diff --git a/modeling_internlm3.py b/modeling_internlm3.py
index 6e877d0..eec4e6b 100644
--- a/modeling_internlm3.py
+++ b/modeling_internlm3.py
@@ -793,7 +793,7 @@ class InternLM3Model(InternLM3PreTrainedModel):
     Args:
         config: InternLM3Config
     """
-
+    _auto_class = "AutoModel"
     def __init__(self, config: InternLM3Config):
         super().__init__(config)
         self.padding_idx = config.pad_token_id
@@ -1070,6 +1070,7 @@ class KwargsForCausalLM(FlashAttentionKwargs, LossKwargs): ...
 
 
 class InternLM3ForCausalLM(InternLM3PreTrainedModel, GenerationMixin):
+    _auto_class = "AutoModelForCausalLM"
     _tied_weights_keys = ["lm_head.weight"]
     _tp_plan = {"lm_head": "colwise_rep"}
 
diff --git a/tokenization_internlm3.py b/tokenization_internlm3.py
index f68147f..fb919ad 100644
--- a/tokenization_internlm3.py
+++ b/tokenization_internlm3.py
@@ -67,7 +67,7 @@ class InternLM3Tokenizer(PreTrainedTokenizer):
             Whether or not to add an initial space to the input. This allows to treat the leading word just as any
             other word. Again, this should be set with `from_slow=True` to make sure it's taken into account.
     """
-
+    _auto_class = "AutoTokenizer"
     vocab_files_names = VOCAB_FILES_NAMES
     model_input_names = ["input_ids", "attention_mask"]