Enable CUDA path in Python FE, update grpo example #1224
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
meta-cla
bot
added
the
CLA Signed
|
@dstaay-fb has exported this pull request. If you are a Meta employee, you can view the originating diff in D82331433. |
) Summary: This diff integrates Monarch RDMA w/ PyTorch's caching allocator for CUDA devices. Typical users will want to create tensors (managed my PyTorch's caching allocator) and just pass to RdmaBuffer(..). However we can't keep generating BAR1 memory registrations every time given limits around total registration space (~128GB est). So in this diff, - monarch RDMA inspects the actual memory segments registered inside caching allocator, and only creates MR when needed. - If a segment expands during program lifecycle, we can use mlx5dv advanced APIs to build rdma regions which are backed by multiple MR regions; and just hand out new lkey/rkey to new tensor within expanded segment (old keys remain valid also). - To spy on PT caching allocator we leverage snapshot API whenever we encounter a new register_buffer request with a virtual address range outside prior known segments. Other stuff: - It doesn't make sense to MR's to RdmaDomain, moving up to Manager level. Reviewed By: zdevito Differential Revision: D81736949
Summary: Enable device='cuda' for RDMABuffer Reviewed By: zdevito Differential Revision: D82331433
dstaay-fb
added a commit
to dstaay-fb/monarch
that referenced
this pull request
Sep 16, 2025
Summary: Enable device='cuda' for RDMABuffer Reviewed By: zdevito Differential Revision: D82331433
dstaay-fb
force-pushed
the
export-D82331433
branch
from
c6e2092 to
9b03dda
Compare
|
@dstaay-fb has exported this pull request. If you are a Meta employee, you can view the originating diff in D82331433. |
dstaay-fb
force-pushed
the
export-D82331433
branch
from
9b03dda to
d6af921
Compare
|
@dstaay-fb has exported this pull request. If you are a Meta employee, you can view the originating diff in D82331433. |
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.
Summary: Enable device='cuda' for RDMABuffer
Reviewed By: zdevito
Differential Revision: D82331433