Fix onnxrt split op (#1686)

Signed-off-by: yuwenzho <yuwen.zhou@intel.com>

Author: yuwenzho

examples/onnxrt/nlp/huggingface_model/token_classification/layoutlm/quantization/ptq_dynamic/main.py

@@ -244,6 +244,7 @@ def get_label_list(labels):
         cache_dir=model_args.cache_dir,
         revision=model_args.model_revision,
         use_auth_token=True if model_args.use_auth_token else None,
+        use_safetensors = False,
     )
 
     # Tokenizer check: this script requires a fast tokenizer.

examples/onnxrt/nlp/huggingface_model/token_classification/layoutlm/quantization/ptq_static/main.py

@@ -248,6 +248,7 @@ def get_label_list(labels):
         cache_dir=model_args.cache_dir,
         revision=model_args.model_revision,
         use_auth_token=True if model_args.use_auth_token else None,
+        use_safetensors = False,
     )
 
     # Tokenizer check: this script requires a fast tokenizer.

neural_compressor/adaptor/ox_utils/operators/split.py

@@ -43,7 +43,10 @@ def convert_check(self, convert_format):
         node = self.node
         assert convert_format in ["static"], "convert format for {} should be in ['static']".format(node.op_type)
 
-        parent = self.quantizer.model.get_parents(node)[0]
+        parents = self.quantizer.model.get_parents(node)
+        if len(parents) == 0:
+            return False
+        parent = parents[0]
         children = self.quantizer.model.get_children(node)
         if (
             parent.op_type != "DequantizeLinear" or len(children) == 0 or not node.name.endswith("_quant")