`torch.concat` is supported when reproducing results with docker

Many thanks for the kind sharing!!

When reproducing training results, the docker file specifies `torch==1.8.1`. However, torch at 1.8.1 had not introduced `torch.concat` yet, which will cause errors on this line of code:

https://github.com/EleutherAI/gpt-neox/blob/71df4d5017f9f4919566a11454fe3a507ffdc632/megatron/model/transformer.py#L432

Simply changing `torch.concat` to `torch.cat` can make it compatible.