Merge pull request #360 from kohya-ss/dev

fix for merge_lora.py

README.md

@@ -127,101 +127,75 @@ The majority of scripts is licensed under ASL 2.0 (including codes from Diffuser
 
 ## Change History
 
-- 10 Mar. 2023, 2023/3/10: release v0.5.1
-  - Fix to LoRA modules in the model are same to the previous (before 0.5.0) if Conv2d-3x3 is disabled (no `conv_dim` arg, default).
-    - Conv2D with kernel size 1x1 in ResNet modules were accidentally included in v0.5.0.
-    - Trained models with v0.5.0 will work with Web UI's built-in LoRA and Additional Networks extension.
-  - Fix an issue that dim (rank) of LoRA module is limited to the in/out dimensions of the target Linear/Conv2d (in case of the dim > 320).
-  - `resize_lora.py` now have a feature to `dynamic resizing` which means each LoRA module can have different ranks (dims). Thanks to mgz-dev for this great work!
-    - The appropriate rank is selected based on the complexity of each module with an algorithm specified in the command line arguments. For details: https://github.com/kohya-ss/sd-scripts/pull/243
-  - Multiple GPUs training is finally supported in `train_network.py`. Thanks to ddPn08 to solve this long running issue!
-  - Dataset with fine-tuning method (with metadata json) now works without images if `.npz` files exist. Thanks to rvhfxb!
-  - `train_network.py` can work if the current directory is not the directory where the script is in. Thanks to mio2333!
-  - Fix `extract_lora_from_models.py` and `svd_merge_lora.py` doesn't work with higher rank (>320).
+- 1 Apr. 2023, 2023/4/1:
+  - Fix an issue that `merge_lora.py` does not work with the latest version.
+  - Fix an issue that `merge_lora.py` does not merge Conv2d3x3 weights.
+  - 最新のバージョンで`merge_lora.py` が動作しない不具合を修正しました。
+  - `merge_lora.py` で `no module found for LoRA weight: ...` と表示され Conv2d3x3 拡張の重みがマージされない不具合を修正しました。
+- 31 Mar. 2023, 2023/3/31:
+  - Fix an issue that the VRAM usage temporarily increases when loading a model in `train_network.py`.
+  - Fix an issue that an error occurs when loading a `.safetensors` model in `train_network.py`. [#354](https://github.com/kohya-ss/sd-scripts/issues/354)
+  - `train_network.py` でモデル読み込み時にVRAM使用量が一時的に大きくなる不具合を修正しました。
+  - `train_network.py` で `.safetensors` 形式のモデルを読み込むとエラーになる不具合を修正しました。[#354](https://github.com/kohya-ss/sd-scripts/issues/354)
+- 30 Mar. 2023, 2023/3/30:
+  - Support [P+](https://prompt-plus.github.io/) training. Thank you jakaline-dev!
+    - See [#327](https://github.com/kohya-ss/sd-scripts/pull/327) for details.
+    - Use `train_textual_inversion_XTI.py` for training. The usage is almost the same as `train_textual_inversion.py`. However, sample image generation during training is not supported.
+    - Use `gen_img_diffusers.py` for image generation (I think Web UI is not supported). Specify the embedding with `--XTI_embeddings` option.
+  - Reduce RAM usage at startup in `train_network.py`. [#332](https://github.com/kohya-ss/sd-scripts/pull/332)  Thank you guaneec!
+  - Support pre-merge for LoRA in `gen_img_diffusers.py`. Specify `--network_merge` option. Note that the `--am` option of the prompt option is no longer available with this option.
 
