app.py

#!/usr/bin/env python3
"""
MiloMusic - Hugging Face Spaces Version
AI-powered music generation platform optimized for cloud deployment with high-performance configuration.
"""

import os
import sys
import subprocess
import tempfile
import gradio as gr
import soundfile as sf
from dataclasses import dataclass, field
from typing import Any
import xxhash
import numpy as np
import spaces
import groq

# Import environment setup for Spaces
def setup_spaces_environment():
    """Setup environment variables and paths for Hugging Face Spaces"""
    # Set HuggingFace cache directory
    os.environ["HF_HOME"] = "/tmp/hf_cache"
    os.environ["TRANSFORMERS_CACHE"] = "/tmp/transformers_cache"
    os.environ["HF_HUB_CACHE"] = "/tmp/hf_hub_cache"
    
    # PyTorch CUDA memory optimization
    os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "expandable_segments:True"
    
    # Set temp directory for audio files
    os.environ["TMPDIR"] = "/tmp"
    
    print("🚀 Environment setup complete for Spaces")

# Install flash-attn if not already installed
def install_flash_attn():
    """Install flash-attn from source with proper compilation flags"""
    try:
        import flash_attn
        print("✅ flash-attn already installed")
        return True
    except ImportError:
        print("📦 Installing flash-attn from source...")
        try:
            # Install with optimized settings for Spaces
            cmd = [
                sys.executable, "-m", "pip", "install", 
                "--no-build-isolation", 
                "--no-cache-dir",
                "flash-attn",
                "--verbose"
            ]
            
            # Use more parallel jobs for faster compilation in Spaces
            env = os.environ.copy()
            env["MAX_JOBS"] = "4"  # Utilize more CPU cores
            env["NVCC_PREPEND_FLAGS"] = "-ccbin /usr/bin/gcc"
            
            result = subprocess.run(cmd, env=env, capture_output=True, text=True, timeout=1800)  # 30 min timeout
            
            if result.returncode == 0:
                print("✅ flash-attn installed successfully")
                return True
            else:
                print(f"❌ flash-attn installation failed: {result.stderr}")
                return False
                
        except subprocess.TimeoutExpired:
            print("⏰ flash-attn installation timed out")
            return False
        except Exception as e:
            print(f"❌ Error installing flash-attn: {e}")
            return False

# Setup environment first
setup_spaces_environment()

# Install flash-attn if needed
flash_attn_available = install_flash_attn()

# Apply transformers patches for performance optimization
def apply_transformers_patch():
    """
    Apply YuEGP transformers patches for high-performance generation.
    
    This function applies optimized transformers patches that provide:
    - 2x speed improvement for low VRAM profiles
    - 3x speed improvement for Stage 1 generation (16GB+ VRAM)
    - 2x speed improvement for Stage 2 generation (all profiles)
    
    The patches replace two key files in the transformers library:
    - models/llama/modeling_llama.py (LLaMA model optimizations)
    - generation/utils.py (generation utilities optimizations)
    
    Includes smart detection to avoid re-applying patches on restart.
    """
    try:
        import shutil
        import site
        import hashlib
        
        # Define source and target directories
        source_dir = os.path.join(project_root, "YuEGP", "transformers")
        
        # Get the site-packages directory where transformers is installed
        site_packages = site.getsitepackages()
        if not site_packages:
            # Fallback for some environments
            import transformers
            transformers_path = os.path.dirname(transformers.__file__)
            target_base = os.path.dirname(transformers_path)
        else:
            target_base = site_packages[0]
        
        target_dir = os.path.join(target_base, "transformers")
        
        # Check if source patches exist
        if not os.path.exists(source_dir):
            print("⚠️  YuEGP transformers patches not found, skipping optimization")
            return False
        
        if not os.path.exists(target_dir):
            print("⚠️  Transformers library not found, skipping patches")
            return False
        
        # Check if patches are already applied by comparing file hashes
        def get_file_hash(filepath):
            """Get MD5 hash of file content"""
            if not os.path.exists(filepath):
                return None
            with open(filepath, 'rb') as f:
                return hashlib.md5(f.read()).hexdigest()
        
