set proper output layer name

convert.py

@@ -13,6 +13,6 @@ inputs = {"input_ids": torch.ones(1, max_seq_length, dtype=torch.int64),
 symbolic_names = {0: 'batch_size', 1: 'max_seq_len'}
 
 torch.onnx.export(model, args=tuple(inputs.values()), f='pytorch_model.onnx', export_params=True, 
-    input_names=['input_ids', 'attention_mask', 'token_type_ids'], output_names=['output'],
+    input_names=['input_ids', 'attention_mask', 'token_type_ids'], output_names=['last_hidden_state'],
     dynamic_axes={'input_ids': symbolic_names, 'attention_mask': symbolic_names, 'token_type_ids': symbolic_names})
 

pytorch_model.onnx

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:79c87071eeceb3a40ef53b96aa8b3f9bfd4d3152023a5ae25ed134eed59c724e
-size 90984197
+oid sha256:47206d2e5b865151da68245745c426ebc496d21200bbd73154c3a2ddc699b0a1
+size 90984263