hcpdiff 0.9.0__py3-none-any.whl → 2.1__py3-none-any.whl

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/controlnet.yaml

@@ -1,53 +0,0 @@
-_base_:
-  - cfgs/train/train_base.yaml
-  - cfgs/plugins/plugin_controlnet.yaml # include controlnet plugin
-
-tokenizer_pt:
-  train: null
-
-train:
-  gradient_accumulation_steps: 1
-  save_step: 100
-
-  scheduler:
-    name: 'constant_with_warmup'
-    num_warmup_steps: 50
-    num_training_steps: 600
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  tokenizer_repeats: 1
-  ema_unet: 0
-  ema_text_encoder: 0
-
-data:
-  dataset1:
-    _target_: hcpdiff.data.TextImageCondPairDataset
-    _partial_: True # Not directly instantiate the object here. There are other parameters to be added in the runtime.
-    batch_size: 4
-    cache_latents: True
-    att_mask_encode: False
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        _target_: hcpdiff.data.source.Text2ImageCondSource
-        img_root: 'imgs/'
-        cond_root: 'cond_imgs/'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null # path to image captions (file_words)
-        att_mask: null
-        bg_color: [ 255, 255, 255 ] # RGB; for ARGB -> RGB
-
-        text_transforms:
-          _target_: torchvision.transforms.Compose
-          transforms:
-            - _target_: hcpdiff.utils.caption_tools.TagShuffle
-            - _target_: hcpdiff.utils.caption_tools.TagDropout
-              p: 0.1
-            - _target_: hcpdiff.utils.caption_tools.TemplateFill
-              word_names: { }
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 5

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/ema.yaml

@@ -1,10 +0,0 @@
-_base_:
-  - cfgs/train/examples/fine-tuning.yaml
-
-model:
-  ema:
-    _target_: hcpdiff.utils.ema.ModelEMA
-    _partial_: True
-    decay_max: 0.9997
-    power: 0.85
-

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/fine-tuning.yaml

@@ -1,53 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-unet:
-  -
-    lr: 1e-6
-    layers:
-      - '' # fine-tuning all layers in unet
-
-## fine-tuning text-encoder
-#text_encoder:
-#  - lr: 1e-6
-#    layers:
-#      - ''
-
-tokenizer_pt:
-  train: null
-
-train:
-  gradient_accumulation_steps: 1
-  save_step: 100
-
-  scheduler:
-    name: 'constant_with_warmup'
-    num_warmup_steps: 50
-    num_training_steps: 600
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  tokenizer_repeats: 1
-  ema_unet: 0
-  ema_text_encoder: 0
-
-# The dataset configuration inherits base_dataset.yaml
-# 数据集配置继承自base_dataset.yaml， 只需修改部分参数
-data:
-  dataset1:
-    batch_size: 4
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: 'imgs/'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null # path to image captions (file_words)
-        att_mask: null
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 5

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/locon.yaml

@@ -1,24 +0,0 @@
-_base_: [cfgs/train/examples/lora_conventional.yaml]
-
-lora_unet:
-  - # Linear
-    lr: 1e-4
-    rank: 8
-    layers:
-      - 're:.*\.attn.?$'
-      - 're:.*\.ff$'
-  - # Conv2d
-    lr: 1e-4
-    rank: 8
-    layers:
-      - 're:.*\.resnets$'
-      - 're:.*\.proj_in$'
-      - 're:.*\.proj_out$'
-      - 're:.*\.conv$'
-
-lora_text_encoder:
-  - lr: 1e-5
-    rank: 4
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/lora_anime_character.yaml

