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

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/te_struct.txt

@@ -1,193 +0,0 @@
-CLIPTextModel(
-  (text_model): CLIPTextTransformer(
-    (embeddings): CLIPTextEmbeddings(
-      (token_embedding): Embedding(49408, 768)
-      (position_embedding): Embedding(77, 768)
-    )
-    (encoder): CLIPEncoder(
-      (layers): ModuleList(
-        (0): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (1): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (2): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (3): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (4): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (5): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (6): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (7): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (8): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (9): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (10): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-        (11): CLIPEncoderLayer(
-          (self_attn): CLIPAttention(
-            (k_proj): Linear(in_features=768, out_features=768, bias=True)
-            (v_proj): Linear(in_features=768, out_features=768, bias=True)
-            (q_proj): Linear(in_features=768, out_features=768, bias=True)
-            (out_proj): Linear(in_features=768, out_features=768, bias=True)
-          )
-          (layer_norm1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-          (mlp): CLIPMLP(
-            (activation_fn): QuickGELUActivation()
-            (fc1): Linear(in_features=768, out_features=3072, bias=True)
-            (fc2): Linear(in_features=3072, out_features=768, bias=True)
-          )
-          (layer_norm2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-        )
-      )
-    )
-    (final_layer_norm): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
-  )
-)

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/dataset/base_dataset.yaml

@@ -1,29 +0,0 @@
-data:
-  dataset1:
-    _target_: hcpdiff.data.TextImagePairDataset
-    _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.Text2ImageAttMapSource
-        img_root: '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
-
-        word_names: {}
-
-        text_transforms:
-          _target_: torchvision.transforms.Compose
-          transforms:
-            - _target_: hcpdiff.utils.caption_tools.TemplateFill
-              word_names: ${....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/dataset/regularization_dataset.yaml

@@ -1,31 +0,0 @@
-_base_: [cfgs/train/dataset/base_dataset.yaml]
-
-data:
-  dataset_class:
-    _target_: hcpdiff.data.TextImagePairDataset
-    _partial_: True
-    batch_size: 1
-    cache_latents: True
-    att_mask_encode: False
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        _target_: hcpdiff.data.source.Text2ImageAttMapSource
-        img_root: 'imgs/db_class'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null
-        att_mask: null
-        bg_color: [ 255, 255, 255 ] # RGB; for ARGB -> RGB
-
-        word_names:
-          pt1: ''
-
-        text_transforms:
-          _target_: torchvision.transforms.Compose
-          transforms:
-            - _target_: hcpdiff.utils.caption_tools.TemplateFill
-              word_names: ${....word_names}
-    bucket:
-      _target_: hcpdiff.data.bucket.FixedBucket
-      target_size: [ 512, 512 ]

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

@@ -1,74 +0,0 @@
-_base_:
-  - cfgs/train/dataset/regularization_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-unet:
-  -
-    lr: 1e-6
-    layers: # k,v of cross attention
-      - 're:.*attn2\.to_k$'
-      - 're:.*attn2\.to_v$'
-
-## lora version of CustomDiffusion
-#lora_unet:
-#  -
-#    lr: 1e-4
-#    layers:
-#      - 're:.*attn2\.to_k$'
-#      - 're:.*attn2\.to_v$'
-
-tokenizer_pt:
-  train: # prompt tuning embeddings, needs to be created in advance
-    - { name: 'pt-dog1', lr: 0.003 }
-
-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:
-    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: sks
-          class: dog
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 1
-
-  dataset_class:
-    batch_size: 1
-    cache_latents: True
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        img_root: 'imgs/db_class'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null
-
-        word_names:
-          class: dog
-    bucket:
-      _target_: hcpdiff.data.bucket.FixedBucket
-      target_size: [512, 512]

