qa_prompt
from conversation_synthesis import CONVERSATION_PROMPTS, get_conversation_prompt
from multistyle_rewrite import MULTISTYLE_PROMPTS, get_multistyle_prompt
from knowledge_textbook import (
    get_knowledge_extraction_prompt,
    get_textbook_exercise_prompt,
    TEXTBOOK_EXERCISE_PROMPTS,
)
from run_synthesis import (
    parse_qa_output,
    parse_conversation_output,
    parse_rewrite_output,
    parse_knowledge_output,
    parse_textbook_output,
)

# API 配置从环境变量读取（通过 HF Secrets 设置）
API_KEY = os.getenv("OPENAI_API_KEY")
BASE_URL = os.getenv("OPENAI_BASE_URL", "https://llm-center.ali.modelbest.cn/llm/openai/v1")
DEFAULT_MODEL = "GLM_ar7snd"


async def call_api(prompt: str, model: str = DEFAULT_MODEL, temperature: float = 0.7) -> str:
    """调用 API 生成内容"""
    if not API_KEY:
        return "Error: API Key not configured. Please contact administrator."
    
    client = AsyncOpenAI(api_key=API_KEY, base_url=BASE_URL)
    try:
        response = await client.chat.completions.create(
            model=model,
            messages=[{"role": "user", "content": prompt}],
            temperature=temperature,
            max_tokens=8192,
        )
        # 处理 reasoning model 的返回格式
        message = response.choices[0].message
        content = message.content
        # 如果 content 为空，尝试获取 reasoning_content
        if not content and hasattr(message, 'reasoning_content') and message.reasoning_content:
            content = message.reasoning_content
        return content or ""
    except Exception as e:
        return f"Error: {str(e)}"


def run_async(coro):
    """运行异步函数"""
    try:
        loop = asyncio.get_event_loop()
    except RuntimeError:
        loop = asyncio.new_event_loop()
        asyncio.set_event_loop(loop)
    return loop.run_until_complete(coro)


# ============================================================================
# Task Handlers
# ============================================================================

def qa_synthesis(text: str, level: str, model: str, temperature: float):
    """Q&A 问答对合成"""
    if not text.strip():
        return "", "", ""
    
    prompt_template = get_qa_prompt(level)
    prompt = prompt_template.format(text=text)
    
    response = run_async(call_api(prompt, model, temperature))
    parsed = parse_qa_output(response)
    
    return (
        parsed.get("problem", ""),
        parsed.get("solution", ""),
        response
    )


def conversation_synthesis(text: str, style: str, model: str, temperature: float):
    """多轮对话合成"""
    if not text.strip():
        return "", ""
    
    prompt_template = get_conversation_prompt(style)
    prompt = prompt_template.format(text=text)
    
    response = run_async(call_api(prompt, model, temperature))
    parsed = parse_conversation_output(response)
    
    return parsed.get("content", response), response


def rewrite_synthesis(text: str, style: str, model: str, temperature: float):
    """多风格改写"""
    if not text.strip():
        return "", ""
    
    prompt_template = get_multistyle_prompt(style)
    prompt = prompt_template.format(text=text)
    
    response = run_async(call_api(prompt, model, temperature))
    parsed = parse_rewrite_output(response)
    
    return parsed.get("rewritten", response), response


def knowledge_extraction(text: str, model: str, temperature: float):
    """知识点提取"""
    if not text.strip():
        return "", ""
    
    prompt_template = get_knowledge_extraction_prompt()
    prompt = prompt_template.format(text=text)
    
    response = run_async(call_api(prompt, model, temperature))
    parsed = parse_knowledge_output(response)
    
    knowledge_points = parsed.get("knowledge_points", [])
    formatted = "\n\n---\n\n".join(knowledge_points) if knowledge_points else "No knowledge points extracted."
    
