Translate so slow

Support
1

I have my data about 3.4M sentence. My train take me 40h to complete 10000 step. But my translate take 7 day to finish 640000 lines. Is this normal ?? and how can i improve speed of this. I train and translate on CPU with 32 cores. And batch_size is 30

2

You need a GPU, otherwise, no matter how many cores, it will take forever.

1 Like
3

I don’t know if there is a more uptodate tutorial, but if you don’t have a Nvidia that support CUDA… or if you want the easy way… just use Google colab (free).

have a look at this tutorial:

4

there is one more problem.In translate.py, I want to translate only one sentence that i input not using src file. So how can i do that ? I cant find it anywhere

5

I have never done that so far, but I would recommend you to have a look at this post:

https://forum.opennmt.net/t/simple-web-interface/4527