1.delete apply patches for YuE 2.modify YuEGP to YuE

app.py

@@ -78,7 +78,7 @@ def install_flash_attn() -> bool:
 # Setup environment first
 setup_spaces_environment()
 
-# Download required models for YuEGP inference
+# Download required models for YuE inference
 def download_required_models():
     """Download required model files at startup"""
     try:
@@ -103,116 +103,6 @@ models_ready = download_required_models()
 # 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
@@ -224,9 +114,6 @@ 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():
@@ -267,9 +154,9 @@ def validate_api_keys():
 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",
+        "YuE/inference/infer.py",
+        "YuE/inference/codecmanipulator.py", 
+        "YuE/inference/mmtokenizer.py",
         "tools/generate_lyrics.py",
         "tools/groq_client.py",
         "schemas/lyrics.py"  # Required for lyrics structure models
@@ -306,8 +193,8 @@ def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progre
             genre_file.write(f"instrumental,{genre},{mood},male vocals")
             genre_file_path = genre_file.name
         
-        # Convert lyrics format for YuEGP compatibility
-        # YuEGP expects [VERSE], [CHORUS] format, but our AI generates **VERSE**, **CHORUS**
+        # Convert lyrics format for YuE compatibility
+        # YuE expects [VERSE], [CHORUS] format, but our AI generates **VERSE**, **CHORUS**
         import re
         
         # Extract only the actual lyrics content, removing AI commentary
@@ -352,7 +239,7 @@ def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progre
         
         # 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")
+        infer_script_path = os.path.join(os.getcwd(), "YuE", "inference", "infer.py")
         cmd = [
             sys.executable,
             infer_script_path,
@@ -365,9 +252,6 @@ def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progre
             "--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",
-            "--rescale",                    # Enable audio rescaling to proper volume
             "--prompt_start_time", "0",
             "--prompt_end_time", "30",      # Full 30-second clips
         ]
@@ -391,9 +275,9 @@ def generate_music_spaces(lyrics: str, genre: str, mood: str, progress=gr.Progre
         print(f"Working directory: {os.getcwd()}")
         print(f"Command: {' '.join(cmd)}")
         
-        # Change to YuEGP/inference directory for execution
+        # Change to YuE/inference directory for execution
         original_cwd = os.getcwd()
-        inference_dir = os.path.join(os.getcwd(), "YuEGP", "inference")
+        inference_dir = os.path.join(os.getcwd(), "YuE", "inference")
         
         try:
             os.chdir(inference_dir)

download_models.py

@@ -14,7 +14,7 @@ def download_xcodec_models():
     """Download xcodec_mini_infer using git clone (no LFS) + wget for large files"""
 
     # Base path for xcodec models - convert to absolute path to avoid working directory issues
-    xcodec_base = Path("YuEGP/inference/xcodec_mini_infer").resolve()
+    xcodec_base = Path("YuE/inference/xcodec_mini_infer").resolve()
 
     print("📥 Downloading xcodec_mini_infer using git clone + wget strategy...")
 
@@ -183,7 +183,7 @@ def ensure_model_availability():
     Download them if they don't exist.
     """
     
-    xcodec_base = Path("YuEGP/inference/xcodec_mini_infer")
+    xcodec_base = Path("YuE/inference/xcodec_mini_infer")
     
     # Check if critical files exist (both for recons and vocoder stages)
     critical_files = [