Wondering why accuracy in NMTAttn is too low

Hi, I tried the following:
training_loop = training.Loop(NMTAttn(mode=‘train’),
train_task,
eval_tasks=[eval_task],
output_dir=output_dir)
training_loop.run(10)
Step 50: Ran 10 train steps in 616.65 secs
Step 50: train CrossEntropyLoss | 6.81675625
Step 50: eval CrossEntropyLoss | 7.13336086
Step 50: eval Accuracy | 0.04535790
I am wondering why accuracy in NMTAttn is too low?
also, i tried to translate the following sentence:
input: “While blood preasure is important, suggar level in your body is more critical”
output: “Während der Bluthochgesetz wichtig ist, darauf, ist suggansteigen in Ihrem Körper ist re kritischstimulär.”
the result is not satisfiable, can you please suggest methods or steps to improve it to meet production level accuracy?

Hi owis_kweder,

As a belated reply, the best way to improve accuracy is to use a transformer!