Skip to content

Commit

Permalink
Publish v15
Browse files Browse the repository at this point in the history
  • Loading branch information
@bmaltais
bmaltais committed Dec 5, 2022
1 parent 30b4be5 commit e8db30b
Show file tree
Hide file tree
Showing 17 changed files with 513 additions and 2,747 deletions.
75 changes: 42 additions & 33 deletions README.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -21,29 +21,20 @@ Give unrestricted script access to powershell so venv can work:
Open a regular Powershell terminal and type the following inside:

```powershell
# Clone the Kohya_ss repository
git clone https://github.com/bmaltais/kohya_ss.git
# Navigate to the newly cloned directory
cd kohya_ss
# Create a virtual environment using the system-site-packages option
python -m venv --system-site-packages venv
# Activate the virtual environment
.\venv\Scripts\activate
# Install the required packages
pip install torch==1.12.1+cu116 torchvision==0.13.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116
pip install --upgrade -r requirements.txt
pip install -U -I --no-deps https://github.com/C43H66N12O12S2/stable-diffusion-webui/releases/download/f/xformers-0.0.14.dev0-cp310-cp310-win_amd64.whl
# Copy the necessary files to the virtual environment's site-packages directory
cp .\bitsandbytes_windows\*.dll .\venv\Lib\site-packages\bitsandbytes\
cp .\bitsandbytes_windows\cextension.py .\venv\Lib\site-packages\bitsandbytes\cextension.py
cp .\bitsandbytes_windows\main.py .\venv\Lib\site-packages\bitsandbytes\cuda_setup\main.py
# Configure the accelerate utility
accelerate config
```
Expand Down Expand Up @@ -285,20 +276,22 @@ Refer to this url for more details about finetuning: https://note.com/kohya_ss/n
## Options list

```txt
usage: train_db_fixed.py [-h] [--v2] [--v_parameterization] [--pretrained_model_name_or_path PRETRAINED_MODEL_NAME_OR_PATH]
[--fine_tuning] [--shuffle_caption] [--caption_extention CAPTION_EXTENTION]
usage: train_db_fixed.py [-h] [--v2] [--v_parameterization]
[--pretrained_model_name_or_path PRETRAINED_MODEL_NAME_OR_PATH] [--fine_tuning]
[--shuffle_caption] [--caption_extention CAPTION_EXTENTION]
[--caption_extension CAPTION_EXTENSION] [--train_data_dir TRAIN_DATA_DIR]
[--reg_data_dir REG_DATA_DIR] [--dataset_repeats DATASET_REPEATS] [--output_dir OUTPUT_DIR]
[--save_every_n_epochs SAVE_EVERY_N_EPOCHS] [--save_state] [--resume RESUME]
[--reg_data_dir REG_DATA_DIR] [--dataset_repeats DATASET_REPEATS] [--output_dir OUTPUT_DIR]
[--use_safetensors] [--save_every_n_epochs SAVE_EVERY_N_EPOCHS] [--save_state] [--resume RESUME]
[--prior_loss_weight PRIOR_LOSS_WEIGHT] [--no_token_padding]
[--stop_text_encoder_training STOP_TEXT_ENCODER_TRAINING] [--color_aug] [--flip_aug]
[--face_crop_aug_range FACE_CROP_AUG_RANGE] [--random_crop] [--debug_dataset]
[--resolution RESOLUTION] [--train_batch_size TRAIN_BATCH_SIZE] [--use_8bit_adam] [--mem_eff_attn]
[--xformers] [--cache_latents] [--enable_bucket] [--min_bucket_reso MIN_BUCKET_RESO]
[--max_bucket_reso MAX_BUCKET_RESO] [--learning_rate LEARNING_RATE]
[--max_train_steps MAX_TRAIN_STEPS] [--seed SEED] [--gradient_checkpointing]
[--mixed_precision {no,fp16,bf16}] [--save_precision {None,float,fp16,bf16}] [--clip_skip CLIP_SKIP]
[--logging_dir LOGGING_DIR] [--lr_scheduler LR_SCHEDULER] [--lr_warmup_steps LR_WARMUP_STEPS]
[--resolution RESOLUTION] [--train_batch_size TRAIN_BATCH_SIZE] [--use_8bit_adam]
[--mem_eff_attn] [--xformers] [--vae VAE] [--cache_latents] [--enable_bucket]
[--min_bucket_reso MIN_BUCKET_RESO] [--max_bucket_reso MAX_BUCKET_RESO]
[--learning_rate LEARNING_RATE] [--max_train_steps MAX_TRAIN_STEPS] [--seed SEED]
[--gradient_checkpointing] [--mixed_precision {no,fp16,bf16}]
[--save_precision {None,float,fp16,bf16}] [--clip_skip CLIP_SKIP] [--logging_dir LOGGING_DIR]
[--log_prefix LOG_PREFIX] [--lr_scheduler LR_SCHEDULER] [--lr_warmup_steps LR_WARMUP_STEPS]
options:
-h, --help show this help message and exit
Expand All @@ -310,7 +303,7 @@ options:
--fine_tuning fine tune the model instead of DreamBooth / DreamBoothではなくfine tuningする
--shuffle_caption shuffle comma-separated caption / コンマで区切られたcaptionの各要素をshuffleする
--caption_extention CAPTION_EXTENTION
extension of caption files (backward compatiblity) / 読み込むcaptionファイルの拡張子（スペルミスを残し てあります）
