updated

app.py

@@ -68,10 +68,13 @@ def load_model():
         print("Initial model loading...")
         _processor = AutoProcessor.from_pretrained("facebook/musicgen-large")
         _model = MusicgenForConditionalGeneration.from_pretrained(
-            "facebook/musicgen-large",
-            torch_dtype=torch.float16,
-            device="cuda" if torch.cuda.is_available() else "cpu"
+                    "facebook/musicgen-large",
+                    torch_dtype=torch.float16,
+                    device="cuda",
+                    mode="S",
+                    __paged=True,
         )
+        _model.eval()
     return _processor, _model
 
 def generate_music(text_prompt, duration=10, temperature=1.0, top_k=250, top_p=0.0):
@@ -80,11 +83,15 @@ def generate_music(text_prompt, duration=10, temperature=1.0, top_k=250, top_p=0
         processor, model = load_model()
         
         # Process the text prompt
+        print("Processor start")
         inputs = processor(
             text=[text_prompt],
             padding=True,
             return_tensors="pt",
         ).to("cuda")
+        print("Processor end")
+        print(inputs.device)
+
         # Generate audio
         with torch.no_grad():
             audio_values = model.generate(