update

configuration_japanese_clip.py

@@ -24,8 +24,6 @@ class JapaneseCLIPVisionConfig(PretrainedConfig):
         output_tokens: bool = False,
         **kwargs
     ):
-        super().__init__(**kwargs)
-
         self.image_size = image_size
         self.patch_size = patch_size
         self.width = width
@@ -43,6 +41,7 @@ class JapaneseCLIPVisionConfig(PretrainedConfig):
         self.pool_type = pool_type
         self.final_ln_after_pool = final_ln_after_pool
         self.output_tokens = output_tokens
+        super().__init__(**kwargs)
 
 
 class JapaneseCLIPConfig(PretrainedConfig):
@@ -53,12 +52,8 @@ class JapaneseCLIPConfig(PretrainedConfig):
         max_length: int = 77,
         **kwargs
     ):
-        super().__init__(**kwargs)
-
         self.max_length = max_length
-        import pdb
 
-        pdb.set_trace()
         if "vision_config" not in kwargs:
             raise ValueError("vision_config must be provided")
         if "text_config" not in kwargs:
@@ -69,6 +64,7 @@ class JapaneseCLIPConfig(PretrainedConfig):
 
         self.vision_config = JapaneseCLIPVisionConfig(**vision_config)
         self.text_config = RobertaConfig(**text_config)
+        super().__init__(**kwargs)
 
     @classmethod
     def from_vision_text_configs(