Update app.py

app.py

@@ -1,8 +1,5 @@
 import spaces # If using Hugging Face Spaces
 import os
-import torch
-import gradio as gr
-
 # ## GGUF MOD: Unused environment variables for PyTorch have been removed.
 # ## GGUF MOD: ctransformers handles its own memory and GPU management.
 os.putenv('PYTORCH_NVML_BASED_CUDA_CHECK','1')
@@ -14,11 +11,16 @@ os.putenv('PYTORCH_NVML_BASED_CUDA_CHECK','1')
 # os.environ['PYTORCH_CUDA_ALLOC_CONF'] = ','.join(alloc_conf_parts)
 # os.environ["SAFETENSORS_FAST_GPU"] = "1"
 os.putenv('HF_HUB_ENABLE_HF_TRANSFER','1')
+import torch
+import gradio as gr
+
+
 
 # ## GGUF MOD: Import AutoModelForCausalLM from ctransformers instead of transformers.
 # ## GGUF MOD: BitsAndBytesConfig is no longer needed.
 from ctransformers import AutoModelForCausalLM
 from transformers import AutoTokenizer
+from image_gen_aux import UpscaleWithModel
 
 # ## GGUF MOD: PyTorch backend settings are not used by ctransformers.
 torch.backends.cuda.matmul.allow_tf32 = True
@@ -46,6 +48,9 @@ print("Loading GGUF model...")
 #   leading to much faster inference. Adjust this number based on your VRAM.
 # - hf=True: This tells ctransformers to download from the Hugging Face Hub.
 
+upscaler = UpscaleWithModel.from_pretrained("Kim2091/ClearRealityV1").to(torch.device('cuda'))
+upscaler.to(torch.device('cpu'))
+
 def loadModel():
     device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
     model = AutoModelForCausalLM.from_pretrained(
@@ -58,10 +63,7 @@ def loadModel():
     )
     return model
     
-@spaces.GPU()
-def device_wake():
-    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
-device_wake()
+device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
 model = loadModel()
 print("GGUF Model loaded successfully.")