Update app.py

app.py

@@ -49,7 +49,8 @@ def load_llava_model():
         llava_model = AutoModelForCausalLM.from_pretrained(
             model_id,
             torch_dtype=torch.float16,
-            device_map="auto"
+            device_map="auto",
+            cache_dir="./models"  # Add this line
         )
     
     return llava_model, llava_processor