3 ms·
We’re Training AI Twice as Fast This Year as Last
- nharada 4y agoThe scalability of deep nets has made amazing progress since I stated working with them. Even just 5 years ago it was a challenge to train on multi-GPU, and training ResNet would take 24 hours on a standard GPU in those days. Now it's 10 seconds? Insane.