cudaにすれば5sぐらいが目標で

app.py

@@ -1,5 +1,8 @@
 import gradio as gr  # type: ignore
+
 import spaces  # type: ignore
+import torch
+
 
 # Load model directly
 from transformers import AutoTokenizer, AutoModelForCausalLM
@@ -11,6 +14,8 @@ model = AutoModelForCausalLM.from_pretrained(
     "microsoft/Phi-3-mini-4k-instruct", trust_remote_code=True
 )
 
+model.to("cuda")
+
 
 def greet(name, sliderint):
     return "Hellonyaaaaa " + name + "!!" + str(sliderint)