app.py

import os
import subprocess
import sys
from pathlib import Path

# --- 1. Clone the VibeVoice Repository ---
repo_dir = "VibeVoice"
if not os.path.exists(repo_dir):
    print("Cloning the VibeVoice repository...")
    try:
        subprocess.run(
            ["git", "clone", "https://github.com/vibevoice-community/VibeVoice.git"],
            check=True, capture_output=True, text=True
        )
        print("Repository cloned successfully.")
    except subprocess.CalledProcessError as e:
        print(f"Error cloning repository: {e.stderr}")
        sys.exit(1)
else:
    print("Repository already exists. Skipping clone.")

# --- 2. Install Dependencies ---
os.chdir(repo_dir)
print(f"Changed directory to: {os.getcwd()}")

print("Installing the VibeVoice package in editable mode...")
try:
    subprocess.run(
        [sys.executable, "-m", "pip", "install", "-e", "."],
        check=True, capture_output=True, text=True
    )
    print("Package installed successfully.")
except subprocess.CalledProcessError as e:
    print(f"Error installing package: {e.stderr}")
    sys.exit(1)

# --- 3. Refactor the demo script using a direct replacement strategy ---
demo_script_path = Path("demo/gradio_demo.py")
print(f"Refactoring {demo_script_path} for ZeroGPU lazy loading...")

try:
    with open(demo_script_path, 'r') as f:
        modified_content = f.read()

    # --- Add 'import spaces' at the top ---
    if "import spaces" not in modified_content:
        modified_content = "import spaces\n" + modified_content

    # --- Patch 1: Defer model loading in __init__ ---
    original_init_call = "        self.load_model()"
    replacement_init_block = (
        "        # self.load_model() # Patched: Defer model loading\n"
        "        self.model = None\n"
        "        self.processor = None"
    )
    if original_init_call in modified_content:
        modified_content = modified_content.replace(original_init_call, replacement_init_block, 1)
        print("Successfully patched __init__ to prevent startup model load.")
    else:
        print(f"\033[91mError: Could not find '{original_init_call}' to patch. Startup patch failed.\033[0m")
        sys.exit(1)
        
    # --- Patch 2: Add decorator and lazy-loading logic to the generation method ---
    # Define the exact block to find, spanning the full method signature down to the 'try:'.
    # This is sensitive to whitespace but is the most direct way to replace.
    original_method_header = """    def generate_podcast_streaming(self, 
                                 num_speakers: int,
                                 script: str,
                                 speaker_1: str = None,
                                 speaker_2: str = None,
                                 speaker_3: str = None,
                                 speaker_4: str = None,
                                 cfg_scale: float = 1.3) -> Iterator[tuple]:
        try:"""

    # Define the full replacement block with correct indentation.
    replacement_method_header = """    @spaces.GPU(duration=120)
    def generate_podcast_streaming(self,
                                 num_speakers: int,
                                 script: str,
                                 speaker_1: str = None,
                                 speaker_2: str = None,
                                 speaker_3: str = None,
                                 speaker_4: str = None,
                                 cfg_scale: float = 1.3) -> Iterator[tuple]:
        # Patched: Lazy-load model and processor on the GPU worker
        if self.model is None or self.processor is None:
            print("Loading processor & model for the first time on GPU worker...")
            self.processor = VibeVoiceProcessor.from_pretrained(self.model_path)
            self.model = VibeVoiceForConditionalGenerationInference.from_pretrained(
                self.model_path,
                torch_dtype=torch.bfloat16, # Use 16-bit precision for quality
                device_map="auto",
            )
            self.model.eval()
            self.model.model.noise_scheduler = self.model.model.noise_scheduler.from_config(
                self.model.model.noise_scheduler.config,
                algorithm_type='sde-dpmsolver++',
                beta_schedule='squaredcos_cap_v2'
            )
            self.model.set_ddpm_inference_steps(num_steps=self.inference_steps)
            print("Model and processor loaded successfully on GPU worker.")

        try:"""

    if original_method_header in modified_content:
        modified_content = modified_content.replace(original_method_header, replacement_method_header, 1)
        print("Successfully patched generation method for lazy loading.")
    else:
        print(f"\033[91mError: Could not find the method definition for 'generate_podcast_streaming' to patch. This is likely due to a whitespace mismatch. Please check the demo script.\033[0m")
        sys.exit(1)

    # --- Write the modified content back to the file ---
    with open(demo_script_path, 'w') as f:
        f.write(modified_content)
    
    print("Script patching complete.")

except Exception as e:
    print(f"An error occurred while modifying the script: {e}")
    import traceback
    traceback.print_exc()
    sys.exit(1)

# --- 4. Launch the Gradio Demo ---
model_id = "microsoft/VibeVoice-1.5B"
command = ["python", str(demo_script_path), "--model_path", model_id, "--share"]
print(f"Launching Gradio demo with command: {' '.join(command)}")
subprocess.run(command)