-
Notifications
You must be signed in to change notification settings - Fork 62
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
Fix GQA permutation computation and sequential weight initialization / loading when doing TP #531
Conversation
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
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.
Will restore this change before merging.
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.
LGTM. Thanks for the fix!
local_rank = xm.get_local_ordinal() | ||
if num_ranks_per_loading_step < 0: | ||
num_ranks_per_loading_step = get_local_world_size() | ||
for worker in range(math.ceil(get_local_world_size() / num_ranks_per_loading_step)): |
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.
So there will be two workers (0, 1) if get_local_world_size() / num_ranks_per_loading_step
> 1 and < 2?
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.
I can set num_ranks_per_loading_step = min(num_ranks_per_loading_step, get_local_world_size())
as a safety measure.
Ran the distributed tests locally and they pass. It's just a flaky test that needs to be solved. |
What does this PR do?
indicies -> indices