extension of caption files (backward compatiblity) / 読み込むcaptionファイルの拡張子（スペルミスを 残してあります）
--caption_extension CAPTION_EXTENSION
extension of caption files / 読み込むcaptionファイルの拡張子
--train_data_dir TRAIN_DATA_DIR
Expand All @@ -320,15 +313,18 @@ options:
--dataset_repeats DATASET_REPEATS
repeat dataset in fine tuning / fine tuning時にデータセットを繰り返す回数
--output_dir OUTPUT_DIR
directory to output trained model (default format is same to input) /
学習後のモデル出力先ディレクトリ（デフォルトの保存形式は読み込んだ形式と同じ）
directory to output trained model / 学習後のモデル出力先ディレクトリ
--use_safetensors use safetensors format for StableDiffusion checkpoint /
StableDiffusionのcheckpointをsafetensors形式で保存する
--save_every_n_epochs SAVE_EVERY_N_EPOCHS
save checkpoint every N epochs / 学習中のモデルを指定エポックごとに保存します
--save_state save training state additionally (including optimizer states etc.) / optimizerなど学習状態も含めたstateを追加で保存する
--save_state save training state additionally (including optimizer states etc.) /
optimizerなど学習状態も含めたstateを追加で保存する
--resume RESUME saved state to resume training / 学習再開するモデルのstate
--prior_loss_weight PRIOR_LOSS_WEIGHT
loss weight for regularization images / 正則化画像のlossの重み
--no_token_padding disable token padding (same as Diffuser's DreamBooth) / トークンのpaddingを無効にする（Diffusers版DreamBoothと同じ動作）
--no_token_padding disable token padding (same as Diffuser's DreamBooth) /
トークンのpaddingを無効にする（Diffusers版DreamBoothと同じ動作）
--stop_text_encoder_training STOP_TEXT_ENCODER_TRAINING
steps to stop text encoder training / Text Encoderの学習を止めるステップ数
--color_aug enable weak color augmentation / 学習時に色合いのaugmentationを有効にする
Expand All @@ -337,16 +333,17 @@ options:
enable face-centered crop augmentation and its range (e.g. 2.0,4.0) /
学習時に顔を中心とした切り出しaugmentationを有効にするときは倍率を指定する（例：2.0,4.0）
--random_crop enable random crop (for style training in face-centered crop augmentation) /
ランダムな切り出しを有効にする（顔を中心としたaugmentationを行うときに画風の学習用に指定する）
--debug_dataset show images for debugging (do not train) / デバッグ用に学習データを画面表示する（学習は行わない）
ランダムな切り出しを有効にする（顔を中心としたaugmentationを行うときに画風の学習用に指定する）
--debug_dataset show images for debugging (do not train) / デバッグ用に学習データを画面表示する（学習は行わない）
--resolution RESOLUTION
resolution in training ('size' or 'width,height') / 学習時の画像解像度（'サイズ'指定、または'幅,高さ'指定）
resolution in training ('size' or 'width,height') / 学習時の画像解像度（'サイズ'指定、または'幅,高 さ'指定）
--train_batch_size TRAIN_BATCH_SIZE
batch size for training (1 means one train or reg data, not train/reg pair) /
学習時のバッチサイズ（1でtrain/regをそれぞれ1件ずつ学習）
--use_8bit_adam use 8bit Adam optimizer (requires bitsandbytes) / 8bit Adamオプティマイザを使う（bitsandbytesのインストールが必要）
--mem_eff_attn use memory efficient attention for CrossAttention / CrossAttentionに省メモリ版attentionを使う
--mem_eff_attn use memory efficient attention for CrossAttention / CrossAttentionに省メモリ版attentionを使う
--xformers use xformers for CrossAttention / CrossAttentionにxformersを使う
--vae VAE path to checkpoint of vae to replace / VAEを入れ替える場合、VAEのcheckpointファイルまたはディレクトリ
--cache_latents cache latents to reduce memory (augmentations must be disabled) /
メモリ削減のためにlatentをcacheする（augmentationは使用不可）
--enable_bucket enable buckets for multi aspect ratio training / 複数解像度学習のためのbucketを有効にする
Expand All @@ -365,17 +362,29 @@ options:
use mixed precision / 混合精度を使う場合、その精度
--save_precision {None,float,fp16,bf16}
precision in saving (available in StableDiffusion checkpoint) /
保存時に精度を変更して保存する（StableDiffusion形式での保存時のみ有効）
--clip_skip CLIP_SKIP
use output of nth layer from back of text encoder (n>=1) / text encoderの後ろからn番目の層の出力を 用いる（nは1以上）
--logging_dir LOGGING_DIR
enable logging and output TensorBoard log to this directory / ログ出力を有効にしてこのディレクトリにTensorBoard用のログを出力する
enable logging and output TensorBoard log to this directory /
ログ出力を有効にしてこのディレクトリにTensorBoard用のログを出力する
--log_prefix LOG_PREFIX
add prefix for each log directory / ログディレクトリ名の先頭に追加する文字列
--lr_scheduler LR_SCHEDULER
scheduler to use for learning rate / 学習率のスケジューラ: linear, cosine, cosine_with_restarts, polynomial,
constant (default), constant_with_warmup
scheduler to use for learning rate / 学習率のスケジューラ: linear, cosine, cosine_with_restarts,
polynomial, constant (default), constant_with_warmup
--lr_warmup_steps LR_WARMUP_STEPS
Number of steps for the warmup in the lr scheduler (default is 0) / 学習率のスケジューラをウォームアップするステップ数（デフォルト0）
Number of steps for the warmup in the lr scheduler (default is 0) /
学習率のスケジューラをウォームアップするステップ数（デフォルト0）
```

## Change history

* 12/05 (v15) update:
- The script has been divided into two parts
- Support for SafeTensors format has been added. Install SafeTensors with `pip install safetensors`. The script will automatically detect the format based on the file extension when loading. Use the `--use_safetensors` option if you want to save the model as safetensor.
- The vae option has been added to load a VAE model separately.
- The log_prefix option has been added to allow adding a custom string to the log directory name before the date and time.
* 11/30 (v13) update:
- fix training text encoder at specified step (`--stop_text_encoder_training=<step #>`) that was causing both Unet and text encoder training to stop completely at the specified step rather than continue without text encoding training.
* 11/29 (v12) update:
Expand Down Expand Up @@ -405,4 +414,4 @@ options:
- The data format of checkpoint at the time of saving can be specified with the --save_precision option. You can choose float, fp16, and bf16.
- Added a --save_state option to save the learning state (optimizer, etc.) in the middle. It can be resumed with the --resume option.
* 11/9 (v8): supports Diffusers 0.7.2. To upgrade diffusers run `pip install --upgrade diffusers[torch]`
* 11/7 (v7): Text Encoder supports checkpoint files in different storage formats (it is converted at the time of import, so export will be in normal format). Changed the average value of EPOCH loss to output to the screen. Added a function to save epoch and global step in checkpoint in SD format (add values if there is existing data). The reg_data_dir option is enabled during fine tuning (fine tuning while mixing regularized images). Added dataset_repeats option that is valid for fine tuning (specified when the number of teacher images is small and the epoch is extremely short).
* 11/7 (v7): Text Encoder supports checkpoint files in different storage formats (it is converted at the time of import, so export will be in normal format). Changed the average value of EPOCH loss to output to the screen. Added a function to save epoch and global step in checkpoint in SD format (add values if there is existing data). The reg_data_dir option is enabled during fine tuning (fine tuning while mixing regularized images). Added dataset_repeats option that is valid for fine tuning (specified when the number of teacher images is small and the epoch is extremely short).
9 changes: 1 addition & 8 deletions diffusers_fine_tuning/README.md
View file
Original file line number Diff line number Diff line change
@@ -1,10 +1,3 @@
# Diffusers Fine Tuning

This subfolder provide all the required tools to run the diffusers fine tuning version found in this note: https://note.com/kohya_ss/n/nbf7ce8d80f29

## Releases

11/23 (v3):
- Added WD14Tagger tagging script.
- A log output function has been added to the fine_tune.py. Also, fixed the double shuffling of data.
- Fixed misspelling of options for each script (caption_extention→caption_extension will work for the time being, even if it remains outdated).
Code has been moved to dedicated repo at: https://github.com/bmaltais/kohya_diffusers_fine_tuning
125 changes: 0 additions & 125 deletions diffusers_fine_tuning/clean_captions_and_tags.py
View file

This file was deleted.

Loading

0 comments on commit e8db30b

Please sign in to comment.