-  - LoRAのConv2d-3x3拡張を行わない場合（`conv_dim` を指定しない場合）、以前（v0.5.0）と同じ構成になるよう修正しました。
-    - ResNetのカーネルサイズ1x1のConv2dが誤って対象になっていました。
-    - ただv0.5.0で学習したモデルは Additional Networks 拡張、およびWeb UIのLoRA機能で問題なく使えると思われます。
-  - LoRAモジュールの dim (rank) が、対象モジュールの次元数以下に制限される不具合を修正しました（320より大きい dim を指定した場合）。
-  - `resize_lora.py` に `dynamic resizing` （リサイズ後の各LoRAモジュールが異なるrank (dim) を持てる機能）を追加しました。mgz-dev 氏の貢献に感謝します。
-    - 適切なランクがコマンドライン引数で指定したアルゴリズムにより自動的に選択されます。詳細はこちらをご覧ください: https://github.com/kohya-ss/sd-scripts/pull/243
-  - `train_network.py` でマルチGPU学習をサポートしました。長年の懸案を解決された ddPn08 氏に感謝します。
-  - fine-tuning方式のデータセット（メタデータ.jsonファイルを使うデータセット）で `.npz` が存在するときには画像がなくても動作するようになりました。rvhfxb 氏に感謝します。
-  - 他のディレクトリから `train_network.py` を呼び出しても動作するよう変更しました。 mio2333 氏に感謝します。
-  - `extract_lora_from_models.py` および `svd_merge_lora.py` が320より大きいrankを指定すると動かない不具合を修正しました。
-  
-- 9 Mar. 2023, 2023/3/9: release v0.5.0
-  - There may be problems due to major changes. If you cannot revert back to the previous version when problems occur, please do not update for a while.
-  - Minimum metadata (module name, dim, alpha and network_args) is recorded even with `--no_metadata`, issue https://github.com/kohya-ss/sd-scripts/issues/254
-  - `train_network.py` supports LoRA for Conv2d-3x3 (extended to conv2d with a kernel size not 1x1).
-    - Same as a current version of [LoCon](https://github.com/KohakuBlueleaf/LoCon). __Thank you very much KohakuBlueleaf for your help!__
-      - LoCon will be enhanced in the future. Compatibility for future versions is not guaranteed.
-    - Specify `--network_args` option like: `--network_args "conv_dim=4" "conv_alpha=1"`
-    - [Additional Networks extension](https://github.com/kohya-ss/sd-webui-additional-networks) version 0.5.0 or later is required to use 'LoRA for Conv2d-3x3' in Stable Diffusion web UI.
-    - __Stable Diffusion web UI built-in LoRA does not support 'LoRA for Conv2d-3x3' now. Consider carefully whether or not to use it.__
-  - Merging/extracting scripts also support LoRA for Conv2d-3x3.
-  - Free CUDA memory after sample generation to reduce VRAM usage, issue https://github.com/kohya-ss/sd-scripts/issues/260 
-  - Empty caption doesn't cause error now, issue https://github.com/kohya-ss/sd-scripts/issues/258
-  - Fix sample generation is crashing in Textual Inversion training when using templates, or if height/width is not divisible by 8.
-  - Update documents (Japanese only).
+  - [P+](https://prompt-plus.github.io/) の学習に対応しました。jakaline-dev氏に感謝します。 
+    - 詳細は [#327](https://github.com/kohya-ss/sd-scripts/pull/327) をご参照ください。
+    - 学習には `train_textual_inversion_XTI.py` を使用します。使用法は `train_textual_inversion.py` とほぼ同じです。た
+    だし学習中のサンプル生成には対応していません。
+    - 画像生成には `gen_img_diffusers.py` を使用してください（Web UIは対応していないと思われます）。`--XTI_embeddings` オプションで学習したembeddingを指定してください。
+  - `train_network.py` で起動時のRAM使用量を削減しました。[#332](https://github.com/kohya-ss/sd-scripts/pull/332) guaneec氏に感謝します。
+  - `gen_img_diffusers.py` でLoRAの事前マージに対応しました。`--network_merge` オプションを指定してください。なおプロンプトオプションの `--am` は使用できなくなります。
 
