• [sd_scripts]之config

    https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-ja.mdicon-default.png?t=N7T8https://github.com/kohya-ss/sd-scripts/blob/main/docs/config_README-ja.md[Stable Diffusion]训练你的LoRA(Linux) - 知乎简介LoRA 是一种参数高效微调方法（PEFT），最早由 LoRA: Low-Rank Adaptation of Large Language Models 提出并应用于微调语言大模型之中，后来由 Low-rank Adaptation for Fast Text-to-Image Diffusion Fine-tu…icon-default.png?t=N7T8https://zhuanlan.zhihu.com/p/640144661配置文件格式TOML：

    1. [general]
    2. shuffle_caption = true
    3. caption_extension = '.txt'
    4. keep_tokens = 1
    5.  
    6. # This is a DreamBooth-style dataset
    7. [[datasets]]
    8. resolution = 512
    9. batch_size = 4
    10. keep_tokens = 2
    11.  
    12.   [[datasets.subsets]]
    13.   image_dir = 'C:\hoge'
    14.   class_tokens = 'hoge girl'
    15.   # This subset has keep_tokens = 2 (using the value of the parent datasets)
    16.  
    17.   [[datasets.subsets]]
    18.   image_dir = 'C:\fuga'
    19.   class_tokens = 'fuga boy'
    20.   keep_tokens = 3
    21.  
    22.   [[datasets.subsets]]
    23.   is_reg = true
    24.   image_dir = 'C:\reg'
    25.   class_tokens = 'human'
    26.   keep_tokens = 1
    27.  
    28. # This is a fine-tuning-style dataset
    29. [[datasets]]
    30. resolution = [768, 768]
    31. batch_size = 2
    32.  
    33.   [[datasets.subsets]]
    34.   image_dir = 'C:\piyo'
    35.   metadata_file = 'C:\piyo\piyo_md.json'
    36.   # This subset has keep_tokens = 1 (using the general value)

    在此示例中，将训练三个目录作为512x512（批量大小4）的dreambooth数据集，以及一个目录作为768x768（批量大小2）的微调数据集。

    1. C:\
    2. ├─ hoge  ->  [[datasets.subsets]] No.1  ┐                        ┐
    3. ├─ fuga  ->  [[datasets.subsets]] No.2  |->  [[datasets]] No.1   |->  [general]
    4. ├─ reg   ->  [[datasets.subsets]] No.3  ┘                        |
    5. └─ piyo  ->  [[datasets.subsets]] No.4  -->  [[datasets]] No.2   ┘

     所有方法均可使用的参数：[general]

