How do i understand approximate time for the training of dataset with 3 gb size

[JRMeyer]

>>> Denyz_Pylypenko
[February 2, 2021, 10:20am]

I downloaded Russian dataset of size 3 GB and I am running training on
2-core machine. My recent checkpoint just passed point of 150k and it
has been running for a last few months.

Can I somehow estimate the amount of time left until it is done?

[This is an archived TTS discussion thread from discourse.mozilla.org/t/how-do-i-understand-approximate-time-for-the-training-of-dataset-with-3-gb-size]

[JRMeyer]

>>> othiele
[February 2, 2021, 10:24am]

If you are running on CPU only, this can take a while.

Take the nr of chunks and divide by the batch size. This will give you
the steps for each epoch.

Hint: Use a GPU machine on Google Colab. You'll have a model after a
couple of hours. But it is not permament
So download your results.

[Archived Post]

[JRMeyer]

>>> Denyz_Pylypenko
[February 2, 2021, 10:43am]

Sounds amazing. Will definitely try Google Colab.

[Archived Post]

[JRMeyer]

>>> Denyz_Pylypenko
[February 23, 2021, 7:42pm]

hey, Olaf slash
can you please tell me where I can find a full guide on how to train
Mozilla Voice datasets on Google Colab

p.s. I am JavaScript engineer so I have some difficulties understanding
Python

[Archived Post]

[JRMeyer]

>>> othiele
[February 23, 2021, 8:02pm]

I guess you have 2 choices as I don't have a running Colab currently:

1. Find a Linux server you can use and go by the playbook linked in the
guidelines.

2. Search here for a Colab that works. I remember one just recently for
inferencing, but might have been training too. Or simply do one step
after the other as in the readme. It works. Post everything in a new
thread and I am happy to help.

[Archived Post]