Commit 23ec33ae authored by Reed Wanderman-Milne's avatar Reed Wanderman-Milne Committed by TensorFlower Gardener
Browse files

Implement LossScaleOptimizer. 



Loss scaling is necessary for mixed precision training when using float16. This change only implements fixed loss scaling, and dynamic loss scaling will come in a future change.

Co-authored-by: default avatarJames Qin <jamesqin@google.com>
PiperOrigin-RevId: 236948980
parent ff513ac5
Show whitespace changes
Inline Side-by-side
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Please to comment