Update app.py

app.py

@@ -1,15 +1,14 @@
+# Disable CUDA visibility at the start
+os.environ["CUDA_VISIBLE_DEVICES"] = ""  # Prevents CUDA initialization
+
 import spaces
 import gradio as gr
 from huggingface_hub import InferenceClient, login
 import os
 import time
 
-# Disable CUDA visibility at the start
-os.environ["CUDA_VISIBLE_DEVICES"] = ""  # Prevents CUDA initialization
 
-@spaces.GPU  # Forces GPU allocation before execution
-def force_gpu_allocation():
-    pass  # Dummy function to trigger GPU setup
+
 
 # Base model (LLaMA 3.1 8B) from Meta
 base_model_name = "meta-llama/Llama-3.1-8B"
@@ -34,6 +33,10 @@ def chatbot_response(user_input):
     if torch.cuda.is_initialized():
         print("CUDA was already initialized before Accelerator!")
 
+    @spaces.GPU  # Forces GPU allocation before execution
+    def force_gpu_allocation():
+        pass  # Dummy function to trigger GPU setup
+
     accelerator = Accelerator()
 
     # Login because LLaMA 3.1 8B is a gated model