-  - 大きく変更したため不具合があるかもしれません。問題が起きた時にスクリプトを前のバージョンに戻せない場合は、しばらく更新を控えてください。
-  - 最低限のメタデータ（module name, dim, alpha および network_args）が `--no_metadata` オプション指定時にも記録されます。issue https://github.com/kohya-ss/sd-scripts/issues/254
-  - `train_network.py` で LoRAの Conv2d-3x3 拡張に対応しました（カーネルサイズ1x1以外のConv2dにも対象範囲を拡大します）。
-    - 現在のバージョンの [LoCon](https://github.com/KohakuBlueleaf/LoCon) と同一の仕様です。__KohakuBlueleaf氏のご支援に深く感謝します。__
-      - LoCon が将来的に拡張された場合、それらのバージョンでの互換性は保証できません。
-    - `--network_args` オプションを `--network_args "conv_dim=4" "conv_alpha=1"` のように指定してください。
-    - Stable Diffusion web UI での使用には [Additional Networks extension](https://github.com/kohya-ss/sd-webui-additional-networks) のversion 0.5.0 以降が必要です。
-    - __Stable Diffusion web UI の LoRA 機能は LoRAの Conv2d-3x3 拡張に対応していないようです。使用するか否か慎重にご検討ください。__
-  - マージ、抽出のスクリプトについても LoRA の Conv2d-3x3 拡張に対応しました.
-  - サンプル画像生成後にCUDAメモリを解放しVRAM使用量を削減しました。 issue https://github.com/kohya-ss/sd-scripts/issues/260 
-  - 空のキャプションが使えるようになりました。 issue https://github.com/kohya-ss/sd-scripts/issues/258
-  - Textual Inversion 学習でテンプレートを使ったとき、height/width が 8 で割り切れなかったときにサンプル画像生成がクラッシュするのを修正しました。
-  - ドキュメント類を更新しました。
+## Sample image generation during training
+  A prompt file might look like this, for example
 
-  - Sample image generation:
-    A prompt file might look like this, for example
+```
+# prompt 1
+masterpiece, best quality, (1girl), in white shirts, upper body, looking at viewer, simple background --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 768 --h 768 --d 1 --l 7.5 --s 28
 
-    ```
-    # prompt 1
-    masterpiece, best quality, 1girl, in white shirts, upper body, looking at viewer, simple background --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 768 --h 768 --d 1 --l 7.5 --s 28
+# prompt 2
+masterpiece, best quality, 1boy, in business suit, standing at street, looking back --n (low quality, worst quality), bad anatomy,bad composition, poor, low effort --w 576 --h 832 --d 2 --l 5.5 --s 40
+```
 
-    # prompt 2
-    masterpiece, best quality, 1boy, in business suit, standing at street, looking back --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 576 --h 832 --d 2 --l 5.5 --s 40
-    ```
+  Lines beginning with `#` are comments. You can specify options for the generated image with options like `--n` after the prompt. The following can be used.
 
-    Lines beginning with `#` are comments. You can specify options for the generated image with options like `--n` after the prompt. The following can be used.
+  * `--n` Negative prompt up to the next option.
+  * `--w` Specifies the width of the generated image.
+  * `--h` Specifies the height of the generated image.
+  * `--d` Specifies the seed of the generated image.
+  * `--l` Specifies the CFG scale of the generated image.
+  * `--s` Specifies the number of steps in the generation.
 
-    * `--n` Negative prompt up to the next option.
-    * `--w` Specifies the width of the generated image.
-    * `--h` Specifies the height of the generated image.
-    * `--d` Specifies the seed of the generated image.
-    * `--l` Specifies the CFG scale of the generated image.
-    * `--s` Specifies the number of steps in the generation.
+  The prompt weighting such as `( )` and `[ ]` are working.
 
-    The prompt weighting such as `( )` and `[ ]` are not working.
+## サンプル画像生成
+プロンプトファイルは例えば以下のようになります。
 
-  - サンプル画像生成：
-    プロンプトファイルは例えば以下のようになります。
+```
+# prompt 1
+masterpiece, best quality, (1girl), in white shirts, upper body, looking at viewer, simple background --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 768 --h 768 --d 1 --l 7.5 --s 28
 
-    ```
-    # prompt 1
-    masterpiece, best quality, 1girl, in white shirts, upper body, looking at viewer, simple background --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 768 --h 768 --d 1 --l 7.5 --s 28
+# prompt 2
+masterpiece, best quality, 1boy, in business suit, standing at street, looking back --n (low quality, worst quality), bad anatomy,bad composition, poor, low effort --w 576 --h 832 --d 2 --l 5.5 --s 40
+```
 
-    # prompt 2
-    masterpiece, best quality, 1boy, in business suit, standing at street, looking back --n low quality, worst quality, bad anatomy,bad composition, poor, low effort --w 576 --h 832 --d 2 --l 5.5 --s 40
-    ```
+  `#` で始まる行はコメントになります。`--n` のように「ハイフン二個＋英小文字」の形でオプションを指定できます。以下が使用可能できます。
 
-    `#` で始まる行はコメントになります。`--n` のように「ハイフン二個＋英小文字」の形でオプションを指定できます。以下が使用可能できます。
+  * `--n` Negative prompt up to the next option.
+  * `--w` Specifies the width of the generated image.
+  * `--h` Specifies the height of the generated image.
+  * `--d` Specifies the seed of the generated image.
+  * `--l` Specifies the CFG scale of the generated image.
+  * `--s` Specifies the number of steps in the generation.
 
