Multiple GPUs

[wangqian621]

How to use multiple GPUs for training? Do I need to write the function of setting multiple GPUs into the model？

[airsplay]

You can try with distributed data parallel: https://pytorch.org/tutorials/intermediate/ddp_tutorial.html. However, since different nav rollouts have different length and PyTorch DDP uses synchronized update, the speed of each step would be equal the the slowest one.