Training large neural networks faster with TensorFlow's mixed precision
TensorFlow's mixed precision alternates between using float32 and float16 numerical precision during training leading to up to 3x performance speed on modern GPUs and 60% on TPUs!
Original stream date: 09 Mar 2021
Connect elsewhere:
Web - https://www.mrdbourke.com
Main channel - / @mrdbourke
Get email updates on my work - https://www.mrdbourke.com/newsletter
Watch live at / mrdbourke