-
Notifications
You must be signed in to change notification settings - Fork 54
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
docs: Improve documentation for collective operations #1084
Conversation
Force-push for commit cleanup. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks a lot @swaptr, this looks so much better!
In the Alltoallv
function, will you add the description of sendbuf
and recvbuf
as tuples of buffer, counts and displacements (type hints and docstrings)? Thanks!
Thanks for reviewing this @ClaudiaComito . Can you check if its correct now? |
sendbuf: Tuple[DNDarray, torch.Tensor, Any], | ||
recvbuf: Tuple[DNDarray, torch.Tensor, Any], |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
thanks @swaptr , you're actually validating my impression that the documentation is unclear, so thanks a lot for tackling this.
For Alltoallv, sendbuf
and recvbuf
must be a tuple of: data object, counts, displacements. Check out the official OpenMPI documentation. Here's another explanation I get back to often as it's easier for me to understand.
Here's an example in our ht.reshape()
function
Thank you for the PR! |
Hi @swaptr , you can leave this PR open but will you consider allowing us to make changes? We're looking forward to merge your PR. Thanks! |
Description
Current documentation for
communication.Alltoall
andcommunication.Alltoallv
does not represent the different construction ofsendbuf
andrecvbuf
. Also, the documentation of non-blocking methods for this same are also updated.Issue/s resolved: #1072
Changes proposed:
Type of change
Memory requirements
NA
Performance
NA
Due Diligence
Does this change modify the behaviour of other functions? If so, which?
no