[docs] (non) determinism in TF #26147
Labels
stale
This label marks the issue/pr stale - to be closed automatically if no activity
stat:awaiting response
Status - Awaiting response from author
type:docs-bug
Document issues
Describe the feature and the current behavior/state.
A documentation explaining determinism and sources of non determinism in TF applications. It should cover things from having multiple readers, seeds in the specific ops, seed in numpy, PYTHONHASHSEED, non-deterministic ops in CUDA (and why TF does not support passing deterministic flag even though it would be slower),
TF_CUDNN_USE_AUTOTUNE
, and probably some others I do not know about.Who will benefit with this feature?
I take part in many Kaggle competitions and very often people there have problems with non-determinism in NN.
The text was updated successfully, but these errors were encountered: