app.py

"""
Production-grade Gradio Space for Myanmar LLM Code Assistant
Model: amkyawdev/mm-llm-coder-lite-v1
"""

import gradio as gr
from gradio import themes
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, GenerationConfig
from functools import lru_cache
import warnings
import time

warnings.filterwarnings("ignore")

import os

# ==================== CONFIGURATION ====================
MODEL_NAME = "amkyawdev/mm-llm-coder-lite-v1"
HF_TOKEN = os.environ.get("HF_TOKEN", "")  # Set in Space secrets
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"

# System Prompts
SYSTEM_PROMPTS = {
    "General Assistant": "သင်သည် မြန်မာစာ ကျွမ်းကျင်သော AI အကူအညီပေးသူဖြစ်သည်။ သင့်အား မြန်မာဘာသာဖြင့် ဖြေကြားပါ။",
    "Code Expert": "သင်သည် Senior Python Developer ဖြစ်သည်။ အဆင့်မြင့် Code များကို ရှင်းလင်းစွာ ရေးသားပါ။ မြန်မာဘာသာဖြင့် ဖြေကြားပါ။",
    "Translator": "သင်သည် မြန်မာ-အင်္ဂလိပ် ဘာသာပြန်ကျွမ်းကျင်သူဖြစ်သည်။ ဘာသာပြန်လုပ်ပါ။"
}

# ==================== MODEL LOADING ====================
@lru_cache(maxsize=1)
def load_model_and_tokenizer():
    """Load model and tokenizer with caching"""
    print(f"Loading model from {MODEL_NAME}...")
    
    try:
        # Try with trust_remote_code and different settings
        tokenizer = AutoTokenizer.from_pretrained(
            MODEL_NAME,
            trust_remote_code=True,
            use_fast=True,
            token=HF_TOKEN if HF_TOKEN else None
        )
    except Exception as e1:
        print(f"Fast tokenizer failed: {e1}, trying slow...")
        try:
            tokenizer = AutoTokenizer.from_pretrained(
                MODEL_NAME,
                trust_remote_code=True,
                use_fast=False,
                token=HF_TOKEN if HF_TOKEN else None
            )
        except Exception as e2:
            print(f"Slow tokenizer also failed: {e2}")
            raise
    
    # Handle missing pad token
    if tokenizer.pad_token is None:
        tokenizer.pad_token = tokenizer.eos_token
    
    # Load model with lower memory settings
    model = AutoModelForCausalLM.from_pretrained(
        MODEL_NAME,
        torch_dtype=torch.float32,
        device_map="cpu",
        low_cpu_mem_usage=True,
        trust_remote_code=True,
        token=HF_TOKEN if HF_TOKEN else None
    )
    
    print(f"Model loaded successfully")
    return model, tokenizer

# Initialize model at startup
try:
    model, tokenizer = load_model_and_tokenizer()
    MODEL_LOADED = True
except Exception as e:
    print(f"Error loading model: {e}")
    MODEL_LOADED = False
    model = None
    tokenizer = None

# ==================== GENERATION FUNCTIONS ====================
def format_prompt(user_message: str, system_prompt: str, history: list) -> str:
    """Format the prompt for the model"""
    prompt = f"System: {system_prompt}\n\n"
    
    for msg, response in history:
        prompt += f"User: {msg}\n\nAssistant: {response}{tokenizer.eos_token}\n\n"
    
    prompt += f"User: {user_message}\n\nAssistant:"
    return prompt

def generate_response(
    user_message: str,
    system_prompt: str,
    history: list,
    max_new_tokens: int,
    temperature: float,
    top_p: float
) -> tuple:
    """Generate response from the model"""
    
    if not MODEL_LOADED:
        return "❌ မော်ဒယ် မပါ။ ပြန်လည်ကြိုးစားပါ။", history
    
    try:
        # Format prompt
        prompt = format_prompt(user_message, system_prompt, history)
        
        # Tokenize
        inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
        
        # Use CPU if no CUDA
        if torch.cuda.is_available():
            inputs = {k: v.to(model.device) for k, v in inputs.items()}
        
