fix(conv): scale gradients before the fp16 cast so small values survive - #23

fix(conv): scale gradients before the fp16 cast so small values survive#23

AlexanderSoftCode merged 1 commit into

mainAlexanderSoftCode/Aether-ML:mainfrom

fix/conv-gradient-underflowAlexanderSoftCode/Aether-ML:fix/conv-gradient-underflowCopy head branch name to clipboard