Update app.py

app.py

@@ -9,13 +9,18 @@ model_path = hf_hub_download(
     filename="AstroSage-8B-Q8_0.gguf"
 )
 
-llm = Llama(
-    model_path=model_path,
-    n_ctx=2048,
-    chat_format="llama-3",
-    n_gpu_layers=-1,  # ensure all layers are on GPU
-    split_mode=0,
-)
+@space.GPU
+def load_llm():
+    llm = Llama(
+        model_path=model_path,
+        n_ctx=2048,
+        chat_format="llama-3",
+        n_gpu_layers=-1,  # ensure all layers are on GPU
+        split_mode=0,
+    )
+    return llm
+
+llm = load_llm()
 
 # Placeholder responses for when context is empty
 GREETING_MESSAGES = [