deeplab model_variant="nas_pnasnet" - unable to export_model

Hello supporters

I am using tensorflow 1.13.1
with tensorflow model garden version 1.11

2. Describe the bug

export_model.py --checkpoint_path=myCheckpoint num_classes=2 --export_path=MyTarget.pb --model_variant="nas_pnasnet"
fails with InvalidArgumentError: Restoring from checkpoint failed
Assign requires shapes of both tensors to match. lhs shape= [3,3,40,1] rhs shape= [3,3,80,1]

my purpose right now it to train for 100 epochs, export a frozen graph , execute it on a few images and check the runtime of each model variant.

I have already managed to train and export on my dataset using model_variants mobilenet_v2, resnet_v1_50, resnet_v1_50_Beta, resnet_v1_101, resnet_v1_101_beta, xception_41, xception_65, xception_71.
but when i changed the model_variant value to nas_pnasnet ( on both train and export ) , export fails with the error described above.

( this error does looks differently to me from the very common misuse of the dataset[classes] paramter ).

please advise.
Thanks,
Omer.