Integrate MXFP8 weights into experimental MFSDP - #7114
Open
wujingyue wants to merge 3 commits into
Open
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
wujingyue
changed the title
wujingyue
force-pushed
the
agent/mfsdp-grouped-dbuffer
branch
from
September 8, 2026 19:26
56ace1a to
b00d660
Compare
wujingyue
commented
Sep 10, 2026
wujingyue
commented
Sep 10, 2026
wujingyue
commented
Sep 11, 2026
wujingyue
commented
Sep 11, 2026
wujingyue
commented
Sep 11, 2026
wujingyue
changed the title
wujingyue
commented
Sep 11, 2026
wujingyue
commented
Sep 11, 2026
wujingyue
changed the title
wujingyue
force-pushed
the
agent/mfsdp-grouped-dbuffer
branch
from
September 11, 2026 22:02
f506fb8 to
333855f
Compare
wujingyue
changed the base branch from
pull-request/6978
to
pull-request/7257
September 11, 2026 22:02
wujingyue
commented
Sep 11, 2026
wujingyue
commented
Sep 11, 2026
|
|
||
| def redistribute( | ||
| self, new_placements: Iterable[Placement], *, out: "QuantizedDBuffer | None" = None | ||
| ) -> "QuantizedDBuffer": |
Contributor
Author
There was a problem hiding this comment.
Does this return a new QuantizedDBuffer even when out is specified?
wujingyue
changed the title
wujingyue
changed the base branch from
pull-request/7257
to
pull-request/7265
September 11, 2026 23:13
wujingyue
marked this pull request as ready for review
September 11, 2026 23:15
Contributor
Author
|
/ok to test 8689e05 |
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
wujingyue
force-pushed
the
agent/mfsdp-grouped-dbuffer
branch
from
September 11, 2026 23:18
8689e05 to
8ee05d3
Compare
copy-pr-bot
Bot
force-pushed
the
pull-request/7265
branch
from
September 11, 2026 23:18
654e29f to
c1f8617
Compare
Contributor
Author
|
/ok to test 8ee05d3 |
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Signed-off-by: Jingyue Wu <jingyuew@nvidia.com>
Contributor
Author
|
/ok to test 38fa0f6 |
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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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.
Integrate QuantizedDBuffer into experimental MFSDP for MXFP8 training.
Depends on #7265 (QuantizedDBuffer and buffer tests), which depends on #7257 (DBuffer layout allocation) and #6978 (BlockAtomic placement).
Remaining scope:
The split preserves the complete source tree of the previously tested #7114 head; only commit ancestry and PR boundaries change. Buffer implementation and its standalone tests are reviewed in #7265.
Limitations: 2D MXFP8 weights with dimensions divisible by 32 and aligned row shards. Compact scales require padded TE allocations when necessary. Two previously identified CPU/meta initialization test failures remain outside this change.