@@ -1,77 +0,0 @@
-_base_:
-  - cfgs/train/examples/lora_conventional.yaml
-
-model:
-  pretrained_model_name_or_path: 'deepghs/animefull-latest'
-  clip_skip: 1
-
-character_name: surtr_arknights
-dataset_dir: '/root/autodl-tmp/dataset/surtr_3'
-# if exp_dir is not set, a random time-based directory will be used
-# exp_dir: 'exps/surtr'
-
-unet_rank: 8
-text_encoder_rank: 4
-
-tokenizer_pt:
-  emb_dir: 'embs/' #自定义word目录
-  replace: False #训练后是否替换原有word
-  train: 
-    - name: ${character_name}
-      lr: 0.003
-
-lora_unet:
-  - lr: 1e-4
-    rank: ${unet_rank}
-    layers:
-      - 're:.*\.attn.?$'
-      - 're:.*\.ff$'
-
-lora_text_encoder:
-  - lr: 1e-5
-    rank: ${text_encoder_rank}
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-
-data:
-  dataset1:
-    batch_size: 4
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: ${dataset_dir}
-        prompt_template: 'prompt_tuning_template/object_caption.txt'
-        caption_file: ${dataset_dir}  # path to image captions (file_words)
-
-        word_names:
-          pt1: ${character_name}
-
-    # support images with any size, not recommended for anime training
-    # bucket:
-    #   _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-    #   target_area: ${times:512,512}
-    #   num_bucket: 5
-
-    # all images must have the same size, such as 512x704
-    bucket:
-      _target_: hcpdiff.data.bucket.SizeBucket.from_files # aspect ratio bucket
-      target_area: ---
-      num_bucket: 1
-
-logger:
-  - _target_: hcpdiff.loggers.CLILogger
-    _partial_: True
-    out_path: 'train.log'
-    log_step: 20
-  - _target_: hcpdiff.loggers.TBLogger
-    _partial_: True
-    out_path: 'tblog/'
-    log_step: 5
-#  - _target_: hcpdiff.loggers.WanDBLogger
-#    _partial_: True
-#    out_path: null
-#    log_step: 5
-
-

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/lora_conventional.yaml

@@ -1,56 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-lora_unet:
-  -
-    lr: 1e-4
-    rank: 8
-    layers:
-      - 're:.*\.attn.?$'
-      - 're:.*\.ff$'
-
-lora_text_encoder:
-  - lr: 1e-5
-    rank: 4
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-
-tokenizer_pt:
-  train: null
-
-train:
-  gradient_accumulation_steps: 1
-  save_step: 100
-
-  scheduler:
-    name: 'constant_with_warmup'
-    num_warmup_steps: 50
-    num_training_steps: 1000
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  tokenizer_repeats: 1
-  ema_unet: 0
-  ema_text_encoder: 0
-
-data:
-  dataset1:
-    batch_size: 4
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: 'imgs/'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null # path to image captions (file_words)
-
-        word_names:
-          pt1: pt-cat1
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 5

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/lora_sdxl.yaml

@@ -1,41 +0,0 @@
-_base_:
-  - cfgs/train/examples/lora_conventional.yaml
-
-lora_unet:
-  -
-    lr: 1e-4
-    rank: 8
-    layers:
-      - 're:.*\.attn.?$'
-      - 're:.*\.ff$'
-
-lora_text_encoder:
-  - lr: 1e-5
-    rank: 4
-    # for both CLIP
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-    # for CLIP1 (CLIP_B)
-    # layers:
-    #   - 're:clip_B.*self_attn$'
-    #   - 're:clip_B.*mlp$'
-    # for CLIP2 (CLIP_bigG)
-    # layers:
-    #   - 're:clip_bigG.*self_attn$'
-    #   - 're:clip_bigG.*mlp$'
-
-model:
-  pretrained_model_name_or_path: '/mnt/f/models/stable-diffusion-xl-base-1.0'
-  clip_skip: 1
-  clip_final_norm: False
-
-data:
-  dataset1:
-    _target_: hcpdiff.data.CropInfoPairDataset
-    batch_size: 4
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"768*768"}
-      num_bucket: 4

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/min_snr.yaml

@@ -1,7 +0,0 @@
-_base_: [cfgs/train/examples/fine-tuning.yaml]
-
-train:
-  loss:
-    criterion: # min SNR loss
-      _target_: hcpdiff.loss.MinSNRLoss
-      gamma: 2.0

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/preview_in_training.yaml

@@ -1,6 +0,0 @@
-_base_: [cfgs/train/train_base.yaml]
-
-previewer:
-  _target_: hcpdiff.loggers.preview.ImagePreviewer
-  _partial_: True
-  infer_cfg: cfgs/infer/text2img.yaml

