Remove backend argument of init_parallel_env #26773
Conversation
|
Thanks for your contribution! |
| # directly, if they want to switch default place, | ||
| # they need to call a function to change default place, | ||
| # here just set correctly place to users | ||
| place = core.CUDAPlace(ParallelEnv().device_id) |
There was a problem hiding this comment.
Can I call this function with CPU whl?
There was a problem hiding this comment.
thx, add cuda check for this function
guru4elephant
left a comment
guru4elephant
left a comment
There was a problem hiding this comment.
LGTM, I think init_parallel_env() can be used in CPU whl, the difference is we init gloo related functions. For GPU whl, we'd better init both nccl and gloo for compatibility of cpu/gpu collective operators. @sandyhouse @chenwhql please improve this function in the following updates.
thx, of course, we should support init CPU parallel env in this function later |
6 participants
PR types
Function optimization
PR changes
APIs
Describe
Remove backend argument of init_parallel_env, reason:
related PR: #26044