    dreambooth-style 特有的参数：

    fine-tuning-style特有的参数：

    train_db.py参数

    1. [-h] [--v2] [--v_parameterization] 
    2. [--pretrained_model_name_or_path PRETRAINED_MODEL_NAME_OR_PATH] 
    3. [--tokenizer_cache_dir TOKENIZER_CACHE_DIR]
    4. [--train_data_dir TRAIN_DATA_DIR]
    5. [--shuffle_caption]
    6. [--caption_extension CAPTION_EXTENSION]
    7. [--caption_extention CAPTION_EXTENTION] 
    8. [--keep_tokens KEEP_TOKENS]
    9. [--caption_prefix CAPTION_PREFIX]
    10. [--caption_suffix CAPTION_SUFFIX] 
    11. [--color_aug] 
    12. [--flip_aug]
    13. [--face_crop_aug_range FACE_CROP_AUG_RANGE]
    14. [--random_crop] 
    15. [--debug_dataset]
    16. [--resolution RESOLUTION]
    17. [--cache_latents] 
    18. [--vae_batch_size VAE_BATCH_SIZE]
    19. [--cache_latents_to_disk] 
    20. [--enable_bucket]
    21. [--min_bucket_reso MIN_BUCKET_RESO]
    22. [--max_bucket_reso MAX_BUCKET_RESO]
    23. [--bucket_reso_steps BUCKET_RESO_STEPS]
    24. [--bucket_no_upscale] 
    25. [--token_warmup_min TOKEN_WARMUP_MIN]
    26. [--token_warmup_step TOKEN_WARMUP_STEP]
    27. [--dataset_class DATASET_CLASS] 
    28. [--caption_dropout_rate CAPTION_DROPOUT_RATE] 
    29. [--caption_dropout_every_n_epochs CAPTION_DROPOUT_EVERY_N_EPOCHS]
    30. [--caption_tag_dropout_rate CAPTION_TAG_DROPOUT_RATE] 
    31. [--reg_data_dir REG_DATA_DIR] 
    32. [--output_dir OUTPUT_DIR] 
    33. [--output_name OUTPUT_NAME]
    34. [--huggingface_repo_id HUGGINGFACE_REPO_ID] 
    35. [--huggingface_repo_type HUGGINGFACE_REPO_TYPE]
    36. [--huggingface_path_in_repo HUGGINGFACE_PATH_IN_REPO]
    37. [--huggingface_token HUGGINGFACE_TOKEN]
    38. [--huggingface_repo_visibility HUGGINGFACE_REPO_VISIBILITY] 
    39. [--save_state_to_huggingface]
    40. [--resume_from_huggingface]
    41. [--async_upload]
    42. [--save_precision {None,float,fp16,bf16}] 
    43. [--save_every_n_epochs SAVE_EVERY_N_EPOCHS]
    44. [--save_every_n_steps SAVE_EVERY_N_STEPS]
    45. [--save_n_epoch_ratio SAVE_N_EPOCH_RATIO]
    46. [--save_last_n_epochs SAVE_LAST_N_EPOCHS] 
    47. [--save_last_n_epochs_state SAVE_LAST_N_EPOCHS_STATE] 
    48. [--save_last_n_steps SAVE_LAST_N_STEPS]
    49. [--save_last_n_steps_state SAVE_LAST_N_STEPS_STATE]
    50. [--save_state]
    51. [--resume RESUME] 
    52. [--train_batch_size TRAIN_BATCH_SIZE] 
    53. [--max_token_length {None,150,225}] 
    54. [--mem_eff_attn]
    55. [--xformers] 
    56. [--sdpa]
    57. [--vae VAE] 
    58. [--max_train_steps MAX_TRAIN_STEPS] 
    59. [--max_train_epochs MAX_TRAIN_EPOCHS] 
    60. [--max_data_loader_n_workers MAX_DATA_LOADER_N_WORKERS]
    61. [--persistent_data_loader_workers]