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples_noob/DreamBooth.yaml

@@ -1,70 +0,0 @@
-_base_:
-  - cfgs/train/dataset/regularization_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-# 这部分根据需要改
-
-train:
-  save_step: 100
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  lr: 1e-6
-
-data_cfg:
-  batch_size: 4
-  img_root: 'imgs/' # 数据集路径
-  caption_file: 'imgs/' # 标注文件或文件夹
-  regularization_img_root: 'imgs/' # 正则化数据集
-  regularization_caption_file: 'imgs/' # 正则化数据集标注文件
-  target_area: 512*512 # 训练预期分辨率
-
-object_name: pt-cat1 # 训练物体名称
-class_name: cat # 训练类别名称
-
-
-
-# 这部分不用动
-
-unet:
-  -
-    lr: ${model.lr}
-    layers:
-      - ''
-
-data:
-  dataset1:
-    batch_size: ${data_cfg.batch_size}
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: ${data_cfg.img_root}
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: ${data_cfg.caption_file}
-
-        word_names:
-          pt1: ${object_name}
-          class: ${class_name}
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"${data_cfg.target_area}"}
-      num_bucket: 1
-
-  dataset_class:
-    batch_size: 1
-    cache_latents: True
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        img_root: ${data_cfg.regularization_img_root}
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: ${data_cfg.regularization_caption_file}
-
-        word_names:
-          class: ${class_name}
-    bucket:
-      _target_: hcpdiff.data.bucket.FixedBucket
-      target_size: [512, 512]

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples_noob/TextualInversion.yaml

@@ -1,45 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-# 这部分根据需要改
-
-train:
-  save_step: 100
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  lr: 1e-6
-  train_word_name: pt-catgirl1 # 训练单词名(需要提前创建单词)
-
-data_cfg:
-  batch_size: 4
-  img_root: 'imgs/' # 数据集路径
-  caption_file: 'imgs/' # 标注文件或文件夹
-  target_area: 512*512 # 训练预期分辨率
-
-
-
-
-
-# 这部分不用动
-
-data:
-  dataset1:
-    batch_size: ${data_cfg.batch_size}
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: ${data_cfg.img_root}
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: ${data_cfg.caption_file}
-
-        word_names:
-          pt1: ${model.train_word_name}
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"${data_cfg.target_area}"}
-      num_bucket: 3

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples_noob/fine-tuning.yaml

@@ -1,45 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-# 这部分根据需要改
-
-train:
-  save_step: 100
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  lr: 1e-6
-
-data_cfg:
-  batch_size: 4
-  img_root: 'imgs/' # 数据集路径
-  caption_file: 'imgs/' # 标注文件或文件夹
-  target_area: 512*512 # 训练预期分辨率
-
-
-
-
-# 这部分不用动
-unet:
-  -
-    lr: ${model.lr}
-    layers:
-      - '' # fine-tuning all layers in unet
-
-data:
-  dataset1:
-    batch_size: ${data_cfg.batch_size}
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: ${data_cfg.img_root}
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: ${data_cfg.caption_file}
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"${data_cfg.target_area}"}
-      num_bucket: 3

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples_noob/lora.yaml

@@ -1,63 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-
-# 这部分根据需要改
-
-train:
-  save_step: 100
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  lr_unet: 1e-4
-  lr_text_encoder: 1e-5
-  lora_rank: 8
-
-data_cfg:
-  batch_size: 4
-  img_root: 'imgs/' # 数据集路径
-  caption_file: 'imgs/' # 标注文件或文件夹
-  target_area: 512*512 # 训练预期分辨率
-
-object_name: pt-cat1 # lora训练的物体名(触发词)
-
-
-
-
-# 这部分不用动
-
-lora_unet:
-  -
-    lr: ${model.lr_unet}
-    rank: ${model.lora_rank}
-    layers:
-      - 're:.*\.attn.?$'
-      - 're:.*\.ff$'
-
-lora_text_encoder:
-  - lr: ${model.lr_text_encoder}
-    rank: ${model.lora_rank}
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-
-data:
-  dataset1:
-    batch_size: ${data_cfg.batch_size}
-    cache_latents: True
-
-    source:
-      data_source1:
-        img_root: ${data_cfg.img_root}
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: ${data_cfg.caption_file}
-
-        word_names:
-          pt1: ${object_name}
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"${data_cfg.target_area}"}
-      num_bucket: 3

