Could not decode variant with type_name: "tensorflow::DatasetVariantWrapper"

[wangshuaizs]

Hi,

When I tried to modify the source code to support distributed training, the following error happened:

E tensorflow/core/framework/variant.cc:102] Could not decode variant with type_name: "tensorflow::DatasetVariantWrapper".  Perhaps you forgot to register a decoder via REGISTER_UNARY_VARIANT_DECODE_FUNCTION?

It seems that tensorflow::DatasetVariantWrapper is not supported, but I did run it successfully in a standalone machine. Anyone knows why the error happens?

Thanks

[elzino]

I have the same issue. Did you resolve this?

[severussong]

I have the same issue too. It seems that make_initializable_iterator does not work

[lesserror]

My case is that when I bacth the dataset this error appears, but when I cancel it other error happens.