        # Key files to check for patch status
        key_patches = [
            "models/llama/modeling_llama.py",
            "generation/utils.py"
        ]
        
        patches_needed = False
        for patch_file in key_patches:
            source_file = os.path.join(source_dir, patch_file)
            target_file = os.path.join(target_dir, patch_file)
            
            if os.path.exists(source_file):
                source_hash = get_file_hash(source_file)
                target_hash = get_file_hash(target_file)
                
                if source_hash != target_hash:
                    patches_needed = True
                    break
        
        if not patches_needed:
            print("✅ YuEGP transformers patches already applied, skipping re-installation")
            print("  📈 High-performance optimizations are active:")
            print("    • Stage 1 generation: 3x faster (16GB+ VRAM)")
            print("    • Stage 2 generation: 2x faster (all profiles)")
            return True
        
        # Apply patches by copying optimized files
        print("🔧 Applying YuEGP transformers patches for high-performance generation...")
        
        # Copy the patched files, preserving directory structure
        for root, dirs, files in os.walk(source_dir):
            # Calculate relative path from source_dir
            rel_path = os.path.relpath(root, source_dir)
            target_subdir = os.path.join(target_dir, rel_path) if rel_path != '.' else target_dir
            
            # Ensure target subdirectory exists
            os.makedirs(target_subdir, exist_ok=True)
            
            # Copy all Python files in this directory
            for file in files:
                if file.endswith('.py'):
                    src_file = os.path.join(root, file)
                    dst_file = os.path.join(target_subdir, file)
                    
                    shutil.copy2(src_file, dst_file)
                    print(f"  ✅ Patched: {os.path.relpath(dst_file, target_base)}")
        
        print("🚀 Transformers patches applied successfully!")
        print("  📈 Expected performance gains:")
        print("    • Stage 1 generation: 3x faster (16GB+ VRAM)")
        print("    • Stage 2 generation: 2x faster (all profiles)")
        return True
        
    except Exception as e:
        print(f"❌ Error applying transformers patches: {e}")
        print("   Continuing without patches - performance may be reduced")
        return False

# Now import the rest of the dependencies
# Add project root to Python path for imports
project_root = os.path.dirname(os.path.abspath(__file__))
if project_root not in sys.path:
    sys.path.insert(0, project_root)

from tools.groq_client import client as groq_client
from openai import OpenAI
from tools.generate_lyrics import generate_structured_lyrics, format_lyrics

# Apply patches after all imports are set up
patch_applied = apply_transformers_patch()

# Import CUDA info after flash-attn setup
import torch
if torch.cuda.is_available():
    print(f"🎮 GPU: {torch.cuda.get_device_name(0)}")
    print(f"💾 VRAM: {torch.cuda.get_device_properties(0).total_memory / 1e9:.1f}GB")
else:
    print("⚠️  No CUDA GPU detected")

@dataclass
class AppState:
    """
    Maintains the application state throughout user interactions.
    """
    conversation: list = field(default_factory=list)
    stopped: bool = False
    model_outs: Any = None
    lyrics: str = ""
    genre: str = "pop"
    mood: str = "upbeat"
    theme: str = "love"

def validate_api_keys():
    """Validate required API keys for Spaces deployment"""
    required_keys = ["GROQ_API_KEY",  "GEMINI_API_KEY"]
    missing_keys = []
    
    for key in required_keys:
        if not os.getenv(key):
            missing_keys.append(key)
    
    if missing_keys:
        print(f"⚠️  Missing API keys: {missing_keys}")
        return False
    
    print("✅ All API keys validated")
    return True

def validate_file_structure():
    """Validate that required files and directories exist"""
    required_paths = [
        "YuEGP/inference/infer.py",
        "YuEGP/inference/codecmanipulator.py", 
        "YuEGP/inference/mmtokenizer.py",
        "tools/generate_lyrics.py",
        "tools/groq_client.py",
        "schemas/lyrics.py"  # Required for lyrics structure models
    ]
    
    missing_files = []
    for path in required_paths:
        if not os.path.exists(path):
            missing_files.append(path)
    
    if missing_files:
        print(f"⚠️  Missing required files: {missing_files}")
        return False
    
    print("✅ All required files found")
    return True

@spaces.GPU  # Decorator for GPU access in Spaces
def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progress()) -> str:
    """
    Generate music using YuE model with high-performance Spaces configuration
    """
    if not lyrics.strip():
        return "Please provide lyrics to generate music."
    
    try:
        progress(0.1, desc="Preparing lyrics...")
        