    62. [--seed SEED]
    63. [--gradient_checkpointing]
    64. [--gradient_accumulation_steps GRADIENT_ACCUMULATION_STEPS] 
    65. [--mixed_precision {no,fp16,bf16}]
    66. [--full_fp16] 
    67. [--full_bf16] 
    68. [--ddp_timeout DDP_TIMEOUT]
    69. [--clip_skip CLIP_SKIP]
    70. [--logging_dir LOGGING_DIR]
    71. [--log_with {tensorboard,wandb,all}]
    72. [--log_prefix LOG_PREFIX]
    73. [--log_tracker_name LOG_TRACKER_NAME] 
    74. [--log_tracker_config LOG_TRACKER_CONFIG] 
    75. [--wandb_api_key WANDB_API_KEY] 
    76. [--noise_offset NOISE_OFFSET]
    77. [--multires_noise_iterations MULTIRES_NOISE_ITERATIONS]
    78. [--ip_noise_gamma IP_NOISE_GAMMA] 
    79. [--multires_noise_discount MULTIRES_NOISE_DISCOUNT]
    80. [--adaptive_noise_scale ADAPTIVE_NOISE_SCALE] 
    81. [--zero_terminal_snr] 
    82. [--min_timestep MIN_TIMESTEP] 
    83. [--max_timestep MAX_TIMESTEP]
    84. [--lowram]
    85. [--sample_every_n_steps SAMPLE_EVERY_N_STEPS] 
    86. [--sample_every_n_epochs SAMPLE_EVERY_N_EPOCHS] 
    87. [--sample_prompts SAMPLE_PROMPTS]
    88. [--sample_sampler {ddim,pndm,lms,euler,euler_a,heun,dpm_2,dpm_2_a,dpmsolver,dpmsolver++,dpmsingle,k_lms,k_euler,k_euler_a,k_dpm_2,k_dpm_2_a}]
    89. [--config_file CONFIG_FILE]
    90. [--output_config] 
    91. [--metadata_title METADATA_TITLE] 
    92. [--metadata_author METADATA_AUTHOR]
    93. [--metadata_description METADATA_DESCRIPTION] 
    94. [--metadata_license METADATA_LICENSE]
    95. [--metadata_tags METADATA_TAGS]
    96. [--prior_loss_weight PRIOR_LOSS_WEIGHT] 
    97. [--save_model_as {None,ckpt,safetensors,diffusers,diffusers_safetensors}]
    98. [--use_safetensors]
    99. [--optimizer_type OPTIMIZER_TYPE]
    100. [--use_8bit_adam] 
    101. [--use_lion_optimizer] 
    102. [--learning_rate LEARNING_RATE]
    103. [--max_grad_norm MAX_GRAD_NORM]
    104. [--optimizer_args [OPTIMIZER_ARGS [OPTIMIZER_ARGS ...]]] 
    105. [--lr_scheduler_type LR_SCHEDULER_TYPE] 
    106. [--lr_scheduler_args [LR_SCHEDULER_ARGS [LR_SCHEDULER_ARGS ...]]]
    107. [--lr_scheduler LR_SCHEDULER]
    108. [--lr_warmup_steps LR_WARMUP_STEPS] 
    109. [--lr_scheduler_num_cycles LR_SCHEDULER_NUM_CYCLES] 
    110. [--lr_scheduler_power LR_SCHEDULER_POWER]
    111. [--dataset_config DATASET_CONFIG]
    112. [--min_snr_gamma MIN_SNR_GAMMA] 
    113. [--scale_v_pred_loss_like_noise_pred]
    114. [--v_pred_like_loss V_PRED_LIKE_LOSS] 
    115. [--debiased_estimation_loss]
    116. [--weighted_captions] 
    117. [--learning_rate_te LEARNING_RATE_TE]
    118. [--no_token_padding] 
    119. [--stop_text_encoder_training STOP_TEXT_ENCODER_TRAINING]

