Multi-GPU Training Support

[GloryyrolG]

Hi @nkolot et al.,

May I ask if the code was also verified in the case of the gpus > 1 DDP setting?

ProHMR/train/train_prohmr.py

Line 50 in 6f778db

gpus=1,

[GloryyrolG]

Besides, when I tried gpus=2, accelerator='dp', it pumped errors.

RuntimeError: unscale_() has already been called on this optimizer since the last update().

KeyError: 'exp_avg_sq'