Moritz Freidank

@duncanriach Thank you! I can reproduce the resolution and things are now deterministic in the scenario above - should have taken my own advice from the notes based on your...

@duncanriach Thank you a lot for your work and drive on this and for the conclusive summary of where we stand and what we know. I agree with all your...

@dmitriydligach Did you ever get this resolved?

@dmitriydligach Just to verify: your code becomes fully reproducible with pytorch?

> PyTorch has potentially different non-deterministic ops than TensorFlow, and no general mechanism, yet, to enable deterministic op functionality. Both PyTorch and TensorFlow now have the ability to enable deterministic...

@duncanriach Thanks a lot for taking the time to look into this and for your encouragement. I feel much more confident about this now, knowing that someone with your experience...

@duncanriach Non-reproducibility of the code of @dmitriydligach may be related to him training for multiple epochs, see my update on issue #19.

Hi @phqtuyen, @duncanriach Sorry for the wait, I had been pulled into other things. I've picked up work on this now and will have an update to the PR before...

Hi Duncan, Sorry for the delay on this. Yes, I'll pick it up starting tomorrow my time. Duncan Riach schrieb am Fr., 4. Sep. 2020, 20:07: > Hi @Freidank, I'm...

Hi again @duncanriach, I've implemented a first version (on top of the most recent `master` branch with `enable_determinism`) that patches only `tf.nn.sparse_softmax_cross_entropy_with_logits` and `tf.nn.softmax_cross_entropy_with_logits` and integrated t[he unit test case...