"Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes," Jia and Song et al., Tencent: https://t.co/AH3crhdBQH
— Miles Brundage (@Miles_Brundage) July 31, 2018
"Highly Scalable Deep Learning Training System with Mixed-Precision: Training ImageNet in Four Minutes," Jia and Song et al., Tencent: https://t.co/AH3crhdBQH
— Miles Brundage (@Miles_Brundage) July 31, 2018
The latest entry in turning ImageNet into MNIST: 75.8% top-1 test accuracy with ResNet-50 (90 epochs) in 6.6 minutes using 2048 Tesla P40 GPUs https://t.co/AxqiLdIOyQ 64K "mini"-batch size, mixed precision training, LARS, BN&bias weight decay at zero, custom all-reduce
— Andrej Karpathy (@karpathy) August 1, 2018