[DO NOT MERGE] refactor(fsdp)(4.1/5): per-parameter weight-precision control by FSDP2 expansion patch - #98
Closed
Rockdu wants to merge 7 commits into
Closed
Conversation
Compile model precision patterns into FSDP-local maps so numerically sensitive Wan parameters stay in FP32 without giving up BF16 compute. Co-authored-by: Cursor <cursoragent@cursor.com>
Exercise patch guards, packing paths, policy validation, scoped integration, and frozen FP32 parameters while keeping the feature independent of image rebuilds.
Validate the base PyTorch release and source hashes without tying the monkeypatch to one CUDA wheel suffix.
Make FSDP wrapping and precision policy model-owned so Diffusers and native backends share one explicit parallel-plan interface.
Rockdu
changed the title
Show each upstream block directly above its replacement and normalize empty dtype maps to the unpatched policy path.
Replace the prose summary with the original FSDPParam construction code so the monkeypatch remains directly reviewable.
Rockdu
marked this pull request as ready for review
August 6, 2026 22:55
Keep the monkeypatch structurally aligned with the proposed PyTorch change while preserving the zero-temporary mixed-gradient packing path. This PR was authored with Claude. Co-authored-by: Cursor <cursoragent@cursor.com>
This was referenced Aug 7, 2026
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
This design is outdated; parts are split into different subsequent PRs; see the latest #100 #103 #104