app.py

import os
import gradio as gr
from huggingface_hub import InferenceClient
from datasets import load_dataset
import random
import re

# Global datasets - load lazily
math_samples = None

def load_sample_problems():
    """Load sample problems from ALL datasets - FIXED VERSION"""
    global math_samples
    if math_samples is not None:
        return math_samples
    
    samples = []
    try:
        print("🔄 Loading GSM8K...")
        # GSM8K (math problems)
        gsm8k = load_dataset("openai/gsm8k", "main", streaming=True)
        gsm_count = 0
        for i, item in enumerate(gsm8k["train"]):
            samples.append(item["question"])
            gsm_count += 1
            if gsm_count >= 50:
                break
        
        print("🔄 Loading Fineweb-edu...")
        # Fineweb-edu (educational text - extract math-like questions)
        fw = load_dataset("HuggingFaceFW/fineweb-edu", name="sample-10BT", split="train", streaming=True)
        fw_count = 0
        for item in fw:
            # Filter for math-related content
            text_lower = item['text'].lower()
            if any(word in text_lower for word in ['math', 'calculate', 'solve', 'derivative', 'integral', 'triangle', 'equation', 'area', 'volume', 'probability']):
                # Truncate and format as question
                question = item['text'][:150].strip()
                if len(question) > 20:  # Ensure it's substantial
                    samples.append(question + " (Solve this math problem.)")
                    fw_count += 1
                    if fw_count >= 20:
                        break
        
        print("🔄 Loading Ultrachat...")
        # Ultrachat_200k (chat-like math queries)
        ds = load_dataset("HuggingFaceH4/ultrachat_200k", streaming=True)
        ds_count = 0
        for item in ds:
            if len(item['messages']) > 0:
                content = item['messages'][0]['content'].lower()
                if any(word in content for word in ['math', 'calculate', 'solve', 'problem', 'equation', 'derivative', 'integral']):
                    user_msg = item['messages'][0]['content']
                    if len(user_msg) > 10:  # Valid length
                        samples.append(user_msg)
                        ds_count += 1
                        if ds_count >= 20:
                            break
        
        print(f"✅ Loaded {len(samples)} samples: GSM8K ({gsm_count}), Fineweb-edu ({fw_count}), Ultrachat ({ds_count})")
        math_samples = samples
        return samples
        
    except Exception as e:
        print(f"⚠️ Dataset error: {e}, using fallback")
        math_samples = [
            "What is the derivative of f(x) = 3x² + 2x - 1?",
            "A triangle has sides of length 5, 12, and 13. What is its area?",
            "If log₂(x) + log₂(x+6) = 4, find the value of x.",
            "Find the limit: lim(x->0) (sin(x)/x)",
            "Solve the system: x + 2y = 7, 3x - y = 4",
            "Calculate the integral of sin(x) from 0 to pi.",
            "What is the probability of rolling a 6 on a die 3 times in a row?"
        ]
        return math_samples

def create_math_system_message():
    """Specialized system prompt for mathematics with LaTeX"""
    return r"""You are Mathetics AI, an advanced mathematics tutor and problem solver.

🧮 **Your Expertise:**
- Step-by-step problem solving with clear explanations
- Multiple solution approaches when applicable
- Proper mathematical notation and terminology using LaTeX
- Verification of answers through different methods

📐 **Problem Domains:**
- Arithmetic, Algebra, and Number Theory
- Geometry, Trigonometry, and Coordinate Geometry
- Calculus (Limits, Derivatives, Integrals)
- Statistics, Probability, and Data Analysis
- Competition Mathematics (AMC, AIME level)

💡 **Teaching Style:**
1. **Understand the Problem** - Identify what's being asked
2. **Plan the Solution** - Choose the appropriate method
3. **Execute Step-by-Step** - Show all work clearly with LaTeX formatting
4. **Verify the Answer** - Check if the result makes sense
5. **Alternative Methods** - Mention other possible approaches

**LaTeX Guidelines:**
- Use $...$ for inline math: $x^2 + y^2 = z^2$
- Use $$...$$ for display math
- Box final answers: \boxed{answer}
- Fractions: \frac{numerator}{denominator}
- Limits: \lim_{x \to 0}
- Derivatives: \frac{d}{dx} or f'(x)

