Use tvm to optimize savedmodel generated by onnx-tensorflow caused error

[LotuSrc]

I use onnx-tf to convert a onnx model to savedmodel. When I call relay.frontend.from_tensorflow, caused TVMError: unknown type string. This is caused by 'saver_filename' PlaceHolder in tf graph. Why there is a such strange placeholder in converted savedmodel. The nodes in graph are as followed: Const Const Const_1 Const Const_2 Const Const_3 Const Const_4 Const Const_5 Const Const_6 Const Const_7 Const Const_8 Const Const_9 Const Const_10 Const Const_11 Const Const_12 Const Const_13 Const Const_14 Const Const_15 Const Const_16 Const Const_17 Const Const_18 Const Const_19 Const Const_20 Const Const_21 Const Const_22 Const Const_23 Const Const_24 Const Const_25 Const Const_26 Const Const_27 Const Const_28 Const Const_29 Const Const_30 Const Const_31 Const Const_32 Const Const_33 Const Const_34 Const Const_35 Const Const_36 Const Const_37 Const Const_38 Const Const_39 Const Const_40 Const Const_41 Const Const_42 Const Const_43 Const Const_44 Const Const_45 Const Const_46 Const Const_47 Const serving_default_input_ids Placeholder serving_default_input_mask Placeholder serving_default_segment_ids Placeholder PartitionedCall PartitionedCall NoOp NoOp Const_48 Const saver_filename Placeholder StatefulPartitionedCall StatefulPartitionedCall StatefulPartitionedCall_1 StatefulPartitionedCall init NoOp

Why there's just less than 100 nodes? When I convert the onnx model by tf1 verison, there's hundreds of nodes and I can explicitly see the net architecture.