Error when using multi-gpu

[sanchezismael]

If argument expected for --gpu is type int, it is not posible to train with more than two gpus

[codecov-io]

Codecov Report

Merging #1292 into master will not change coverage.
The diff coverage is 0%.

@@          Coverage Diff           @@
##           master   #1292   +/-   ##
======================================
  Coverage      23%     23%           
======================================
  Files          43      43           
  Lines        2535    2535           
======================================
  Hits          576     576           
  Misses       1959    1959

[hgaiser]

Fair point :)

I think this argument is used in more scripts, could you check them?

[wheemyungshin]

It still is causing probelm:
tf.config.experimental.set_visible_devices(gpus[int(gpu_id)], 'GPU')
as it cannot convert from '0,1,2,3' to int.

Also even if I specify some numbers like,
tf.config.experimental.set_visible_devices(gpus[0:2], 'GPU')
It outputs another error which is related to this page: keras-team/keras#13684