fix model loading in custom_st.py

custom_st.py

@@ -6,7 +6,6 @@ from typing import Any, Dict, List, Optional, Tuple, Union
 
 import requests
 import torch
-from PIL import Image
 from torch import nn
 from transformers import AutoConfig, AutoImageProcessor, AutoModel, AutoTokenizer
 
@@ -55,7 +54,7 @@ class Transformer(nn.Module):
             config_args = {}
 
         config = AutoConfig.from_pretrained(model_name_or_path, **config_args, cache_dir=cache_dir)
-        self._load_model(model_name_or_path, config, cache_dir, **model_args)
+        self.auto_model = AutoModel.from_pretrained(model_name_or_path, **config_args, cache_dir=cache_dir, **model_args)
 
         if max_seq_length is not None and "model_max_length" not in tokenizer_args:
             tokenizer_args["model_max_length"] = max_seq_length