Update op to support both models

Signed-off-by: Jael Gu <mengjia.gu@zilliz.com>

auto_transformers.py

@@ -84,8 +84,10 @@ class AutoTransformers(NNOperator):
         self.model_name = model_name
 
         if self.model_name:
-            self.model = Model(model_name=self.model_name, device=self.device, checkpoint_path=checkpoint_path)
-            self.configs = self.model.model.config
+            self.accelerate_model = Model(
+                model_name=self.model_name, device=self.device, checkpoint_path=checkpoint_path)
+            self.model = self.accelerate_model.model
+            self.configs = self.model.config
             if tokenizer is None:
                 try:
                     self.tokenizer = AutoTokenizer.from_pretrained(model_name)
@@ -105,7 +107,7 @@ class AutoTransformers(NNOperator):
             log.error(f'Invalid input for the tokenizer: {self.model_name}')
             raise e
         try:
-            outs = self.model(**inputs)
+            outs = self.accelerate_model(**inputs)
         except Exception as e:
             log.error(f'Invalid input for the model: {self.model_name}')
             raise e
@@ -144,13 +146,13 @@ class AutoTransformers(NNOperator):
             from transformers.onnx.features import FeaturesManager
             from transformers.onnx import export
             model_kind, model_onnx_config = FeaturesManager.check_supported_model_or_raise(
-                self.model.model, feature='default')
+                self.model, feature='default')
             onnx_config = model_onnx_config(self.configs)
             if os.path.isdir(path):
                 shutil.rmtree(path)
             onnx_inputs, onnx_outputs = export(
                 self.tokenizer,
-                self.model.model,
+                self.model,
                 config=onnx_config,
                 opset=13,
                 output=Path(path+'.onnx')