        # Generate
        generation_config = GenerationConfig(
            max_new_tokens=max_new_tokens,
            temperature=temperature,
            top_p=top_p,
            do_sample=temperature > 0,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.pad_token_id,
        )
        
        with torch.no_grad():
            outputs = model.generate(
                **inputs,
                generation_config=generation_config
            )
        
        # Decode response
        response = tokenizer.decode(outputs[0], skip_special_tokens=True)
        
        # Extract only the assistant's response
        response = response.split("Assistant:")[-1].strip()
        
        # Update history
        history.append((user_message, response))
        
        return response, history
        
    except Exception as e:
        print(f"Generation error: {e}")
        return f"❌ အမှားဖြစ်ပါ။: {str(e)}", history

def clear_history():
    """Clear chat history"""
    return [], ""

# ==================== CUSTOM CSS ====================
CUSTOM_CSS = """
/* Premium Dark Theme */
:root {
    --primary: #10a37f;
    --secondary: #1a1a1a;
    --accent: #2d2d2d;
    --text-primary: #ffffff;
    --text-secondary: #a0a0a0;
    --user-bubble: #10a37f;
    --bot-bubble: #2d2d2d;
    --border-color: #404040;
}

/* Light Theme Overrides */
.light {
    --secondary: #ffffff;
    --accent: #f5f5f5;
    --text-primary: #1a1a1a;
    --text-secondary: #666666;
    --bot-bubble: #f0f0f0;
    --border-color: #e0e0e0;
}

/* Main Container */
.gradio-container {
    max-width: 1200px !important;
    margin: auto !important;
}

/* Header */
.header-section {
    text-align: center;
    padding: 20px;
    background: linear-gradient(135deg, #1a1a1a 0%, #2d2d2d 100%);
    border-radius: 16px;
    margin-bottom: 20px;
}

.header-title {
    font-size: 28px;
    font-weight: 700;
    background: linear-gradient(90deg, #10a37f, #00d4aa);
    -webkit-background-clip: text;
    -webkit-text-fill-color: transparent;
    margin-bottom: 8px;
}

.header-subtitle {
    color: #a0a0a0;
    font-size: 14px;
}

/* Chat Interface */
.chat-container {
    border-radius: 16px;
    overflow: hidden;
    box-shadow: 0 4px 20px rgba(0, 0, 0, 0.3);
}

.chat-message {
    padding: 12px 16px;
    margin: 8px 0;
    border-radius: 12px;
    max-width: 85%;
    line-height: 1.6;
}

.chat-message.user {
    background: linear-gradient(135deg, #10a37f, #0d8a66);
    color: white;
    margin-left: auto;
    border-bottom-right-radius: 4px;
}

.chat-message.bot {
    background: var(--bot-bubble);
    color: var(--text-primary);
    border-bottom-left-radius: 4px;
}

/* Code Blocks */
pre {
    background: #1e1e1e !important;
    border-radius: 8px;
    padding: 12px !important;
    margin: 12px 0 !important;
    overflow-x: auto;
}

code {
    font-family: 'Fira Code', 'Consolas', monospace;
    font-size: 13px;
}

/* Input Area */
.input-container {
    background: var(--accent);
    border-radius: 12px;
    padding: 12px;
    border: 1px solid var(--border-color);
}

.input-container:focus-within {
    border-color: #10a37f;
    box-shadow: 0 0 0 2px rgba(16, 163, 127, 0.2);
}

/* Buttons */
.btn-primary {
    background: linear-gradient(135deg, #10a37f, #0d8a66) !important;
    border: none !important;
    border-radius: 8px !important;
    padding: 10px 20px !important;
    font-weight: 600 !important;
    transition: all 0.3s ease !important;
}

.btn-primary:hover {
    transform: translateY(-2px);
    box-shadow: 0 4px 12px rgba(16, 163, 127, 0.4);
}

/* Example Buttons */
.example-btn {
    background: var(--accent) !important;
    border: 1px solid var(--border-color) !important;
    border-radius: 8px !important;
    padding: 8px 16px !important;
    transition: all 0.3s ease !important;
}