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/train_base.yaml

@@ -1,81 +0,0 @@
-
-exp_dir: exps/${hcp.time:}
-mixed_precision: 'fp16'
-allow_tf32: False
-seed: 114514
-ckpt_type: 'safetensors' # [torch, safetensors]
-
-vis_info:
-  prompt: null
-  negative_prompt: ''
-
-train:
-  train_steps: 1000
-  train_epochs: null # Choose one of [train_steps, train_epochs]
-  gradient_accumulation_steps: 1
-  workers: 4
-  max_grad_norm: 1.0
-  set_grads_to_none: False
-  save_step: 100
-  cfg_scale: '1.0' # for DreamArtist
-
-  resume: null
-#  resume:
-#    ckpt_path:
-#      unet: []
-#      TE: []
-#      words: {}
-#    start_step: 0
-
-  loss:
-    criterion:
-      _target_: torch.nn.MSELoss
-      _partial_: True
-      reduction: 'none' # support for attention mask
-    type: 'eps' # 'eps' or 'sample'
-
-  optimizer:
-    _target_: torch.optim.AdamW
-    _partial_: True
-    weight_decay: 1e-3
-
-  optimizer_pt:
-    _target_: torch.optim.AdamW
-    _partial_: True
-    weight_decay: 5e-4
-
-  scale_lr: True # auto scale lr with total batch size
-  scheduler:
-    name: 'one_cycle'
-    num_warmup_steps: 200
-    num_training_steps: 1000
-    scheduler_kwargs: {} # args for scheduler
-
-  scale_lr_pt: True
-  scheduler_pt: ${.scheduler}
-
-logger:
-  -
-    _target_: hcpdiff.loggers.CLILogger
-    _partial_: True
-    out_path: 'train.log'
-    log_step: 20
-
-model:
-  revision: null
-  pretrained_model_name_or_path: null
-  tokenizer_repeats: 2
-  enable_xformers: True
-  gradient_checkpointing: True
-  force_cast_precision: False
-  ema: null
-  clip_skip: 0
-  clip_final_norm: True
-
-  tokenizer: null
-  noise_scheduler: null
-  unet: null
-  text_encoder: null
-  vae: null
-
-previewer: null

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/tuning_base.yaml

@@ -1,42 +0,0 @@
-# down部分总共6个 (Resnet+Transformer) 块，分别对应于:
-#   1:
-#     down_blocks.0.resnets.0
-#     down_blocks.0.attentions.0.transformer_blocks.0
-#   2:
-#     down_blocks.0.resnets.1
-#     down_blocks.0.attentions.1.transformer_blocks.0
-#   3:
-#     down_blocks.1.resnets.0
-#     down_blocks.1.attentions.0.transformer_blocks.0
-#   ......
-
-# up部分总共9个 (Resnet+Transformer) 块，分别对应于:
-#   1:
-#     up_blocks.1.resnets.0
-#     up_blocks.1.attentions.0.transformer_blocks.0
-#   2:
-#     up_blocks.1.resnets.1
-#     up_blocks.1.attentions.1.transformer_blocks.0
-#   3:
-#     up_blocks.1.resnets.2
-#     up_blocks.1.attentions.2.transformer_blocks.0
-#   4:
-#     up_blocks.2.resnets.0
-#     up_blocks.2.attentions.0.transformer_blocks.0
-#   ......
-
-tokenizer_pt:
-  emb_dir: 'embs/'
-  replace: False
-  train: []
-  #train:
-  #  - {name: pt1, lr: 1e-3}
-
-unet: null
-lora_unet: null
-
-text_encoder: null
-lora_text_encoder: null
-
-plugin_unet: null
-plugin_TE: null