Multi gpu training

[Jaster1999]

Feature Description

Split training across multiple gpu's. Not sure if this is already a feature but I can see no docs on it.

[henrytsui000]

Multi-GPU training is enabled by default.

[Jaster1999]

so simply calling: "python yolo/lazy.py task=train dataset=** use_wandb=True" will use all available gpus by default and split the btachsize across gpus or will it use the batchsize per gpu?

[henrytsui000]

the batchsize per GPU, note: multi-GPU training performance may be as good as single-GPU.