* support zero for expert local dp
* fix above codes:
*treat optim.zero_world_size and optim.zero_local_rank as list in model_checkpoint.py and test_model_checkpoint.py
*add overlap and zero check for moe in args_sanity_check(.)
* fix/brocast should not in commu stream
* fix/brocast should not in commu stream
* feat: support allreduce grad using async op
* fix bug of async op
* use reduceop.avg
* use torch flat
* delete unused stream
* delete unused stream
* feat: overap allreduce with memcapy
---------
Co-authored-by: yingtongxiong <974106207@qq.com>
* feat(optimier.py): reduce memory footprint and avoid _check_overflow call
* feat(optimier.py): reduce memory footprint and avoid _check_overflow call
* feat(optimizer.py): overlap compute norm with allreduce
* update var and function name
* update function compute norm (#197)
Co-authored-by: ChenQiaoling00 <qiaoling_chen@u.nus.edu>
* feat(optimizer/hybrid_zero_optim.py): overlap gradients last bucket allreduce and compute norm (#196)
* support gradients allreduce and compute norm overlap
* fix para set error
* remove timer cal_norm for testing
* feat(optimizer/hybrid_zero_optim.py): support group global norm
* format(lint): fix lint error
* feat(optimizer/store.py): update code based on comment
---------
Co-authored-by: ChenQiaoling00 <qiaoling_chen@u.nus.edu>
Co-authored-by: huangting4201 <1538303371@qq.com>
Wenwen Qu