Update app.py

app.py

@@ -7,45 +7,42 @@ from huggingface_hub import snapshot_download
 from dotenv import load_dotenv
 load_dotenv()
 
-# Load models function
-def load_models():
-    # Check if CUDA is available
-    device = "cuda" if torch.cuda.is_available() else "cpu"
 
-    print("Loading SNAC model...")
-    snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
-    snac_model = snac_model.to(device)
-    
-    model_name = "canopylabs/orpheus-3b-0.1-ft"
-    
-    # Download only model config and safetensors
-    snapshot_download(
-        repo_id=model_name,
-        allow_patterns=[
-            "config.json",
-            "*.safetensors",
-            "model.safetensors.index.json",
-        ],
-        ignore_patterns=[
-            "optimizer.pt",
-            "pytorch_model.bin",
-            "training_args.bin",
-            "scheduler.pt",
-            "tokenizer.json",
-            "tokenizer_config.json",
-            "special_tokens_map.json",
-            "vocab.json",
-            "merges.txt",
-            "tokenizer.*"
-        ]
-    )
-    
-    model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
-    model.to(device)
-    tokenizer = AutoTokenizer.from_pretrained(model_name)
-    print(f"Orpheus model loaded to {device}")
-    
-    return snac_model, model, tokenizer, device
+# Check if CUDA is available
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+print("Loading SNAC model...")
+snac_model = SNAC.from_pretrained("hubertsiuzdak/snac_24khz")
+snac_model = snac_model.to(device)
+
+model_name = "canopylabs/orpheus-3b-0.1-ft"
+
+# Download only model config and safetensors
+snapshot_download(
+    repo_id=model_name,
+    allow_patterns=[
+        "config.json",
+        "*.safetensors",
+        "model.safetensors.index.json",
+    ],
+    ignore_patterns=[
+        "optimizer.pt",
+        "pytorch_model.bin",
+        "training_args.bin",
+        "scheduler.pt",
+        "tokenizer.json",
+        "tokenizer_config.json",
+        "special_tokens_map.json",
+        "vocab.json",
+        "merges.txt",
+        "tokenizer.*"
+    ]
+)
+
+model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
+model.to(device)
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+print(f"Orpheus model loaded to {device}")
 
 # Process text prompt
 def process_prompt(prompt, voice, tokenizer, device):
@@ -172,13 +169,6 @@ examples = [
 # Available voices
 VOICES = ["tara", "dan", "josh", "emma"]
 
-# Load models globally
-try:
-    snac_model, model, tokenizer, device = load_models()
-except Exception as e:
-    print(f"Error loading models: {e}")
-    raise
-
 # Create Gradio interface
 with gr.Blocks(title="Orpheus Text-to-Speech") as demo:
     gr.Markdown("""