This leads to a significant performance increase, coupled with simpler code.
One major issue comes from the move to pytorch.distributed under the hood.
pytorch-lightning expects a lot to be managed by slurm. We have to set a range ov env variables in order to overcome this.
This leads to a significant performance increase, coupled with simpler code. One major issue comes from the move to pytorch.distributed under the hood. pytorch-lightning expects a lot to be managed by slurm. We have to set a range ov env variables in order to overcome this.