OpenNMT Forum

Running translate on Quickstart demo .pt

I’m running the Quickstart demo.
Training worked.

To run translate do I run it with the LAST demo .pt file?

onmt_translate -model demo-model_step_100000.pt -src data/en2.txt -output data/pred2.txt

Q1. Is that right? I use the latest .pt?
Q2. It’s 100K sentences I think. How bad should I expect?

It can’t translate

‘I would like milk and bread’!

I get:

‘Ich möchte Nicole Soulier aufführen’

A1: Yes, the latest pt-file is good. You can average the last checkpoints in tensorflow with the option average_last_checkpoints
A2: 100K sentences is a low resource setting and I think it is quite good that a grammatical correct sentence was generated.

Try to adapt pretrained models to see faster results with your dataset.