    return formatted, response


def textbook_exercise(knowledge_point: str, difficulty: str, model: str, temperature: float):
    """教材练习生成"""
    if not knowledge_point.strip():
        return "", ""
    
    prompt_template = get_textbook_exercise_prompt(difficulty)
    prompt = prompt_template.format(mathematical_knowledge_point=knowledge_point)
    
    response = run_async(call_api(prompt, model, temperature))
    parsed = parse_textbook_output(response)
    
    return parsed.get("material", response), response


# ============================================================================
# Gradio UI
# ============================================================================

custom_css = """
.gradio-container {
    font-family: 'Segoe UI', Tahoma, Geneva, Verdana, sans-serif !important;
    background: linear-gradient(135deg, #1a1a2e 0%, #16213e 50%, #0f3460 100%) !important;
}

.main-title {
    font-weight: 700 !important;
    font-size: 2.2rem !important;
    background: linear-gradient(90deg, #e94560, #f39c12, #00d9ff) !important;
    -webkit-background-clip: text !important;
    -webkit-text-fill-color: transparent !important;
    background-clip: text !important;
    text-align: center !important;
}

.subtitle {
    text-align: center !important;
    color: #94a3b8 !important;
    font-size: 1rem !important;
    margin-bottom: 1.5rem !important;
}

.gr-button-primary {
    background: linear-gradient(135deg, #e94560 0%, #f39c12 100%) !important;
    border: none !important;
    font-weight: 600 !important;
}

.gr-button-primary:hover {
    transform: translateY(-2px) !important;
    box-shadow: 0 8px 25px rgba(233, 69, 96, 0.4) !important;
}

footer {
    display: none !important;
}
"""

with gr.Blocks(title="UltraData-Math L3 Generator", css=custom_css) as demo:
    gr.HTML('<h1 class="main-title">🧮 UltraData-Math L3 Generator</h1>')
    gr.HTML('<p class="subtitle">LLM-based Mathematical Data Synthesis Tool</p>')
    
    with gr.Row():
        model_select = gr.Dropdown(
            choices=["GLM_ar7snd", "GLM_pq0dvd", "GLM_35a7cn", "QWEN_czrd3t", "DEEPSEEK_5jcwxs"],
            value="GLM_ar7snd",
            label="Model",
            scale=1,
        )
        temperature = gr.Slider(
            minimum=0.0, maximum=1.5, value=0.7, step=0.1,
            label="Temperature",
            scale=1,
        )
    
    with gr.Tabs():
        # Q&A Synthesis Tab
        with gr.TabItem("📝 Q&A Synthesis"):
            gr.Markdown("根据数学内容生成问答对，按教育难度分级。")
            with gr.Row():
                with gr.Column():
                    qa_input = gr.Textbox(
                        label="Input Mathematical Content",
                        placeholder="Enter mathematical content here...",
                        lines=8,
                    )
                    qa_level = gr.Radio(
                        choices=list(QA_PROMPTS.keys()),
                        value="high_school",
                        label="Difficulty Level",
                    )
                    qa_btn = gr.Button("🚀 Generate Q&A", variant="primary")
                with gr.Column():
                    qa_problem = gr.Textbox(label="Generated Problem", lines=4)
                    qa_solution = gr.Textbox(label="Generated Solution", lines=8)
                    qa_raw = gr.Textbox(label="Raw Response", lines=4, visible=False)
            
            qa_btn.click(
                qa_synthesis,
                inputs=[qa_input, qa_level, model_select, temperature],
                outputs=[qa_problem, qa_solution, qa_raw],
            )
        