.example-btn:hover {
    background: #10a37f !important;
    color: white !important;
    border-color: #10a37f !important;
}

/* Sliders */
.slider-container label {
    color: var(--text-primary);
    font-weight: 500;
}

.slider-container .slider-value {
    color: #10a37f;
    font-weight: 600;
}

/* Dropdown */
.dropdown-container select {
    background: var(--accent);
    border: 1px solid var(--border-color);
    border-radius: 8px;
    padding: 8px 12px;
    color: var(--text-primary);
}

/* Loading Animation */
.loading-spinner {
    display: flex;
    justify-content: center;
    align-items: center;
    padding: 20px;
}

.loading-spinner::after {
    content: '';
    width: 40px;
    height: 40px;
    border: 3px solid var(--border-color);
    border-top-color: #10a37f;
    border-radius: 50%;
    animation: spin 1s linear infinite;
}

@keyframes spin {
    to { transform: rotate(360deg); }
}

/* Footer */
.footer {
    text-align: center;
    padding: 20px;
    color: #a0a0a0;
    font-size: 12px;
    border-top: 1px solid var(--border-color);
    margin-top: 20px;
}

.footer a {
    color: #10a37f;
    text-decoration: none;
}

.footer a:hover {
    text-decoration: underline;
}

/* Dark/Light Toggle */
.theme-toggle {
    display: flex;
    align-items: center;
    gap: 8px;
    padding: 8px 12px;
    background: var(--accent);
    border-radius: 20px;
    cursor: pointer;
}

/* Animations */
@keyframes fadeIn {
    from { opacity: 0; transform: translateY(10px); }
    to { opacity: 1; transform: translateY(0); }
}

.fade-in {
    animation: fadeIn 0.3s ease-out;
}

/* Scrollbar */
::-webkit-scrollbar {
    width: 8px;
    height: 8px;
}

::-webkit-scrollbar-track {
    background: var(--secondary);
}

::-webkit-scrollbar-thumb {
    background: var(--border-color);
    border-radius: 4px;
}

::-webkit-scrollbar-thumb:hover {
    background: #10a37f;
}
"""

# ==================== GRADIO APP ====================
# Simple version for testing

with gr.Blocks(title="Myanmar LLM") as app:
    
    # State for theme
    theme_state = gr.State(value="dark")
    
    # Header
    gr.HTML("""
    <div class="header-section">
        <div class="header-title"> Myanmar LLM Code Assistant</div>
        <div class="header-subtitle">Powered by amkyawdev/mm-llm-coder-lite-v1</div>
    </div>
    """)
    
    # Main Layout
    with gr.Row():
        with gr.Column(scale=3):
            # Chat Interface
            chatbot = gr.Chatbot(
                label="💬 စကားပြောပါ။",
                height=500,
            )
            
            # Example Prompts
            gr.HTML("<div style='text-align: center; margin: 10px 0; color: #a0a0a0;'>Example Prompts:</div>")
            with gr.Row():
                btn1 = gr.Button("🔢 Fibonacci", size="sm", variant="secondary")
                btn2 = gr.Button("🔤 Unicode → Zawgyi", size="sm", variant="secondary")
                btn3 = gr.Button("Data Cleaning", size="sm", variant="secondary")
            
            # Input Area
            with gr.Row():
                msg_input = gr.Textbox(
                    label="သင့်မေးခွန်း",
                    placeholder="မေးခွန်းရေးသားပါ။...",
                    lines=3,
                    scale=4
                )
                submit_btn = gr.Button(" စာပို့ပါ ", variant="primary", scale=1)
            
            # Clear Button
            clear_btn = gr.Button("🗑️ သန့်ရှင်းပါ။", variant="stop")
        
        with gr.Column(scale=1):
            # Settings Panel
            gr.HTML("<div style='text-align: center; margin-bottom: 15px;'><b>⚙️ အပြင်အဆင့်များ</b></div>")
            
            # System Prompt
            system_prompt = gr.Dropdown(
                choices=list(SYSTEM_PROMPTS.keys()),
                value="General Assistant",
                label="System Prompt",
                info="AI ရဲ့ အပြုအမူ"
            )
            
            # Max Tokens
            max_tokens = gr.Slider(
                minimum=50,
                maximum=512,
                value=256,
                step=10,
                label="Max New Tokens",
                info="အများဆုံး စကားပါးပါး"
            )
            
            # Temperature
            temperature = gr.Slider(
                minimum=0.1,
                maximum=1.5,
                value=0.7,
                step=0.1,
                label="Temperature",
                info="ပိုမိုးတော်တော် (0.1 = တိကျ၊ 1.5 = ဖန်းဆန်း)"
            )
            
            # Top-p
            top_p = gr.Slider(
                minimum=0.5,
                maximum=1.0,
                value=0.95,
                step=0.05,
                label="Top-p",
                info="Nucleus sampling"
            )
            
            # Theme info
            gr.HTML("<div style='text-align: center; color: #a0a0a0;'>🌙 Dark Mode</div>")
    
    # Footer
    gr.HTML("""
    <div class="footer">
        <p>🤖 ဤအက်ပလီမော်ဒယ်သည် <a href="https://huggingface.co/amkyawdev/mm-llm-coder-lite-v1" target="_blank">amkyawdev/mm-llm-coder-lite-v1</a> ဖြစ်ပါ။</p>
        <p>📊 Hardware: T4 Small (Recommended) | ဖန်တီးသူ: Amkyaw Dev </p>
    </div>
    """)
    
    # ==================== EVENT HANDLERS ====================
    
    def respond(
        message: str,
        history: list,
        system_prompt_key: str,
        max_tokens: int,
        temperature: float,
        top_p: float
    ):
        """Handle message submission"""
        if not message.strip():
            return "", history, gr.update()
        
        system_prompt = SYSTEM_PROMPTS.get(system_prompt_key, SYSTEM_PROMPTS["General Assistant"])
        
        response, history = generate_response(
            user_message=message,
            system_prompt=system_prompt,
            history=history,
            max_new_tokens=max_tokens,
            temperature=temperature,
            top_p=top_p
        )
        
        return "", history, gr.update()
    
    # Submit button click
    submit_btn.click(
        fn=respond,
        inputs=[
            msg_input,
            chatbot,
            system_prompt,
            max_tokens,
            temperature,
            top_p
        ],
        outputs=[msg_input, chatbot, chatbot]
    )
    
    # Enter key submit
    msg_input.submit(
        fn=respond,
        inputs=[
            msg_input,
            chatbot,
            system_prompt,
            max_tokens,
            temperature,
            top_p
        ],
        outputs=[msg_input, chatbot, chatbot]
    )
    
    # Clear button
    clear_btn.click(
        fn=clear_history,
        inputs=[],
        outputs=[chatbot, msg_input]
    )
    
    # Example buttons
    btn1.click(
        fn=lambda: ("Python နဲ့ Fibonacci စီးရီးထုတ်တဲ့ function ရေးပေးပါ", []),
        inputs=[],
        outputs=[msg_input, chatbot]
    )
    
    btn2.click(
        fn=lambda: ("မြန်မာ Unicode ကို Zawgyi ပြောင်းတဲ့ code ရေးပါ", []),
        inputs=[],
        outputs=[msg_input, chatbot]
    )
    
    btn3.click(
        fn=lambda: ("ဒေတာ (Data) သန့်ရှင်းရေးလုပ်နည်း အဆင့်ဆင့်ရှင်းပြပါ", []),
        inputs=[],
        outputs=[msg_input, chatbot]
    )
    
    # Theme toggle - simplified (remove for now)
    pass

# ==================== LAUNCH ====================
if __name__ == "__main__":
    print("🚀 Starting Myanmar LLM Code Assistant...")
    print(f"📱 Device: {DEVICE}")
    print(f"📦 Model: {MODEL_NAME}")
    
    app.launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=True
    )