Which Machine configuration works best for Machine Translation using OpenNMT?

I am working on Hindi to English Translation using OpenNMT. In my 2nd try I got this result

Train perplexity: 13.3963
Train accuracy: 51.4974
Validation perplexity: 23.7748
Validation accuracy: 44.4444
Decaying learning rate to 0.00195312

I am Yet to understand it correctly but results are pretty good. It took around 34 + hours. I want increase the dataset, epochs, layers to increase the accuracy and reduce execution time.

Currently I have 16GB CPU system. Is this enough? which system can reduce the execution time?

Please suggest me best suitable system configuration for this experiments.


  • What GPU are you using?
  • Is 16GB your GPU or system memory?

Currently I am using CPU system and 16GB is RAM for CPU machine.

So the answer is a system with a GPU. Currently, the GTX 1080 TI seems to be a solid choice.