* deprecate fairscale's ShardedDDP * fix code style * roll back * deprecate the `sharded_ddp` training argument --------- Co-authored-by: jihuazhong <jihuazhong1@huawei.com>
* deprecate fairscale's ShardedDDP * fix code style * roll back * deprecate the `sharded_ddp` training argument --------- Co-authored-by: jihuazhong <jihuazhong1@huawei.com>