Update app.py

app.py

@@ -40,25 +40,20 @@ MODEL_CONFIG = {
 # -------------------------------
 # Model Manager with Cache
 # -------------------------------
+
 class ModelManager:
-    def __init__(self):
-        self.llama_pipelines = {}
-        self.musicgen_model = None
-        self.tts_models = {}
-        
     def get_llama_pipeline(self, model_id, token):
         if model_id not in self.llama_pipelines:
             tokenizer = AutoTokenizer.from_pretrained(
                 model_id,
-                use_auth_token=token,
-                legacy=False  # Important for compatibility
+                token=token,
+                legacy=False  # Critical for tokenizers 0.19.x compatibility
             )
             model = AutoModelForCausalLM.from_pretrained(
                 model_id,
-                use_auth_token=token,
+                token=token,
                 torch_dtype=torch.float16,
-                device_map="auto",
-                low_cpu_mem_usage=True  # Reduces memory pressure
+                device_map="auto"
             )
             self.llama_pipelines[model_id] = pipeline(
                 "text-generation",
@@ -67,7 +62,7 @@ class ModelManager:
                 device_map="auto"
             )
         return self.llama_pipelines[model_id]
-
+        
     def get_musicgen_model(self):
         if not self.musicgen_model:
             self.musicgen_model = MusicgenForConditionalGeneration.from_pretrained(