How can I control epoch num?

[playma]

The README.md said that the training log will like this: # Start epoch 0, step 0, lr 1, Tue Apr 25 23:17:41 2017

but I got # Start step 0, lr 1, Thu Dec 21 10:44:55 2017

There is no epoch information, and how can I input the epoch I want to train.

[StupidL]

You can add a flag --num_train_steps . For example:

python -m nmt.nmt \
    ... \
    --num_train_steps=10000 \
    ... 

[oahziur]

@playma

In addition to the --num_train_steps, you can get some epoch info here.

[nbro]

@oahziur Anyway, you may want to clarify (in the tutorial) why every epoch apparently takes 1000 steps.