Closed ReDeiPirati closed 7 years ago
(this seems to be the same error I mentioned in #6 )
Agree with @jmhessel. Did you use an external global_step?
The global_step argument in this line https://github.com/JianGoForIt/YellowFin/blob/master/tuner_utils/yellowfin.py#L204 is an dummy argument.
Yes there is another global_step variable but it's correctly initialized. Unfortunately the last merge on tensor2tensor
has brought some bugs on the models in which i've tested YellowFin. I need to investigate deeper, for the moment i close.
Something should probably be done with that dummy argument, but I didn't want to mess anything up (i.e., I wasn't 100% sure global step tracked by YF was the same as the one passed by keras)
I am trying to adapt YellowFin to be usable as optimizer in tensor2tensor(it's use tensorflow>=1.2.0rc1) but unfortunately i cannot debug this error:
Step to reproduce
starter.sh
script (inside a Docker container is better).nvidia-docker run -it -v $(pwd):/t2t -p 6006:6006 -w /t2t tensorflow/tensorflow:latest-devel-gpu
.Error
If you do not want to help or contribute, please close the issue and forgive me. Otherwise, i will appreciate any help :)
I've also tried to write YellowFin as an tf.train.Optimizer, but going at C++ level seems to be out of my skills at the moment...