Update custom_st.py

custom_st.py

@@ -1,10 +1,13 @@
+import math
+import logging
 from io import BytesIO
 from typing import Any, Dict, Optional, List
 import torch
 from PIL import Image
 from sentence_transformers.models import Transformer as BaseTransformer
 from transformers import AutoModelForVision2Seq, AutoProcessor
-
+from packaging import version
+import transformers
 
 class MultiModalTransformer(BaseTransformer):
     def __init__(
@@ -51,7 +54,10 @@ class MultiModalTransformer(BaseTransformer):
         self, features: Dict[str, torch.Tensor], **kwargs
     ) -> Dict[str, torch.Tensor]:       
         if features.get("inputs_embeds", None) is None:
-            features["inputs_embeds"] = self.auto_model.base_model.embed_tokens(features["input_ids"])
+            if version.parse(transformers.__version__) >= version.parse("4.52.0"):
+                features["inputs_embeds"] = self.auto_model.base_model.language_model.embed_tokens(features["input_ids"])
+            else:
+                features["inputs_embeds"] = self.auto_model.base_model.embed_tokens(features["input_ids"])
             if features.get("pixel_values", None) is not None:
                 features["pixel_values"] = features["pixel_values"].type(self.auto_model.visual.get_dtype())
                 image_embeds = self.auto_model.visual(