-    * `--n` Negative prompt up to the next option.
-    * `--w` Specifies the width of the generated image.
-    * `--h` Specifies the height of the generated image.
-    * `--d` Specifies the seed of the generated image.
-    * `--l` Specifies the CFG scale of the generated image.
-    * `--s` Specifies the number of steps in the generation.
-
-    `( )` や `[ ]` などの重みづけは動作しません。
+  `( )` や `[ ]` などの重みづけも動作します。
 
 Please read [Releases](https://github.com/kohya-ss/sd-scripts/releases) for recent updates.
 最近の更新情報は [Release](https://github.com/kohya-ss/sd-scripts/releases) をご覧ください。

XTI_hijack.py

@@ -0,0 +1,209 @@
+import torch
+from typing import Union, List, Optional, Dict, Any, Tuple
+from diffusers.models.unet_2d_condition import UNet2DConditionOutput
+
+def unet_forward_XTI(self,
+        sample: torch.FloatTensor,
+        timestep: Union[torch.Tensor, float, int],
+        encoder_hidden_states: torch.Tensor,
+        class_labels: Optional[torch.Tensor] = None,
+        return_dict: bool = True,
+    ) -> Union[UNet2DConditionOutput, Tuple]:
+        r"""
+        Args:
+            sample (`torch.FloatTensor`): (batch, channel, height, width) noisy inputs tensor
+            timestep (`torch.FloatTensor` or `float` or `int`): (batch) timesteps
+            encoder_hidden_states (`torch.FloatTensor`): (batch, sequence_length, feature_dim) encoder hidden states
+            return_dict (`bool`, *optional*, defaults to `True`):
+                Whether or not to return a [`models.unet_2d_condition.UNet2DConditionOutput`] instead of a plain tuple.
+
+        Returns:
+            [`~models.unet_2d_condition.UNet2DConditionOutput`] or `tuple`:
+            [`~models.unet_2d_condition.UNet2DConditionOutput`] if `return_dict` is True, otherwise a `tuple`. When
+            returning a tuple, the first element is the sample tensor.
+        """
+        # By default samples have to be AT least a multiple of the overall upsampling factor.
+        # The overall upsampling factor is equal to 2 ** (# num of upsampling layears).
+        # However, the upsampling interpolation output size can be forced to fit any upsampling size
+        # on the fly if necessary.
+        default_overall_up_factor = 2**self.num_upsamplers
+
+        # upsample size should be forwarded when sample is not a multiple of `default_overall_up_factor`
+        forward_upsample_size = False
+        upsample_size = None
+
+        if any(s % default_overall_up_factor != 0 for s in sample.shape[-2:]):
+            logger.info("Forward upsample size to force interpolation output size.")
+            forward_upsample_size = True
+
+        # 0. center input if necessary
+        if self.config.center_input_sample:
+            sample = 2 * sample - 1.0
+
+        # 1. time
+        timesteps = timestep
+        if not torch.is_tensor(timesteps):
+            # TODO: this requires sync between CPU and GPU. So try to pass timesteps as tensors if you can
+            # This would be a good case for the `match` statement (Python 3.10+)
+            is_mps = sample.device.type == "mps"
+            if isinstance(timestep, float):
+                dtype = torch.float32 if is_mps else torch.float64
+            else:
+                dtype = torch.int32 if is_mps else torch.int64
+            timesteps = torch.tensor([timesteps], dtype=dtype, device=sample.device)
+        elif len(timesteps.shape) == 0:
+            timesteps = timesteps[None].to(sample.device)
+
+        # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+        timesteps = timesteps.expand(sample.shape[0])
+
+        t_emb = self.time_proj(timesteps)
+
+        # timesteps does not contain any weights and will always return f32 tensors
+        # but time_embedding might actually be running in fp16. so we need to cast here.
+        # there might be better ways to encapsulate this.
+        t_emb = t_emb.to(dtype=self.dtype)
+        emb = self.time_embedding(t_emb)
+
+        if self.config.num_class_embeds is not None:
+            if class_labels is None:
+                raise ValueError("class_labels should be provided when num_class_embeds > 0")
+            class_emb = self.class_embedding(class_labels).to(dtype=self.dtype)
+            emb = emb + class_emb
+
+        # 2. pre-process
+        sample = self.conv_in(sample)
+
+        # 3. down
+        down_block_res_samples = (sample,)
+        down_i = 0
+        for downsample_block in self.down_blocks:
+            if hasattr(downsample_block, "has_cross_attention") and downsample_block.has_cross_attention:
+                sample, res_samples = downsample_block(
+                    hidden_states=sample,
+                    temb=emb,
+                    encoder_hidden_states=encoder_hidden_states[down_i:down_i+2],
