adds _no_split_block

adds _no_split_block to EncoderDecoder model for auto device mapping

modeling_codet5p.py

@@ -755,7 +755,8 @@ def shift_tokens_right(input_ids: torch.Tensor, pad_token_id: int, decoder_start
 # Adapted from transformers.models.encoder_decoder.modeling_encoder_decoder.EncoderDecoderModel
 class CodeT5pEncoderDecoderModel(PreTrainedModel):
     config_class = CodeT5pConfig
-
+    _no_split_modules = ["CodeT5pBlock"]
+    
     def __init__(
             self,
             config: Optional[PretrainedConfig] = None,