[Option] Parallelize preconditioners across ranks; multi-node FSDP #100
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
luciaquirke
changed the title
luciaquirke
force-pushed
the
multi-node
branch
from
8066d72 to
fa7f1b3
Compare
luciaquirke
force-pushed
the
multi-node
branch
from
9277080 to
a9d1531
Compare
luciaquirke
force-pushed
the
multi-node
branch
from
a9d1531 to
4061982
Compare
luciaquirke
changed the title
LouisYRYJ
reviewed
Dec 27, 2025
|
|
||
|
|
||
| @dataclass(kw_only=True) | ||
| class MultiNodeGradientCollector(HookCollectorBase): |
Contributor
There was a problem hiding this comment.
is this going to be a replacement for GradientCollector? It seems like we don't it, if we have this one
Collaborator
Author
There was a problem hiding this comment.
Yes, I will merge this as a separate class for dogfooding and then replace the GradientCollector when we're convinced it's stable
LouisYRYJ
reviewed
Dec 27, 2025
|
|
||
| def build_worker( | ||
| rank: int, | ||
| local_rank: int, |
Contributor
There was a problem hiding this comment.
add to doc what this does
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
More VRAM efficient variant where preconditioners can be spread across an arbitrary number of nodes to compute large outer products. This is useful because preconditioners are often applied to a query and then the query is run across a large dataset, so slow but VRAM-efficient preconditioner computation and usage is a scalable pattern.
Because the preconditioners don't necessarily fit on a single GPU we use GLOO to do distributed CPU operations.