+                )
+                down_i += 2
+            else:
+                sample, res_samples = downsample_block(hidden_states=sample, temb=emb)
+
+            down_block_res_samples += res_samples
+
+        # 4. mid
+        sample = self.mid_block(sample, emb, encoder_hidden_states=encoder_hidden_states[6])
+
+        # 5. up
+        up_i = 7
+        for i, upsample_block in enumerate(self.up_blocks):
+            is_final_block = i == len(self.up_blocks) - 1
+
+            res_samples = down_block_res_samples[-len(upsample_block.resnets) :]
+            down_block_res_samples = down_block_res_samples[: -len(upsample_block.resnets)]
+
+            # if we have not reached the final block and need to forward the
+            # upsample size, we do it here
+            if not is_final_block and forward_upsample_size:
+                upsample_size = down_block_res_samples[-1].shape[2:]
+
+            if hasattr(upsample_block, "has_cross_attention") and upsample_block.has_cross_attention:
+                sample = upsample_block(
+                    hidden_states=sample,
+                    temb=emb,
+                    res_hidden_states_tuple=res_samples,
+                    encoder_hidden_states=encoder_hidden_states[up_i:up_i+3],
+                    upsample_size=upsample_size,
+                )
+                up_i += 3
+            else:
+                sample = upsample_block(
+                    hidden_states=sample, temb=emb, res_hidden_states_tuple=res_samples, upsample_size=upsample_size
+                )
+        # 6. post-process
+        sample = self.conv_norm_out(sample)
+        sample = self.conv_act(sample)
+        sample = self.conv_out(sample)
+
+        if not return_dict:
+            return (sample,)
+
+        return UNet2DConditionOutput(sample=sample)
+
+def downblock_forward_XTI(
+    self, hidden_states, temb=None, encoder_hidden_states=None, attention_mask=None, cross_attention_kwargs=None
+):
+    output_states = ()
+    i = 0
+
+    for resnet, attn in zip(self.resnets, self.attentions):
+        if self.training and self.gradient_checkpointing:
+
+            def create_custom_forward(module, return_dict=None):
+                def custom_forward(*inputs):
+                    if return_dict is not None:
+                        return module(*inputs, return_dict=return_dict)
+                    else:
+                        return module(*inputs)
+
+                return custom_forward
+
+            hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
+            hidden_states = torch.utils.checkpoint.checkpoint(
+                create_custom_forward(attn, return_dict=False), hidden_states, encoder_hidden_states[i]
+            )[0]
+        else:
+            hidden_states = resnet(hidden_states, temb)
+            hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states[i]).sample
+
+        output_states += (hidden_states,)
+        i += 1
+
+    if self.downsamplers is not None:
+        for downsampler in self.downsamplers:
+            hidden_states = downsampler(hidden_states)
+
+        output_states += (hidden_states,)
+
+    return hidden_states, output_states
+
+def upblock_forward_XTI(
+    self,
+    hidden_states,
+    res_hidden_states_tuple,
+    temb=None,
+    encoder_hidden_states=None,
+    upsample_size=None,
+):
+    i = 0
+    for resnet, attn in zip(self.resnets, self.attentions):
+        # pop res hidden states
+        res_hidden_states = res_hidden_states_tuple[-1]
+        res_hidden_states_tuple = res_hidden_states_tuple[:-1]
+        hidden_states = torch.cat([hidden_states, res_hidden_states], dim=1)
+
+        if self.training and self.gradient_checkpointing:
+
+            def create_custom_forward(module, return_dict=None):
+                def custom_forward(*inputs):
+                    if return_dict is not None:
+                        return module(*inputs, return_dict=return_dict)
+                    else:
+                        return module(*inputs)
+
+                return custom_forward
+
+            hidden_states = torch.utils.checkpoint.checkpoint(create_custom_forward(resnet), hidden_states, temb)
+            hidden_states = torch.utils.checkpoint.checkpoint(
+                create_custom_forward(attn, return_dict=False), hidden_states, encoder_hidden_states[i]
+            )[0]
+        else:
+            hidden_states = resnet(hidden_states, temb)
+            hidden_states = attn(hidden_states, encoder_hidden_states=encoder_hidden_states[i]).sample
+        
+        i += 1
+
+    if self.upsamplers is not None:
+        for upsampler in self.upsamplers:
+            hidden_states = upsampler(hidden_states, upsample_size)
+
+    return hidden_states

