Update app.py

app.py

@@ -49,10 +49,13 @@ def load_models():
         logger.info("🚀 Loading vLLM model on GPU...")
         model = LLM(
             model=LLASA_MODEL_ID,
-            gpu_memory_utilization=0.8,
-            max_model_len=2048,
-            enforce_eager=True,
-            device="cuda"
+            gpu_memory_utilization=0.90, 
+            max_model_len=2048, 
+            enable_prefix_caching=True, 
+            dtype='auto',
+            quantization=None,
+            enforce_eager=False,
+            kv_cache_dtype='auto'
         )
         
     if codec_model is None: