Update app.py

app.py

@@ -4,11 +4,13 @@ import torch
 from huggingface_hub import hf_hub_download
 from llama_cpp import Llama, LlamaGrammar
 
+
 zero = torch.Tensor([0]).cuda()
 print(zero.device) # <-- 'cpu' 🤔
 
 @spaces.GPU
-def greet(llm):
+def greet(n):
+    global llm
     print(zero.device) # <-- 'cuda:0' 🤗
     grammar = LlamaGrammar.from_string('''
     root ::= sentence