Update joycaption.py

joycaption.py

@@ -1,6 +1,6 @@
 import os
-if os.environ.get("SPACES_ZERO_GPU") is not None:
-    import spaces
+#if os.environ.get("SPACES_ZERO_GPU") is not None:
+#    import spaces
 else:
     class spaces:
         @staticmethod
@@ -23,10 +23,10 @@ from typing import Union
 
 LOAD_IN_NF4 = True
 
-if os.environ.get("SPACES_ZERO_GPU") is not None:
-    import subprocess
-    LOAD_IN_NF4 = False # If true, Custom VLM LoRA doesn't work initially. The rest are fine.
-    subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
+#if os.environ.get("SPACES_ZERO_GPU") is not None:
+#    import subprocess
+#    LOAD_IN_NF4 = False # If true, Custom VLM LoRA doesn't work initially. The rest are fine.
+#    subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)
 
 BASE_DIR = Path(__file__).resolve().parent # Define the base directory
 device = "cuda" if torch.cuda.is_available() else "cpu"
@@ -474,7 +474,7 @@ def get_repo_gguf(repo_id: str):
     else: return gr.update(value=files[0], choices=files)
 
 
-@spaces.GPU
+#@spaces.GPU
 def change_text_model(model_name: str=MODEL_PATH, use_client: bool=False, gguf_file: Union[str, None]=None,
                       is_nf4: bool=True, is_lora: bool=True, progress=gr.Progress(track_tqdm=True)):
     global use_inference_client, llm_models