     train_network.py

    1. [-h] [--v2] [--v_parameterization] 
    2. [--pretrained_model_name_or_path PRETRAINED_MODEL_NAME_OR_PATH] 
    3. [--tokenizer_cache_dir TOKENIZER_CACHE_DIR]
    4. [--train_data_dir TRAIN_DATA_DIR] 
    5. [--shuffle_caption] 
    6. [--caption_extension CAPTION_EXTENSION]
    7. [--caption_extention CAPTION_EXTENTION]
    8. [--keep_tokens KEEP_TOKENS] 
    9. [--caption_prefix CAPTION_PREFIX] 
    10. [--caption_suffix CAPTION_SUFFIX]
    11. [--color_aug] 
    12. [--flip_aug]
    13. [--face_crop_aug_range FACE_CROP_AUG_RANGE] 
    14. [--random_crop] 
    15. [--debug_dataset] 
    16. [--resolution RESOLUTION] 
    17. [--cache_latents] 
    18. [--vae_batch_size VAE_BATCH_SIZE]
    19. [--cache_latents_to_disk] 
    20. [--enable_bucket] 
    21. [--min_bucket_reso MIN_BUCKET_RESO] 
    22. [--max_bucket_reso MAX_BUCKET_RESO] 
    23. [--bucket_reso_steps BUCKET_RESO_STEPS]
    24. [--bucket_no_upscale] 
    25. [--token_warmup_min TOKEN_WARMUP_MIN] 
    26. [--token_warmup_step TOKEN_WARMUP_STEP] 
    27. [--dataset_class DATASET_CLASS]
    28. [--caption_dropout_rate CAPTION_DROPOUT_RATE]
    29. [--caption_dropout_every_n_epochs CAPTION_DROPOUT_EVERY_N_EPOCHS]
    30. [--caption_tag_dropout_rate CAPTION_TAG_DROPOUT_RATE] 
    31. [--reg_data_dir REG_DATA_DIR] 
    32. [--in_json IN_JSON] 
    33. [--dataset_repeats DATASET_REPEATS]
    34. [--output_dir OUTPUT_DIR] 
    35. [--output_name OUTPUT_NAME] 
    36. [--huggingface_repo_id HUGGINGFACE_REPO_ID] 
    37. [--huggingface_repo_type HUGGINGFACE_REPO_TYPE]
    38. [--huggingface_path_in_repo HUGGINGFACE_PATH_IN_REPO] 
    39. [--huggingface_token HUGGINGFACE_TOKEN] 
    40. [--huggingface_repo_visibility HUGGINGFACE_REPO_VISIBILITY]
    41. [--save_state_to_huggingface] 
    42. [--resume_from_huggingface] 
    43. [--async_upload] 
    44. [--save_precision {None,float,fp16,bf16}]
    45. [--save_every_n_epochs SAVE_EVERY_N_EPOCHS] 
    46. [--save_every_n_steps SAVE_EVERY_N_STEPS] 
    47. [--save_n_epoch_ratio SAVE_N_EPOCH_RATIO]
    48. [--save_last_n_epochs SAVE_LAST_N_EPOCHS] 
    49. [--save_last_n_epochs_state SAVE_LAST_N_EPOCHS_STATE] 
    50. [--save_last_n_steps SAVE_LAST_N_STEPS]
    51. [--save_last_n_steps_state SAVE_LAST_N_STEPS_STATE]
    52. [--save_state] 
    53. [--resume RESUME] 
    54. [--train_batch_size TRAIN_BATCH_SIZE] 
    55. [--max_token_length {None,150,225}]
    56. [--mem_eff_attn] 
    57. [--xformers] 
    58. [--sdpa] 
    59. [--vae VAE] 
    60. [--max_train_steps MAX_TRAIN_STEPS] 
    61. [--max_train_epochs MAX_TRAIN_EPOCHS]
    62. [--max_data_loader_n_workers MAX_DATA_LOADER_N_WORKERS] 
    63. [--persistent_data_loader_workers] 
    64. [--seed SEED] 
    65. [--gradient_checkpointing]
    66. [--gradient_accumulation_steps GRADIENT_ACCUMULATION_STEPS] 
    67. [--mixed_precision {no,fp16,bf16}] 
    68. [--full_fp16] 
    69. [--full_bf16] 
    70. [--ddp_timeout DDP_TIMEOUT]
    71. [--clip_skip CLIP_SKIP] 
    72. [--logging_dir LOGGING_DIR] 
    73. [--log_with {tensorboard,wandb,all}] 
    74. [--log_prefix LOG_PREFIX] 
