Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

This adds smddp backend support for training on Amazon SageMaker as discussed in #567

Modification

SMDDP launches with MPI, so this updates get_comm_device and _init_dist_mpi to support this backend.

BC-breaking (Optional)

I don't think there's any breaking changes.

Use cases (Optional)

This is for use with the SageMaker PyTorch Estimator. For example:

from sagemaker.pytorch import PyTorch

est = PyTorch(
    image_uri=image_uri,
    source_dir='code',
    entry_point='sagemaker_mmdetection_train.py',
    instance_type='ml.p4d.24xlarge',
    instance_count=2,
    hyperparameters={
        'config': '/opt/ml/code/configs/yolox/yolox_s_8xb8-300e_coco.py',
        'work-dir': '/opt/ml/checkpoints',
        'launcher': 'mpi',
        'cfg-options': "env_cfg.dist_cfg.backend='smddp'",
    },
    distribution = {'smdistributed': {'dataparallel': {'enabled': True}}}
)

est.fit({'coco' : s3_data_path})

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.