[DataFlow runtime 6/7] Training: TrainerCore/Controller, FSDP backend, strategies#599
Merged
jiapingW merged 1 commit intoJun 25, 2026
Conversation
maocheng23
requested review from
FlamingoPg,
FrankLeeeee,
shuaills and
sleepcoo
as code owners
Contributor
|
Warning You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again! |
…ies) Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
maocheng23
force-pushed
the
dataflow-up-6-training
branch
from
6f47ecd to
2775580
Compare
jiapingW
approved these changes
Jun 25, 2026
jiapingW
merged commit 39b3d11
into
sgl-project:dataflow-up-5-inference
2 checks passed
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
DataFlow runtime — stacked PR. Stacked on #598 — true-stacked: this PR's base is the previous PR's branch, so the diff below shows only this layer.
Part 6/7 — training (TrainerCore/Controller, FSDP backend, strategies).
Adds
specforge/runtime/training/.TrainerCore/TrainerControlleris a branch-free loop (global_stepcounts optimizer steps; gradient accumulation; checkpoint/eval hooks; ack at the optimizer boundary).FSDPTrainingBackend+ParallelConfigwrap the model preserving TP + Ulysses/Ring SP, with the optimizer built over the wrapped module so FSDP is in the forward/backward path.Eagle3TrainStrategy(+DFlashTrainStrategy) own loss/projection so the core stays branch-free (theFeatureSpec.target_reprtagged union selects hidden_state vs (pruned-)logits). Tests:test_trainer,test_seam_fixes. Additive.Part of a 7-PR series adding the DataFlow runtime (
specforge/runtime/, milestones M1–M4). Verified on current upstreammain: all subpackages import and 65 component tests pass. The integration launcher (launch.py+train_eagle3_dataflow.py) and the end-to-end equivalence gates are a deliberate follow-up, not in this series.