Description

I am using the checkpoints available for the English to German translation task.

The problem is translate_ende_wmt32k.
The model is transformer.
The hyper-parameters set used is transformer_base.

These are the translations got from the model:
EX1:
ENGLISH: Today is a hot day in Folsom. The temperature is about a 30 degree Celsius.
GERMAN: Heute ist ein heißer Tag in Folsom, die Temperatur beträgt etwa 30 Grad Celsius.

EX2:
ENGLISH: Today is a hot day in Folsom. The temperature is about a 30 degree Celsius. Don't forget your sunglasses.
GERMAN: Die Temperaturen liegen bei 30 Grad Celsius und vergessen Sie nicht Ihre Sonnenbrille.

EX3:
ENGLISH: Today is a hot day in Folsom. The temperature is about a 30 degree celsius. Don't forget your sunglasses and sunscreen. Also, make sure you drink lots of water.
GERMAN: Die Temperaturen liegen bei 30 Grad Celsius und die Sonnenbrille und die Sonnencreme sind ebenfalls sehr angenehm.

These were the observations I made:
The German translation will never have multiple sentences, even if the input English text had. The German translation is always comma separated.
When the English text has multiple sentences, it is not translating all of them.

Is anyone else getting similar outputs? Am I doing something wrong to get such outputs?

Does the model always behave this way?

Any help will be greatly appreciated.

Environment information

OS: <your answer here> $ pip freeze | grep tensor # your output here $ python -V # your output here 

For bugs: reproduction and error logs

# Steps to reproduce: ... 

# Error logs: ...