Internal: No unary variant device copy function found for direction: 1 and Variant type_index: tensorflow::data::(anonymous namespace)::DatasetVariantWrapper

[maria364]

I am using this toolkit for language modelling and I would like to train a model with wiki103 dataset.
The training will take place without using Horovod and by using 2GPUs.
I have checked the parameters in the "lstm-wkt103-mixed.py" file and edited them according to the information provided in this page https://nvidia.github.io/OpenSeq2Seq/html/language-model.html.

The command used for training is:
CUDA_VISIBLE_DEVICES=0,1 python run.py --config_file=example_configs/lstmlm/lstm-wkt103-mixed.py --mode=train_eval --enable_logs

The error I am facing is available in the image below:

Could anyone please help?