split "Latent Caching" into "Image Caching" and "Text Caching"#1462
Open
dxqb wants to merge 3 commits into
Open
split "Latent Caching" into "Image Caching" and "Text Caching"#1462dxqb wants to merge 3 commits into
dxqb wants to merge 3 commits into
Conversation
Rename `config.latent_caching` to `config.image_caching` and add a new orthogonal `config.text_caching` field. Adds config migration 10 -> 11 mapping the old field to both new fields, preserving existing behavior. This enables the previously-impossible (no image cache, yes text cache) configuration. Updates modelSetup placement decisions (VAE/effnet -> image_caching, text encoder -> text_caching), dataloader mixin gating, trainer cache-clear/start-next-epoch logic, UI switches, and tracked preset JSONs.
dxqb
added a commit
that referenced
this pull request
Jun 3, 2026
- BaseAnimaSetup: per-component checkpointing_or_offloading_enabled(), remove weight_list from create_autocast_context / disable_fp16_autocast_context - AnimaFineTune/LoRASetup: latent_caching → image_caching / text_caching - ModelType: add ANIMA to _MODEL_PARTS and supported_training_methods Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
dxqb
added a commit
to TheForgotten69/OneTrainer
that referenced
this pull request
Jun 3, 2026
…d "Text Caching") into preview
This comment was marked as resolved.
This comment was marked as resolved.
dxqb
added a commit
that referenced
this pull request
Jun 14, 2026
# Conflicts: # modules/ui/TrainUI.py # modules/util/config/TrainConfig.py
dxqb
added a commit
that referenced
this pull request
Jun 14, 2026
Anima, Lens, and Ideogram setup files used the pre-#1476/#1462 4-arg create_autocast_context/disable_fp16_autocast_context (with a weight-dtype list), the old config.gradient_checkpointing.enabled() global check, and the renamed config.latent_caching field. Update them to the current 3-arg autocast helpers, per-part checkpointing via enable_checkpointing_for_*, and config.image_caching/config.text_caching.
Resolve conflicts: renumber the image/text caching split as config migration 11, since master had already taken slot 10 for an unrelated offload fan-out migration. Drop modules/ui/TrainUI.py, removed on master in favor of the Base/Ctk/PySide6 view split.
Ernie, Ideogram, Anima, and Krea2 were added to master after this branch was created, so they still used the old latent_caching field. Ernie and Ideogram: gate prune_masked_tokens/pad_masked_tokens on text_caching (Ideogram also gets a real text_caching value instead of a hardcoded True). Krea2: same, following ChromaBaseDataLoader's train_text_encoder_or_embedding()-guarded pattern. Anima/Ideogram/ Krea2 setups: split VAE placement (image_caching) from text encoder placement (text_caching). Also updates the ctk data tab, whose old "Latent Caching" switch never got split.
dxqb
added a commit
to dxqb/OneTrainer
that referenced
this pull request
Jul 11, 2026
…d "Text Caching") into preview # Conflicts: # modules/ui/BaseTrainUIView.py
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.
with text caches becoming very large on recent models, you might want to disable text caching - but still do image caching
this PR splits this setting into two: