OpenNMT Forum

Python OpenNMT training

The program gives a lot of unk as output after training.

You might want to give more details about your setup, what you’re trying to achieve etc. if you want some help.

The Program is here.

I don’t have time to look at this in details but it seems you’re trying to translate with a model that 1. has trained only 1000 steps 2. on very little data.
Also, you might want to train first, then translate. The training logs were truncated here.

You may look at the Translation example in the docs for a base setup.

Copy of Copy of Copy of OpenNMT-py Quickstart - Colaboratory (google.com)

This program has training and translation split and has large dataset but also gives unks.

Check your training logs, with such training and validation accuracy it looks like your model hasn’t learned much. You may try training longer, or maybe simplifying the task to get started.

Copy of Copy of Copy of OpenNMT-py Quickstart - Colaboratory (google.com)

I increased the steps and I added replace unk but translations seem strange.

May be 150k sentences, with 10.000 steps may not be enough what say you @francoishernandez @Plkmoi ?