hcpdiff-0.9.0.data/data/hcpdiff/cfgs/train/examples/DreamArtist++.yaml

@@ -1,135 +0,0 @@
-_base_:
-  - cfgs/train/dataset/regularization_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-unet: null
-
-lora_unet:
-  - lr: 1e-4
-    rank: 0.01875
-    branch: p
-    dropout: 0.1
-    layers:
-      - 're:.*\.to_k$'
-      - 're:.*\.to_v$'
-      - 're:.*\.ff$'
-      #- 're:.*\.attn.?$' # Increases fitness, but potentially reduces controllability
-  - lr: 4e-5 # Low negative unet lr prevents image collapse
-    rank: 0.01875
-    branch: n
-    dropout: 0.1
-    layers:
-      - 're:.*\.to_k$'
-      - 're:.*\.to_v$'
-      - 're:.*\.ff$'
-      #- 're:.*\.attn.?$' # Increases fitness, but potentially reduces controllability
-  #  - lr: 1e-4
-  #    rank: 0.01875
-  #    type: p
-  #    layers:
-  #      - 're:.*\.resnets$' # Increases fitness, but potentially reduces controllability and change style
-  #  - lr: 4e-5
-  #    rank: 0.01875
-  #    type: n
-  #    layers:
-  #      - 're:.*\.resnets$' # Increases fitness, but potentially reduces controllability and change style
-
-lora_text_encoder:
-  - lr: 2e-5
-    rank: 2
-    branch: p
-    dropout: 0.1
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-  - lr: 2e-5
-    rank: 2
-    branch: n
-    dropout: 0.1
-    layers:
-      - 're:.*self_attn$'
-      - 're:.*mlp$'
-
-tokenizer_pt:
-  train: # prompt tuning embeddings
-    - { name: 'pt-botdog1', lr: 0.0025 }
-    - { name: 'pt-botdog1-neg', lr: 0.0025 }
-
-train:
-  gradient_accumulation_steps: 1
-  save_step: 100
-
-  #cfg_scale: '1.0-3.0:cos' # dynamic CFG with timestamp
-  cfg_scale: '3.0'
-
-  loss:
-    criterion: # min SNR loss
-      _target_: hcpdiff.loss.MinSNRLoss
-      gamma: 2.0
-
-  scheduler:
-    name: one_cycle
-    num_warmup_steps: 200
-    num_training_steps: 1000
-    scheduler_kwargs: { }
-
-  scheduler_pt:
-    name: one_cycle
-    num_warmup_steps: 200
-    num_training_steps: 1000
-    scheduler_kwargs: {}
-
-model:
-  pretrained_model_name_or_path: 'runwayml/stable-diffusion-v1-5'
-  tokenizer_repeats: 1
-  ema_unet: 0
-  ema_text_encoder: 0
-  clip_skip: 0
-
-# The dataset configuration inherits regularization_dataset.yaml
-# 数据集配置继承自 regularization_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
-
-        word_names: --- # remove unused item
-        text_transforms:
-          transforms: # without TagShuffle and TagDropout. | 去掉 TagShuffle 和 TagDropout.
-            - _target_: hcpdiff.utils.caption_tools.TemplateFill
-              word_names:
-                pt1: [pt-botdog1, pt-botdog1-neg]
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 1
-
-  # Add regularization to prevent image crashes
-  # Regularization images is generated by model itself with prompt from dataset
-  dataset_class:
-    batch_size: 1
-    cache_latents: True
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        img_root: 'imgs/v15'
-        prompt_template: 'prompt_tuning_template/caption.txt'
-        caption_file:
-          _targe_: hcpdiff.data.JsonCaptionLoader
-          path: 'imgs/v15/image_captions.json'
-        att_mask: null
-
-        word_names:
-          pt1: ['', '']
-    bucket:
-      _target_: hcpdiff.data.bucket.FixedBucket
-      target_size: [512, 512]

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

@@ -1,45 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-tokenizer_pt:
-  train: # prompt tuning embeddings, needs to be created in advance
-    - { name: 'pt-catgirl1', lr: 0.003 }
-    - { name: 'pt-catgirl1-neg', lr: 0.003 }
-
-train:
-  gradient_accumulation_steps: 1
-  save_step: 100
-
-  #cfg_scale: '1.0-3.0:cos' # dynamic CFG with timestamp
-  cfg_scale: '3.0'
-
-  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:
-    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-catgirl1, pt-catgirl1-neg ] # A pair of word for positive and negative branches respectively.
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 1

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

@@ -1,62 +0,0 @@
-_base_:
-  - cfgs/train/dataset/regularization_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-unet:
-  -
-    lr: 1e-6
-    layers:
-      - ''
-
-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
-
-# The dataset configuration inherits regularization_dataset.yaml
-# 数据集配置继承自 regularization_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
-
-        word_names:
-          pt1: sks
-          class: dog
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 1
-
-  dataset_class:
-    batch_size: 1
-    cache_latents: True
-    loss_weight: 1.0
-
-    source:
-      data_source1:
-        img_root: 'imgs/db_class'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: null
-
-        word_names:
-          class: dog
-    bucket:
-      _target_: hcpdiff.data.bucket.FixedBucket
-      target_size: [512, 512]

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

@@ -1,33 +0,0 @@
-_base_:
-  - cfgs/train/examples/fine-tuning.yaml
-
-mixed_precision: 'bf16'
-
-train:
-  optimizer:
-    _target_: transformers.optimization.Adafactor
-    _partial_: True
-    relative_step: False
-    weight_decay: 1e-3
-
-model:
-  pretrained_model_name_or_path: 'stabilityai/stable-diffusion-xl-base-1.0'
-  clip_skip: 1
-  clip_final_norm: False
-  force_cast_precision: True
-
-data:
-  dataset1:
-    _target_: hcpdiff.data.CropInfoPairDataset
-    batch_size: 4
-
-    source:
-      data_source1:
-        img_root: 'imgs/'
-        prompt_template: 'prompt_tuning_template/object.txt'
-        caption_file: 'imgs/image_captions.json' # path to image captions (file_words)
-
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"1024*1024"}
-      num_bucket: 4

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

@@ -1,17 +0,0 @@
-_base_: [cfgs/train/examples/fine-tuning.yaml]
-
-# Install: pip install lion-pytorch
-
-train:
-  optimizer:
-    type: --- # remove unused item
-    _target_: lion_pytorch.Lion
-    _partial_: True
-    weight_decay: 1e-2
-    #use_triton: True # set this to True to use cuda kernel w/ Triton lang (Tillet et al)
-
-  optimizer_pt:
-    type: ---
-    _target_: lion_pytorch.Lion
-    _partial_: True
-    weight_decay: 1e-3

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

@@ -1,41 +0,0 @@
-_base_:
-  - cfgs/train/dataset/base_dataset.yaml
-  - cfgs/train/train_base.yaml
-  - cfgs/train/tuning_base.yaml
-
-tokenizer_pt:
-  train: # prompt tuning embeddings, needs to be created in advance
-    - { name: 'pt-catgirl1', lr: 0.003 }
-
-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:
-    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-catgirl1
-    bucket:
-      _target_: hcpdiff.data.bucket.RatioBucket.from_files # aspect ratio bucket
-      target_area: ${hcp.eval:"512*512"}
-      num_bucket: 1

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

@@ -1,15 +0,0 @@
-_base_: [cfgs/train/train_base.yaml]
-
-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
-    log_step: 5