        # Conversation Synthesis Tab
        with gr.TabItem("💬 Conversation Synthesis"):
            gr.Markdown("将数学内容转换为多轮对话格式。")
            with gr.Row():
                with gr.Column():
                    conv_input = gr.Textbox(
                        label="Input Mathematical Content",
                        placeholder="Enter mathematical content here...",
                        lines=8,
                    )
                    conv_style = gr.Radio(
                        choices=list(CONVERSATION_PROMPTS.keys()),
                        value="teacher_student",
                        label="Conversation Style",
                    )
                    conv_btn = gr.Button("🚀 Generate Conversation", variant="primary")
                with gr.Column():
                    conv_output = gr.Textbox(label="Generated Conversation", lines=15)
                    conv_raw = gr.Textbox(label="Raw Response", lines=4, visible=False)
            
            conv_btn.click(
                conversation_synthesis,
                inputs=[conv_input, conv_style, model_select, temperature],
                outputs=[conv_output, conv_raw],
            )
        
        # Rewrite Tab
        with gr.TabItem("✨ Multi-style Rewrite"):
            gr.Markdown("将数学内容改写为不同风格。")
            with gr.Row():
                with gr.Column():
                    rewrite_input = gr.Textbox(
                        label="Input Mathematical Content",
                        placeholder="Enter mathematical content here...",
                        lines=8,
                    )
                    rewrite_style = gr.Radio(
                        choices=list(MULTISTYLE_PROMPTS.keys()),
                        value="textbook",
                        label="Rewrite Style",
                    )
                    rewrite_btn = gr.Button("🚀 Rewrite", variant="primary")
                with gr.Column():
                    rewrite_output = gr.Textbox(label="Rewritten Content", lines=15)
                    rewrite_raw = gr.Textbox(label="Raw Response", lines=4, visible=False)
            
            rewrite_btn.click(
                rewrite_synthesis,
                inputs=[rewrite_input, rewrite_style, model_select, temperature],
                outputs=[rewrite_output, rewrite_raw],
            )
        
        # Knowledge Extraction Tab
        with gr.TabItem("📚 Knowledge Extraction"):
            gr.Markdown("从数学内容中提取定义、定理、性质等知识点。")
            with gr.Row():
                with gr.Column():
                    know_input = gr.Textbox(
                        label="Input Mathematical Content",
                        placeholder="Enter mathematical content here...",
                        lines=10,
                    )
                    know_btn = gr.Button("🚀 Extract Knowledge", variant="primary")
                with gr.Column():
                    know_output = gr.Textbox(label="Extracted Knowledge Points", lines=15)
                    know_raw = gr.Textbox(label="Raw Response", lines=4, visible=False)
            
            know_btn.click(
                knowledge_extraction,
                inputs=[know_input, model_select, temperature],
                outputs=[know_output, know_raw],
            )
        
        # Textbook Exercise Tab
        with gr.TabItem("📖 Textbook Exercise"):
            gr.Markdown("基于知识点生成不同难度的教材式练习。")
            with gr.Row():
                with gr.Column():
                    textbook_input = gr.Textbox(
                        label="Input Knowledge Point",
                        placeholder="Enter a mathematical knowledge point...",
                        lines=6,
                    )
                    textbook_diff = gr.Radio(
                        choices=list(TEXTBOOK_EXERCISE_PROMPTS.keys()),
                        value="easy",
                        label="Difficulty",
                    )
                    textbook_btn = gr.Button("🚀 Generate Exercise", variant="primary")
                with gr.Column():
                    textbook_output = gr.Textbox(label="Generated Exercise Material", lines=15)
                    textbook_raw = gr.Textbox(label="Raw Response", lines=4, visible=False)
            
            textbook_btn.click(
                textbook_exercise,
                inputs=[textbook_input, textbook_diff, model_select, temperature],
                outputs=[textbook_output, textbook_raw],
            )
    
    gr.HTML("""
    <div style="text-align: center; margin-top: 2rem; padding: 1rem; color: #64748b; font-size: 0.85rem;">
        <p>🔬 <strong>UltraData-Math L3 Generator</strong> - Part of the UltraData-Math Project</p>
        <p>LLM-based data synthesis for Q&A, conversations, rewriting, and more.</p>
    </div>
    """)


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