fine_tune.py

@@ -5,6 +5,8 @@ import argparse
 import gc
 import math
 import os
+import toml
+from multiprocessing import Value
 
 from tqdm import tqdm
 import torch
@@ -15,351 +17,414 @@ from diffusers import DDPMScheduler
 import library.train_util as train_util
 import library.config_util as config_util
 from library.config_util import (
-  ConfigSanitizer,
-  BlueprintGenerator,
+    ConfigSanitizer,
+    BlueprintGenerator,
 )
-
-def collate_fn(examples):
-  return examples[0]
+import library.custom_train_functions as custom_train_functions
+from library.custom_train_functions import apply_snr_weight
 
 
 def train(args):
-  train_util.verify_training_args(args)
-  train_util.prepare_dataset_args(args, True)
+    train_util.verify_training_args(args)
+    train_util.prepare_dataset_args(args, True)
 
-  cache_latents = args.cache_latents
+    cache_latents = args.cache_latents
 
-  if args.seed is not None:
-    set_seed(args.seed)                           # 乱数系列を初期化する
+    if args.seed is not None:
+        set_seed(args.seed)  # 乱数系列を初期化する
 
-  tokenizer = train_util.load_tokenizer(args)
+    tokenizer = train_util.load_tokenizer(args)
 
-  blueprint_generator = BlueprintGenerator(ConfigSanitizer(False, True, True))
-  if args.dataset_config is not None:
-    print(f"Load dataset config from {args.dataset_config}")
-    user_config = config_util.load_user_config(args.dataset_config)
-    ignored = ["train_data_dir", "in_json"]
-    if any(getattr(args, attr) is not None for attr in ignored):
-      print("ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(', '.join(ignored)))
-  else:
-    user_config = {
-      "datasets": [{
-        "subsets": [{
-          "image_dir": args.train_data_dir,
-          "metadata_file": args.in_json,
-        }]
-      }]
-    }
-
-  blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
-  train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
-
-  if args.debug_dataset:
-    train_util.debug_dataset(train_dataset_group)
-    return
-  if len(train_dataset_group) == 0:
-    print("No data found. Please verify the metadata file and train_data_dir option. / 画像がありません。メタデータおよびtrain_data_dirオプションを確認してください。")
-    return
-
-  if cache_latents:
-    assert train_dataset_group.is_latent_cacheable(), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
-
-  # acceleratorを準備する
-  print("prepare accelerator")
-  accelerator, unwrap_model = train_util.prepare_accelerator(args)
-
-  # mixed precisionに対応した型を用意しておき適宜castする
-  weight_dtype, save_dtype = train_util.prepare_dtype(args)
-
-  # モデルを読み込む
-  text_encoder, vae, unet, load_stable_diffusion_format = train_util.load_target_model(args, weight_dtype)
-
-  # verify load/save model formats
-  if load_stable_diffusion_format:
-    src_stable_diffusion_ckpt = args.pretrained_model_name_or_path
-    src_diffusers_model_path = None
-  else:
-    src_stable_diffusion_ckpt = None
-    src_diffusers_model_path = args.pretrained_model_name_or_path
-
-  if args.save_model_as is None:
-    save_stable_diffusion_format = load_stable_diffusion_format
-    use_safetensors = args.use_safetensors
-  else:
-    save_stable_diffusion_format = args.save_model_as.lower() == 'ckpt' or args.save_model_as.lower() == 'safetensors'
-    use_safetensors = args.use_safetensors or ("safetensors" in args.save_model_as.lower())
-
-  # Diffusers版のxformers使用フラグを設定する関数
-  def set_diffusers_xformers_flag(model, valid):
