Hello, I’m newbie in Neural Machine Translation world,
I have one question…
for train 1M parallel zh-en corpus with zh corpus tokenized with jiagu
What is the more recommended configuration?
Current i’m use the default train arguments:
python train.py -data data/demo -save_model demo-model
@woolz, Using OpenNMT you can play with different configurations. Currently, you are using Sequence to Sequence model with very minimal configuration. Just single Layer Encode-Decoder.
For some better results, try with Transform model, script with the exact configuration that Google used to produce State of art results are given in the FAQ section.
Happy Coding !!