5fa1a76
1
To configure PyTorch AMP-like fp16 mixed precision reduces memory usage and accelerates training speed.
