Uable to load model when using more_general_trainer_metric branch

Hey everyone
I tried to use code from bert2gpt2-cnn_dailymail-fp16 in order to train for different models but whe I want to load GPT-2 model

model = EncoderDecoderModel.from_encoder_decoder_pretrained("HooshvareLab/bert-fa-base-uncased", "flax-community/gpt2-medium-persian")

i got this:

OSError: Can't load config for 'flax-community/gpt2-medium-persian'. Make sure that:

- 'flax-community/gpt2-medium-persian' is a correct model identifier listed on ''

I also tried to clone the model repo manually and load it locally but I got this error

OSError: Unable to load weights from pytorch checkpoint file. If you tried to load a PyTorch model from a TF 2.0 checkpoint, please set from_tf=True.

Note that I use colab and I install more_general_trainer_metric branch of transformer with this command:

!pip install git+git://

Any help would be appreciated
Thanks :slightly_smiling_face: