Fix Flux 2 Klein dreambooth default text_encoder_out_layers#13509
Open
Ricardo-M-L wants to merge 1 commit intohuggingface:mainfrom
Open
Fix Flux 2 Klein dreambooth default text_encoder_out_layers#13509Ricardo-M-L wants to merge 1 commit intohuggingface:mainfrom
Ricardo-M-L wants to merge 1 commit intohuggingface:mainfrom
Conversation
Align the `--text_encoder_out_layers` CLI default in `train_dreambooth_lora_flux2_klein.py` with the layers the Klein pipeline actually consumes internally: `(9, 18, 27)` instead of `(10, 20, 30)`. The vanilla Flux 2 pipeline (`pipeline_flux2.py`) uses hidden states `(10, 20, 30)` (Mistral text encoder), and its training script was correctly set to that. The Klein script was copy-pasted from the non-Klein one but the default was never adjusted, while Klein's pipelines (`pipeline_flux2_klein*.py`) expect `(9, 18, 27)` — an off-by-one that silently sends the training run to the wrong layers unless the user overrides the flag on the CLI. Fixes huggingface#13445
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What does this PR do?
Fixes the off-by-one in the `--text_encoder_out_layers` CLI default of `examples/dreambooth/train_dreambooth_lora_flux2_klein.py`: the script was defaulting to `[10, 20, 30]` (the Flux 2 non-Klein default, appropriate for its Mistral text encoder), but every Klein pipeline — `pipeline_flux2_klein.py`, `pipeline_flux2_klein_kv.py`, `pipeline_flux2_klein_inpaint.py` — expects `(9, 18, 27)` and uses that layer set on every call in the training loop.
The result is a silent mismatch whenever a user trains with the default flag value: Klein's internal `hidden_states_layers=(9, 18, 27)` is what the pipeline advertises, but the training script's default CLI flag suggests `[10, 20, 30]`, which produces the wrong embeddings if actually passed. The issue reporter lost ~72 GPU hours before catching it.
Change
One-line fix in `examples/dreambooth/train_dreambooth_lora_flux2_klein.py`: change the argparse default for `--text_encoder_out_layers` from `[10, 20, 30]` → `[9, 18, 27]` so it matches the layers Klein's pipelines already use.
The non-Klein Flux 2 training script (`train_dreambooth_lora_flux2.py`) stays on `[10, 20, 30]`, which is correct for the non-Klein pipeline.
Fixes #13445
Before submitting
Who can review?
cc @sayakpaul @yiyixuxu