        # Use lyrics directly (already formatted from chat interface)
        formatted_lyrics = lyrics
        
        # Create temporary files
        with tempfile.NamedTemporaryFile(mode='w', suffix='.txt', delete=False) as genre_file:
            genre_file.write(f"instrumental,{genre},{mood},male vocals")
            genre_file_path = genre_file.name
        
        with tempfile.NamedTemporaryFile(mode='w', suffix='.txt', delete=False) as lyrics_file:
            lyrics_file.write(formatted_lyrics)
            lyrics_file_path = lyrics_file.name
        
        progress(0.2, desc="Setting up generation...")
        
        # Generate music with high-performance Spaces configuration
        output_dir = tempfile.mkdtemp()
        
        # High-performance command based on Spaces GPU resources
        # In Spaces, working directory is /app
        infer_script_path = os.path.join(os.getcwd(), "YuEGP", "inference", "infer.py")
        cmd = [
            sys.executable,
            infer_script_path,
            "--cuda_idx", "0",
            "--stage1_model", "m-a-p/YuE-s1-7B-anneal-en-cot",
            "--stage2_model", "m-a-p/YuE-s2-1B-general",
            "--genre_txt", genre_file_path,
            "--lyrics_txt", lyrics_file_path,
            "--run_n_segments", "2",        # Full segments for better quality
            "--stage2_batch_size", "4",     # Higher batch size for speed
            "--output_dir", output_dir,
            "--max_new_tokens", "3000",     # Full token count
            "--profile", "1",               # Highest performance profile
            "--verbose", "3",
            "--prompt_start_time", "0",
            "--prompt_end_time", "30",      # Full 30-second clips
        ]
        
        # Use flash attention if available, otherwise fallback
        if not flash_attn_available:
            cmd.append("--sdpa")
        
        progress(0.3, desc="Starting music generation (Stage 1)...")
        
        # Run with generous timeout for high-quality generation
        print("🎵 Starting high-quality music generation...")
        print(f"Working directory: {os.getcwd()}")
        print(f"Infer script path: {infer_script_path}")
        print(f"Command: {' '.join(cmd)}")
        
        # Change to YuEGP/inference directory for execution
        original_cwd = os.getcwd()
        inference_dir = os.path.join(os.getcwd(), "YuEGP", "inference")
        
        try:
            os.chdir(inference_dir)
            print(f"Changed to inference directory: {inference_dir}")
            
            # Update command to use relative path since we're in the inference directory
            cmd[1] = "infer.py"
            
            # Run with real-time output for debugging
            print(f"🚀 Executing command: {' '.join(cmd)}")
            result = subprocess.run(cmd, capture_output=True, text=True, timeout=1200)  # 20 minutes
            
            # Print stdout and stderr for debugging
            if result.stdout:
                print(f"✅ Command output:\n{result.stdout}")
            if result.stderr:
                print(f"⚠️ Command stderr:\n{result.stderr}")
            print(f"📊 Return code: {result.returncode}")
        finally:
            # Always restore original working directory
            os.chdir(original_cwd)
        
        progress(0.9, desc="Finalizing audio output...")
        
        # Clean up input files
        os.unlink(genre_file_path)
        os.unlink(lyrics_file_path)
        
        if result.returncode == 0:
            # Find generated audio file
            import glob
            audio_files = glob.glob(os.path.join(output_dir, "**/*.mp3"), recursive=True)
            
            if audio_files:
                progress(1.0, desc="Music generation complete!")
                return audio_files[0]  # Return path to generated audio
            else:
                return "Music generation completed but no audio file found."
        else:
            error_msg = f"Return code: {result.returncode}\n"
            if result.stderr:
                error_msg += f"Error: {result.stderr[-1000:]}\n"
            if result.stdout:
                error_msg += f"Output: {result.stdout[-1000:]}"
            return f"Music generation failed:\n{error_msg}"
            
    except subprocess.TimeoutExpired:
        return "Music generation timed out after 20 minutes. Please try again."
    except Exception as e:
        return f"Error during music generation: {str(e)}"

def respond(message, state):
    """Enhanced response function for lyrics generation"""
    try:
        # Add user message to conversation
        state.conversation.append({"role": "user", "content": message})
        
        # Generate response using your existing lyrics generation logic
        song_structure = generate_structured_lyrics(
            state.conversation,
            state.genre, 
            state.mood, 
            state.theme
        )
        
