Repository navigation
0.18.0: epochs, validation pack, held-out photos, Fizgig target - #4
Merged
Merged
Conversation
- ⑤ counts epochs (one checkpoint + one sample set each); Repeats 0 = auto, sized to n x 75 target steps; Fizgig auto = 1 - ⑤ live exposure line (images x repeats / batch = steps/epoch x epochs) with low/high warnings - every config writes validation/: 8 fixed prompts at seed 42, validation.md, per-epoch score sheet - ai-toolkit saves/samples once per epoch, keeps every epoch, seed 42 with walk_seed off - musubi/kohya/Fizgig sample every epoch from the prompt file; musubi Krea 2 samples via --turbo_dit - Fizgig trainer target (Krea 2, Qwen Image 2.1 at 704, Klein) with its cache/train commands - ① writes a dd_source provenance PNG chunk; ⑤ names upscaled sources; ④ records provenance in metadata.json - ④ "Hold out N reference photos" (+ cli export --holdout) copies the largest real photos to <dataset>-heldout/ - dataset naming skips names whose -heldout sibling exists - version 0.18.0; AGENTS.md, ARCHITECTURE.md, README updated
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Phase 1 of the LoRA best-practice update: training export.
validation/pack: 8 fixed prompts at seed 42,validation.md(how to pick the epoch, inference settings, held-out photos), and a per-epoch score sheet.--turbo_dit.docs/CLI.md, with<<FILL>>model paths.dd_sourceprovenance PNG chunk. ⑤ names the sources that were upscaled, and ④ records real-photo provenance inmetadata.json.cli export --holdout). The largest real photos are copied to<dataset>-heldout/, never into the training folder.Deferred to Future ideas: the optimizer dropdown and the attention-only toggle.
Tests:
pytest -q(616 passed),ruff check .clean.