fix(sft): drop the cond cast deleted with loss_hub.utils - #107
Merged
Conversation
Rockdu
force-pushed
the
fix/sft-cond-cast
branch
from
August 8, 2026 08:39
7116181 to
e5fe336
Compare
Rockdu
force-pushed
the
fix/sft-cond-cast
branch
from
August 8, 2026 08:41
e5fe336 to
8c89aaf
Compare
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
loss_hub/sft.py's import of the deletedloss_hub.utils.cast_cond_to_dtypeand its cond pre-cast, mirroring what refactor(fsdp)(4.3/5): family-declared input dtype policy at the model boundary #103 did toflow_grpo/nft(whose leftover comment lines are dropped too).Why
#90 (SFT) merged first and imports
loss_hub.utils; #103 merged after and deleted it — both PRs were green, but every stage-a-cpu run on the merged main now fails at collection withModuleNotFoundError. Boundary input dtypes are familyinput_dtype_policyterritory since #103, and SFT batches flow through the same actor boundary; Wan declares no policy today, deliberately left as a separate family decision.Checklist
pre-commit run --all-filespassestest_loss_hub_sft.pypytest -xis green — not run locally (no local torch); stage-a-cpu re-runs the suite