Update app.py

app.py

@@ -15,6 +15,10 @@ llm = Llama(
     chat_format="llama-3",
     n_gpu_layers=-1,  # ensure all layers are on GPU
     n_threads=1, # no CPU multi-threading
+    offload_kqv=True, # store kqv on GPU
+    vocab_only=False,
+    use_mmap=True,
+    use_mlock=False,
 )
 
 # Placeholder responses for when context is empty