Val loss (sd3 wip) (need help) #1856
Closed
Conversation
This file contains 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
val loss
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Outdated suggestions cannot be applied.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.
I want to move the functionality of the validation loss to a new structure so that it can be used, but there is an issue.
Traceback (most recent call last):
File "D:\SDXL\sd-scripts\sdxl_train_network.py", line 228, in
trainer.train(args)
File "D:\SDXL\sd-scripts\train_network.py", line 469, in train
train_dataset_group, val_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
File "D:\SDXL\sd-scripts\venv\lib\site-packages\torch\utils\data\dataset.py", line 350, in getitem
return self.datasets[dataset_idx][sample_idx]
File "D:\SDXL\sd-scripts\library\train_util.py", line 1638, in getitem
input_ids = [ids[0] for ids in self.tokenize_strategy.tokenize(caption)] # remove batch dimension
AttributeError: 'NoneType' object has no attribute 'tokenize'
I'm currently stuck and need assistance.
The plan is to first support it on the sd train network:
(1) Separate a validation dataset, create a validation loss recorder, and use an independent function to calculate the validation loss to eliminate the difference of hyperparameters.
(2) Allow free setting of the number of steps for validation execution and the number of validation set samples to be used.
(3) Use five time steps [10, 350, 500, 650, 990] for averaging, and using apply_debiased_estimation to adjust weight reduce timestep sampling bias .
I would like to know if Kohya has any plans to support validation loss in the future, as it is very useful during training.
If so, these points could serve as a starting point