    75. [--log_tracker_name LOG_TRACKER_NAME]
    76. [--log_tracker_config LOG_TRACKER_CONFIG] 
    77. [--wandb_api_key WANDB_API_KEY] 
    78. [--noise_offset NOISE_OFFSET]
    79. [--multires_noise_iterations MULTIRES_NOISE_ITERATIONS]
    80. [--ip_noise_gamma IP_NOISE_GAMMA] 
    81. [--multires_noise_discount MULTIRES_NOISE_DISCOUNT] 
    82. [--adaptive_noise_scale ADAPTIVE_NOISE_SCALE] 
    83. [--zero_terminal_snr]
    84. [--min_timestep MIN_TIMESTEP] 
    85. [--max_timestep MAX_TIMESTEP]
    86. [--lowram] 
    87. [--sample_every_n_steps SAMPLE_EVERY_N_STEPS]
    88. [--sample_every_n_epochs SAMPLE_EVERY_N_EPOCHS] 
    89. [--sample_prompts SAMPLE_PROMPTS]
    90. [--sample_sampler {ddim,pndm,lms,euler,euler_a,heun,dpm_2,dpm_2_a,dpmsolver,dpmsolver++,dpmsingle,k_lms,k_euler,k_euler_a,k_dpm_2,k_dpm_2_a}]
    91. [--config_file CONFIG_FILE] 
    92. [--output_config] 
    93. [--metadata_title METADATA_TITLE] 
    94. [--metadata_author METADATA_AUTHOR]
    95. [--metadata_description METADATA_DESCRIPTION] 
    96. [--metadata_license METADATA_LICENSE] 
    97. [--metadata_tags METADATA_TAGS] 
    98. [--prior_loss_weight PRIOR_LOSS_WEIGHT]
    99. [--optimizer_type OPTIMIZER_TYPE] 
    100. [--use_8bit_adam] 
    101. [--use_lion_optimizer] 
    102. [--learning_rate LEARNING_RATE] 
    103. [--max_grad_norm MAX_GRAD_NORM]
    104. [--optimizer_args [OPTIMIZER_ARGS ...]] 
    105. [--lr_scheduler_type LR_SCHEDULER_TYPE] 
    106. [--lr_scheduler_args [LR_SCHEDULER_ARGS ...]] 
    107. [--lr_scheduler LR_SCHEDULER]
    108. [--lr_warmup_steps LR_WARMUP_STEPS] 
    109. [--lr_scheduler_num_cycles LR_SCHEDULER_NUM_CYCLES] 
    110. [--lr_scheduler_power LR_SCHEDULER_POWER]
    111. [--dataset_config DATASET_CONFIG] 
    112. [--min_snr_gamma MIN_SNR_GAMMA] 
    113. [--scale_v_pred_loss_like_noise_pred] 
    114. [--v_pred_like_loss V_PRED_LIKE_LOSS]
    115. [--debiased_estimation_loss] 
    116. [--weighted_captions] 
    117. [--no_metadata] 
    118. [--save_model_as {None,ckpt,pt,safetensors}] 
    119. [--unet_lr UNET_LR]
    120. [--text_encoder_lr TEXT_ENCODER_LR] 
    121. [--network_weights NETWORK_WEIGHTS] 
    122. [--network_module NETWORK_MODULE] 
    123. [--network_dim NETWORK_DIM]
    124. [--network_alpha NETWORK_ALPHA] 
    125. [--network_dropout NETWORK_DROPOUT] 
    126. [--network_args [NETWORK_ARGS ...]] 
    127. [--network_train_unet_only]
    128. [--network_train_text_encoder_only] 
    129. [--training_comment TRAINING_COMMENT] 
    130. [--dim_from_weights] 
    131. [--scale_weight_norms SCALE_WEIGHT_NORMS]
    132. [--base_weights [BASE_WEIGHTS ...]] 
    133. [--base_weights_multiplier [BASE_WEIGHTS_MULTIPLIER ...]] 
    134. [--no_half_vae]
    135.  
    136.  
    137.

  • 相关阅读:
    CoDeSys系列-4、基于Ubuntu的codesys运行时扩展包搭建Profinet主从环境
    力扣-463.岛屿的周长
    杭州ALIENWARE外星人电脑（大悦城旗舰店），玩起来就是不一样
    直播带货平台有哪些
    CSS文字文本样式总结
    chatGPT培训老师AIGC培训讲师叶梓:大模型这么火，我们在使用时应该关注些什么？-6
    MyBatis 源码系列：MyBatis 解析配置文件、二级缓存、SQL
    CSS精灵图和字体图标的使用
    IP-guard WebServer 远程命令执行漏洞
    Harmony OS学习2
  • 原文地址：https://blog.csdn.net/u012193416/article/details/134317526