Always be precise, educational, and encourage mathematical thinking."""

def render_latex(text):
    """Minimal LaTeX cleanup - let the model do the work"""
    if not text:
        return text
    
    try:
        # Convert LaTeX bracket notation to dollar signs
        text = re.sub(r'\\\[(.*?)\\\]', r'$$\1$$', text, flags=re.DOTALL)
        text = re.sub(r'\\\((.*?)\\\)', r'$\1$', text, flags=re.DOTALL)
        
        # Fix boxed answers ONLY if not already in math mode
        if '\\boxed' in text and not re.search(r'\$.*\\boxed.*\$', text):
            text = re.sub(r'\\boxed\{([^}]+)\}', r'$\\boxed{\1}$', text)
        
    except Exception as e:
        print(f"⚠️ LaTeX error: {e}")
    
    return text

def respond(message, history, system_message, max_tokens, temperature, top_p):
    """Streaming response with yield"""
    client = InferenceClient(model="Qwen/Qwen2.5-Math-7B-Instruct")
    
    messages = [{"role": "system", "content": system_message}]
    for msg in history:
        if msg["role"] == "user":
            messages.append({"role": "user", "content": msg["content"]})
        elif msg["role"] == "assistant":
            messages.append({"role": "assistant", "content": msg["content"]})
    messages.append({"role": "user", "content": message})
    
    response = ""
    try:
        for chunk in client.chat_completion(
            messages,
            max_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p,
            stream=True  # Enable streaming
        ):
            if chunk.choices[0].delta.content:
                response += chunk.choices[0].delta.content
                yield render_latex(response)  # Yield progressively
                
    except Exception as e:
        yield f"❌ Error: {str(e)[:100]}..."

def get_random_sample():
    """Get a random sample problem - loads datasets if needed"""
    global math_samples
    if math_samples is None:
        math_samples = load_sample_problems()
    return random.choice(math_samples)

def insert_sample_to_chat(difficulty):
    """Insert random sample into chat input"""
    return get_random_sample()

def show_help():
    return """**🧮 Math Help Tips:**

1. Be Specific: "Find the derivative of x² + 3x" instead of "help with calculus"
2. Request Steps: "Show me step-by-step how to solve..."
3. Ask for Verification: "Check if my answer x=5 is correct"
4. Alternative Methods: "What's another way to solve this integral?"
5. Use Clear Notation: "lim(x->0)" for limits

Pro Tip: Crank tokens to 1500+ for competition problems!"""

# Simple Chatbot interface
with gr.Blocks(title="🧮 Mathetics AI") as demo:
    gr.Markdown("# 🧮 **Mathetics AI** - Math Tutor\nPowered by Qwen 2.5-Math")
    
    chatbot = gr.Chatbot(height=500, label="Conversation", type='messages')
    help_text = gr.Markdown(visible=False)
    
    msg = gr.Textbox(placeholder="Ask a math problem...", show_label=False)
    
    with gr.Row():
        submit = gr.Button("Solve", variant="primary")
        clear = gr.Button("Clear", variant="secondary")
        sample = gr.Button("Random Problem", variant="secondary")
        help_btn = gr.Button("Help", variant="secondary")
    
    gr.Examples(
        examples=[
            ["derivative of x^2 sin(x)"],
            ["area of triangle 5-12-13"],
            ["∫x^2 dx"]
        ],
        inputs=msg
    )
    
    def chat_response(message, history):
    history.append({"role": "user", "content": message})
    history.append({"role": "assistant", "content": ""})
    
    for partial_response in respond(message, history[:-1], create_math_system_message(), 1024, 0.3, 0.85):
        history[-1]["content"] = partial_response
        yield history, ""
    
    def clear_chat():
        """Clear the chat history and textbox."""
        return [], ""
    
    msg.submit(chat_response, [msg, chatbot], [chatbot, msg])
    submit.click(chat_response, [msg, chatbot], [chatbot, msg])
    clear.click(clear_chat, outputs=[chatbot, msg])
    sample.click(insert_sample_to_chat, outputs=msg)
    help_btn.click(lambda: (show_help(), gr.update(visible=True)), outputs=[help_text, help_text]).then(
        lambda: gr.update(visible=False), outputs=help_text
    )

demo.launch()