Update opensora/serve/gradio_web_server.py

opensora/serve/gradio_web_server.py

@@ -63,16 +63,16 @@ if __name__ == '__main__':
     device = torch.device('cuda:0')
 
     # Load model:
-    transformer_model = LatteT2V.from_pretrained(args.model_path, subfolder=args.version, torch_dtype=torch.float16, cache_dir='cache_dir').to(device)
+    transformer_model = LatteT2V.from_pretrained(args.model_path, subfolder=args.version, torch_dtype=torch.float16).to(device)
 
-    vae = getae_wrapper(args.ae)(args.model_path, subfolder="vae", cache_dir='cache_dir').to(device, dtype=torch.float16)
+    vae = getae_wrapper(args.ae)(args.model_path, subfolder="vae").to(device, dtype=torch.float16)
     vae.vae.enable_tiling()
     image_size = int(args.version.split('x')[1])
     latent_size = (image_size // ae_stride_config[args.ae][1], image_size // ae_stride_config[args.ae][2])
     vae.latent_size = latent_size
     transformer_model.force_images = args.force_images
-    tokenizer = T5Tokenizer.from_pretrained(args.text_encoder_name, cache_dir="cache_dir")
-    text_encoder = T5EncoderModel.from_pretrained(args.text_encoder_name, cache_dir="cache_dir",
+    tokenizer = T5Tokenizer.from_pretrained(args.text_encoder_name)
+    text_encoder = T5EncoderModel.from_pretrained(args.text_encoder_name,
                                                   torch_dtype=torch.float16).to(device)
 
     # set eval mode