[WIP] Initial implementation of AMP support#707

Closed

BenjaminBossan wants to merge 5 commits intomasterfrom

feature/support-automatic-mixed-precision

Commits on Oct 3, 2020

[WIP] Initial implementation of AMP support
BenjaminBossan
committed
Skip AMP test if PyTorch version below 1.6
BenjaminBossan
committed

Commits on Nov 1, 2020

GradientNormClipping checks if optimizer is unscaled
BenjaminBossan
committed

Commits on Jan 16, 2021

Merge branch 'master' into feature/support-automatic-mixed-precision
BenjaminBossan
committed
Using cuda + amp in MNIST example makes no sense
BenjaminBossan
committed