Change wrecked ability to train subjects

[Enyakk]

One of the very recent changes after commit

cbe04acdf120aa4b5b1506e3120c421fb92df49a

wrecked the ability to train subjects in Dreambooth mode. I can A/B the repositories and there's a stark difference in likeness and quality. Possibly one of the CLIP changes?

[Claxiz]

I am having this issue too, except I couldnt even start training. Switching to an older version 3 commits ago solved my problems.

[geocine]

I am having this issue too, except I couldnt even start training. Switching to an older version 3 commits ago solved my problems.

I am having the same issue as @Enyakk . But I don't think the commit he referenced would break anything. Which hash did you revert to @Claxiz ?

[devilismyfriend]

I reverted from the latest big change, please check again

[devilismyfriend]

also make sure to post your configs

[Jordach]

Was regular image captions used alongside the class caption/prompt/token setting? This may hint at further issues.

[blackmagic24]

I reverted from the latest big change, please check again

It works again for me.

[geocine]

@devilismyfriend works again for me. However the Model Playground now is broken https://github.com/devilismyfriend/StableTuner/issues/72

Traceback (most recent call last):
  File "C:\Users\moo\miniconda3\envs\ST\lib\tkinter\__init__.py", line 1921, in __call__
    return self.func(*args)
  File "C:\Users\moo\miniconda3\envs\ST\lib\site-packages\customtkinter\windows\widgets\ctk_button.py", line 531, in _clicked
    self._command()
  File "E:\AIML\StableTuner\scripts\configuration_gui.py", line 1907, in <lambda>
    self.play_toolbox_label = ctk.CTkLabel(self.playground_frame_subframe, text="Toolbox")
  File "E:\AIML\StableTuner\scripts\configuration_gui.py", line 2271, in play_generate_image
    with torch.autocast("cuda"), torch.inference_mode():
AttributeError: play_current_image

And my config is here if you want to check

{
    "concepts": [
        {
            "instance_prompt": "zwx",
            "class_prompt": "person",
            "instance_data_dir": "E:/AIML/iloader/mooos_data",
            "class_data_dir": "E:/AIML/dreamstall/train/class_images/person",
            "flip_p": "",
            "do_not_balance": 0,
            "use_sub_dirs": 0
        }
    ],
    "sample_prompts": [],
    "add_controlled_seed_to_sample": [],
    "model_path": "runwayml/stable-diffusion-v1-5",
    "vae_path": "stabilityai/sd-vae-ft-mse",
    "output_path": "models/new_model",
    "send_telegram_updates": 0,
    "telegram_token": "",
    "telegram_chat_id": "",
    "resolution": "512",
    "batch_size": "1",
    "train_epocs": "5",
    "mixed_precision": "fp16",
    "use_8bit_adam": 1,
    "use_gradient_checkpointing": 1,
    "accumulation_steps": "1",
    "learning_rate": "1e-6",
    "warmup_steps": "0",
    "learning_rate_scheduler": "constant",
    "regenerate_latent_cache": 1,
    "train_text_encoder": 1,
    "with_prior_loss_preservation": 1,
    "prior_loss_preservation_weight": "1.0",
    "use_image_names_as_captions": 0,
    "auto_balance_concept_datasets": 0,
    "add_class_images_to_dataset": 0,
    "number_of_class_images": "200",
    "save_every_n_epochs": "5",
    "number_of_samples_to_generate": "1",
    "sample_height": "512",
    "sample_width": "512",
    "sample_random_aspect_ratio": 0,
    "sample_on_training_start": 1,
    "aspect_ratio_bucketing": 1,
    "seed": "3434554",
    "dataset_repeats": "50",
    "limit_text_encoder_training": "",
    "use_text_files_as_captions": 0,
    "ckpt_version": null,
    "convert_to_ckpt_after_training": 0,
    "execute_post_conversion": 0,
    "disable_cudnn_benchmark": 1,
    "sample_step_interval": "500",
    "conditional_dropout": "",
    "clip_penultimate": 0,
    "use_ema": 0,
    "aspect_ratio_bucketing_mode": "Dynamic Fill",
    "dynamic_bucketing_mode": "Duplicate",
    "model_variant": "Regular",
    "fallback_mask_prompt": "",
    "attention": "xformers",
    "batch_prompt_sampling": 0,
    "shuffle_dataset_per_epoch": 1
}