        # Format the structured lyrics for display
        response = format_lyrics(song_structure)
        
        # Add assistant response
        state.conversation.append({"role": "assistant", "content": response})
        
        # Update lyrics if this looks like final lyrics
        if any(marker in response.lower() for marker in ["[verse", "[chorus", "[bridge"]):
            state.lyrics = response
        
        # Return conversation for display
        conversation_display = []
        for msg in state.conversation:
            role = "User" if msg["role"] == "user" else "Assistant"
            conversation_display.append([msg["content"] if msg["role"] == "user" else None, 
                                       msg["content"] if msg["role"] == "assistant" else None])
        
        return "", conversation_display, state
        
    except Exception as e:
        error_response = f"Sorry, I encountered an error: {str(e)}"
        state.conversation.append({"role": "assistant", "content": error_response})
        
        # Format conversation for display
        conversation_display = []
        for msg in state.conversation:
            conversation_display.append([msg["content"] if msg["role"] == "user" else None, 
                                       msg["content"] if msg["role"] == "assistant" else None])
        
        return "", conversation_display, state

def build_interface():
    """Build the Gradio interface optimized for Spaces with high performance"""
    
    with gr.Blocks(
        title="MiloMusic - AI Music Generation", 
        theme=gr.themes.Soft(),
        css="""
        .container { max-width: 1400px; margin: auto; }
        .performance-notice { background-color: #d4edda; padding: 15px; border-radius: 5px; margin: 10px 0; }
        .generation-status { background-color: #f8f9fa; padding: 10px; border-radius: 5px; }
        """
    ) as demo:
        
        # Header
        gr.Markdown("""
        # 🎵 MiloMusic - AI Music Generation
        ### Professional AI-powered music creation from natural language
        """)
        
        # Performance notice for Spaces
        gr.Markdown("""
        <div class="performance-notice">
        🚀 <strong>High-Performance Mode:</strong> Running on Spaces GPU with optimized settings for best quality.
        Generation time: ~3-5 minutes for professional-grade music with vocals and instruments.
        </div>
        """)
        
        state = gr.State(AppState())
        
        with gr.Row():
            with gr.Column(scale=2):
                # Input controls
                with gr.Group():
                    gr.Markdown("### 🎛️ Music Settings")
                    with gr.Row():
                        genre = gr.Dropdown(
                            choices=["pop", "rock", "jazz", "classical", "electronic", "folk", "r&b", "country", "hip-hop"],
                            value="pop", label="Genre"
                        )
                        mood = gr.Dropdown(
                            choices=["upbeat", "melancholic", "energetic", "calm", "romantic", "dark", "mysterious", "joyful"],
                            value="upbeat", label="Mood"
                        )
                        theme = gr.Dropdown(
                            choices=["love", "friendship", "adventure", "nostalgia", "freedom", "hope", "dreams", "nature"],
                            value="love", label="Theme"
                        )
                
                # Voice Input
                with gr.Group():
                    gr.Markdown("### 🎤 Voice Input")
                    input_audio = gr.Audio(
                        label="Speak Your Musical Ideas",
                        sources=["microphone"],
                        type="numpy",
                        streaming=False,
                        waveform_options=gr.WaveformOptions(waveform_color="#B83A4B"),
                    )
                
                # Chat interface
                with gr.Group():
                    gr.Markdown("### 💬 Lyrics Creation Chat")
                    chatbot = gr.Chatbot(height=400, label="AI Lyrics Assistant", show_copy_button=True)
                    
                    with gr.Row():
                        text_input = gr.Textbox(
                            placeholder="Or type your song idea here...", 
                            show_label=False, 
                            scale=4,
                            lines=2
                        )
                        send_btn = gr.Button("Send", scale=1, variant="primary")
            
            with gr.Column(scale=1):
                # Output controls
                with gr.Group():
                    gr.Markdown("### 🎵 Music Generation")
                    lyrics_display = gr.Textbox(
                        label="Current Lyrics", 
                        lines=12, 
                        interactive=True,
                        placeholder="Your generated lyrics will appear here..."
                    )
                    
                    generate_btn = gr.Button("🎼 Generate High-Quality Music", variant="primary", size="lg")
                    
                    with gr.Column():
                        music_output = gr.Audio(label="Generated Music", type="filepath", show_download_button=True)
                        
