torch.distribute.launch -> torchrun

Need to replace the deprecated distribute wrapper found here https://github.com/facebookresearch/FAMBench/blob/main/benchmarks/bevt/ootb/run_bevt_train.sh#L52
with `torchrun`.

See here: https://pytorch.org/docs/stable/distributed.html

The deprecated command leads to a syntax error in BEVT, sending in `--local-rank` but expects `--local_rank`