Running Code Consistent Zero-shot NMT on multi-GPU has an error

I tried to run the run_nmt_experiment.sh with parse-args.sh
EXP_CONF_NAME="ag_gnmt_luong_att"
EXP_MODEL_NAME="agreement_multilingual_nmt"
EXP_PROBLEM_NAME="translate_iwslt17_nonoverlap"
and set --worker_gpu=2，but when in evaluation

> INFO:tensorflow:Evaluation [10/100]

got an error

> Traceback (most recent call last):
File "/home/jingzhao/.conda/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1334, in _do_call
return fn(*args)
File "/home/jingzhao/.conda/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1319, in _run_fn
options, feed_dict, fetch_list, target_list, run_metadata)
File "/home/jingzhao/.conda/envs/tensorflow/lib/python3.6/site-packages/tensorflow/python/client/session.py", line 1407, in _call_tf_sessionrun
run_metadata)
tensorflow.python.framework.errors_impl.InvalidArgumentError: ConcatOp : Dimensions of inputs should match: shape[0] = [12,77,1,1,32460] vs. shape[1] = [12,78,1,1,32460]
[[{{node agreement_multilingual_nmt/concat}}]]
[[{{node ConstantFoldingCtrl/mean/broadcast_weights/assert_broadcastable/AssertGuard/Switch_0}}]]

and final error info is

> InvalidArgumentError (see above for traceback): ConcatOp : Dimensions of inputs should match: shape[0] = [12,77,1,1,32460] vs. shape[1] = [12,78,1,1,32460]
[[node agreement_multilingual_nmt/concat (defined at /home/jingzhao/.conda/envs/tensorflow/lib/python3.6/site-packages/tensor2tensor/utils/t2t_model.py:240) ]]
[[{{node ConstantFoldingCtrl/mean/broadcast_weights/assert_broadcastable/AssertGuard/Switch_0}}]]

but set --worker_gpu=1, it seems ok.