                        gr.Markdown("""
                        <div class="generation-status">
                        <strong>Generation Features:</strong><br>
                        • Full 30-second clips<br>
                        • Professional vocals<br>
                        • Rich instrumentation<br>
                        • High-fidelity audio
                        </div>
                        """)
                
                # Controls
                with gr.Group():
                    gr.Markdown("### 🔧 Controls")
                    new_song_btn = gr.Button("🆕 Start New Song")
                    clear_btn = gr.Button("🧹 Clear Chat")
        
        # Event handlers
        def update_state_settings(genre_val, mood_val, theme_val, state):
            state.genre = genre_val
            state.mood = mood_val  
            state.theme = theme_val
            return state
        
        # Update state when settings change
        for component in [genre, mood, theme]:
            component.change(
                fn=update_state_settings,
                inputs=[genre, mood, theme, state],
                outputs=[state]
            )
        
        # Voice recording functionality (from app.py)
        stream = input_audio.start_recording(
            process_audio,
            [input_audio, state],
            [input_audio, state],
        )

        respond_audio = input_audio.stop_recording(
            response_audio, [state, input_audio, genre, mood, theme], [state, chatbot]
        )

        restart = respond_audio.then(start_recording_user, [state], [input_audio]).then(
            lambda state: state, state, state, js=js_reset
        )
        
        # Text chat functionality
        send_btn.click(
            fn=respond,
            inputs=[text_input, state],
            outputs=[text_input, chatbot, state],
            queue=True
        )
        
        text_input.submit(
            fn=respond,
            inputs=[text_input, state],
            outputs=[text_input, chatbot, state],
            queue=True
        )
        
        # Music generation with progress
        generate_btn.click(
            fn=generate_music_spaces,
            inputs=[lyrics_display, genre, mood],
            outputs=[music_output],
            queue=True,
            show_progress=True
        )
        
        # Control buttons
        new_song_btn.click(
            fn=lambda: (AppState(), [], "", None, gr.Audio(recording=False)),
            outputs=[state, chatbot, lyrics_display, music_output, input_audio],
            cancels=[respond_audio, restart]
        )
        
        clear_btn.click(
            fn=lambda: [],
            outputs=[chatbot]
        )
        
        # Auto-update lyrics display when state changes
        state.change(
            fn=lambda s: s.lyrics,
            inputs=[state],
            outputs=[lyrics_display]
        )
        
        # Instructions
        gr.Markdown("""
        ### 📖 How to create your music:
        1. **Set your preferences**: Choose genre, mood, and theme
        2. **Voice or chat**: Either speak your ideas or type them in the chat
        3. **Refine the lyrics**: Ask for changes, different verses, or style adjustments  
        4. **Generate music**: Click the generate button for professional-quality output
        5. **Download & enjoy**: Your high-fidelity music with vocals and instruments
        
        **Tips**: Be specific about your vision - mention instruments, vocal style, or song structure!
        """)
        
        # Footer
        gr.Markdown("""
        ---
        <center>
        Made with ❤️ by the MiloMusic Team | Powered by YuE (乐) Model | 🤗 Hugging Face Spaces
        </center>
        """)
    
    return demo

# Audio transcription functions (from app.py)
def process_whisper_response(completion):
    """
    Process Whisper transcription response and filter out silence.
    """
    if completion.segments and len(completion.segments) > 0:
        no_speech_prob = completion.segments[0].get('no_speech_prob', 0)
        print("No speech prob:", no_speech_prob)

        if no_speech_prob > 0.7:
            print("No speech detected")
            return None

        return completion.text.strip()

    return None

def transcribe_audio(client, file_name):
    """
    Transcribe an audio file using the Whisper model via the Groq API.
    """
    if file_name is None:
        return None

    try:
        with open(file_name, "rb") as audio_file:
            with open("audio.wav", "wb") as f:
                f.write(audio_file.read())

            response = client.audio.transcriptions.create(
                model="whisper-large-v3-turbo",
                file=("audio.wav", audio_file),
                response_format="text",
                language="en",
            )
            
            # Process the response to filter out silence
            # For text response format, we need to check if response is meaningful
            if response and len(response.strip()) > 0:
                return response.strip()
            else:
                return None
                
    except Exception as e:
        print(f"Transcription error: {e}")
        return f"Error in audio transcription: {str(e)}"

def start_recording_user(state: AppState):
    """
    Reset the audio recording component for a new user input.
    """
    return None

def process_audio(audio: tuple, state: AppState):
    """
    Process recorded audio in real-time during recording.
    """
    return audio, state

@spaces.GPU(duration=40, progress=gr.Progress(track_tqdm=True))
def response_audio(state: AppState, audio: tuple, genre_value, mood_value, theme_value):
    """
    Process recorded audio and generate a response based on transcription.
    """
    if not audio:
        return state, []

    # Update state with current dropdown values
    state.genre, state.mood, state.theme = genre_value, mood_value, theme_value

    temp_dir = tempfile.gettempdir()
    file_name = os.path.join(temp_dir, f"{xxhash.xxh32(bytes(audio[1])).hexdigest()}.wav")

    sf.write(file_name, audio[1], audio[0], format="wav")

    api_key = os.environ.get("GROQ_API_KEY")
    if not api_key:
        raise ValueError("Please set the GROQ_API_KEY environment variable.")
    client = groq.Client(api_key=api_key)

    # Transcribe the audio file
    transcription = transcribe_audio(client, file_name)
    if transcription:
        if isinstance(transcription, str) and transcription.startswith("Error"):
            transcription = "Error in audio transcription."

        state.conversation.append({"role": "user", "content": transcription})

        assistant_message = generate_chat_completion(client, state.conversation, state.genre, state.mood, state.theme)

        state.conversation.append({"role": "assistant", "content": assistant_message})

        # Update lyrics from conversation
        state.lyrics = extract_lyrics_from_conversation(state.conversation)

        os.remove(file_name)

    # Format conversation for display
    conversation_display = []
    for msg in state.conversation:
        conversation_display.append([msg["content"] if msg["role"] == "user" else None, 
                                   msg["content"] if msg["role"] == "assistant" else None])
    
    return state, conversation_display

def extract_lyrics_from_conversation(conversation):
    """
    Extract lyrics from conversation history.
    """
    lyrics = ""
    for message in reversed(conversation):
        if message["role"] == "assistant" and "verse" in message["content"].lower() and "chorus" in message["content"].lower():
            lyrics = message["content"]
            break
    return lyrics

def generate_chat_completion(client, history, genre, mood, theme):
    """
    Generate an AI assistant response based on conversation history and song parameters.
    """
    messages = []
    system_prompt = f"""You are a creative AI music generator assistant. Help users create song lyrics in the {genre} genre with a {mood} mood about {theme}.
When generating lyrics, create a chorus and at least one verse. Format lyrics clearly with VERSE and CHORUS labels.
Ask if they like the lyrics or want changes. Be conversational, friendly, and creative.
Keep the lyrics appropriate for the selected genre, mood, and theme unless the user specifically requests changes."""

    messages.append({
        "role": "system",
        "content": system_prompt,
    })

    for message in history:
        messages.append(message)

    try:
        completion = client.chat.completions.create(
            model="meta-llama/llama-4-scout-17b-16e-instruct",
            messages=messages,
        )
        return completion.choices[0].message.content
    except Exception as e:
        return f"Error in generating chat completion: {str(e)}"

# JavaScript for frontend enhancements
js_reset = """
() => {
  var record = document.querySelector('.record-button');
  if (record) {
    record.textContent = "Just Start Talking!"
    record.style = "width: fit-content; padding-right: 0.5vw;"
  }
}
"""

# Build the interface
demo = build_interface()

if __name__ == "__main__":
    """
    Spaces entry point - optimized for high-performance deployment
    """
    print("🚀 Starting MiloMusic High-Performance Mode on Hugging Face Spaces...")
    print(f"📁 Working directory: {os.getcwd()}")
    print(f"📂 Directory contents: {os.listdir('.')}")
    
    # Validate file structure
    if not validate_file_structure():
        print("❌ Required files missing - please check your upload")
        sys.exit(1)
    
    # Validate environment
    if not validate_api_keys():
        print("⚠️  Some API keys missing - functionality may be limited")
    
    # Launch with optimized settings for Spaces
    demo.queue(
        default_concurrency_limit=5,  # Allow more concurrent users
        max_size=20
    ).launch(
        server_name="0.0.0.0",
        server_port=7860,
        share=False,  # Spaces handles sharing
        show_error=True,
        quiet=False,
        favicon_path=None,
        ssl_verify=False
    )