-    #   model.set_use_memory_efficient_attention_xformers(valid)            # 次のリリースでなくなりそう
-    # pipeが自動で再帰的にset_use_memory_efficient_attention_xformersを探すんだって(;´Д｀)
-    # U-Netだけ使う時にはどうすればいいのか……仕方ないからコピって使うか
-    # 0.10.2でなんか巻き戻って個別に指定するようになった(;^ω^)
-
-    # Recursively walk through all the children.
-    # Any children which exposes the set_use_memory_efficient_attention_xformers method
-    # gets the message
-    def fn_recursive_set_mem_eff(module: torch.nn.Module):
-      if hasattr(module, "set_use_memory_efficient_attention_xformers"):
-        module.set_use_memory_efficient_attention_xformers(valid)
-
-      for child in module.children():
-        fn_recursive_set_mem_eff(child)
-
-    fn_recursive_set_mem_eff(model)
-
-  # モデルに xformers とか memory efficient attention を組み込む
-  if args.diffusers_xformers:
-    print("Use xformers by Diffusers")
-    set_diffusers_xformers_flag(unet, True)
-  else:
-    # Windows版のxformersはfloatで学習できないのでxformersを使わない設定も可能にしておく必要がある
-    print("Disable Diffusers' xformers")
-    set_diffusers_xformers_flag(unet, False)
-    train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
-
-  # 学習を準備する
-  if cache_latents:
-    vae.to(accelerator.device, dtype=weight_dtype)
-    vae.requires_grad_(False)
-    vae.eval()
-    with torch.no_grad():
-      train_dataset_group.cache_latents(vae)
-    vae.to("cpu")
-    if torch.cuda.is_available():
-      torch.cuda.empty_cache()
-    gc.collect()
-
-  # 学習を準備する：モデルを適切な状態にする
-  training_models = []
-  if args.gradient_checkpointing:
-    unet.enable_gradient_checkpointing()
-  training_models.append(unet)
-
-  if args.train_text_encoder:
-    print("enable text encoder training")
-    if args.gradient_checkpointing:
-      text_encoder.gradient_checkpointing_enable()
-    training_models.append(text_encoder)
-  else:
-    text_encoder.to(accelerator.device, dtype=weight_dtype)
-    text_encoder.requires_grad_(False)             # text encoderは学習しない
-    if args.gradient_checkpointing:
-      text_encoder.gradient_checkpointing_enable()
-      text_encoder.train()                # required for gradient_checkpointing
+    blueprint_generator = BlueprintGenerator(ConfigSanitizer(False, True, True))
+    if args.dataset_config is not None:
+        print(f"Load dataset config from {args.dataset_config}")
+        user_config = config_util.load_user_config(args.dataset_config)
+        ignored = ["train_data_dir", "in_json"]
+        if any(getattr(args, attr) is not None for attr in ignored):
+            print(
+                "ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(
+                    ", ".join(ignored)
+                )
+            )
     else:
-      text_encoder.eval()
+        user_config = {
+            "datasets": [
+                {
+                    "subsets": [
+                        {
+                            "image_dir": args.train_data_dir,
+                            "metadata_file": args.in_json,
+                        }
+                    ]
+                }
+            ]
+        }
 
-  if not cache_latents:
-    vae.requires_grad_(False)
-    vae.eval()
-    vae.to(accelerator.device, dtype=weight_dtype)
+    blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
+    train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
 
-  for m in training_models:
-    m.requires_grad_(True)
-  params = []
-  for m in training_models:
-    params.extend(m.parameters())
-  params_to_optimize = params
+    current_epoch = Value("i", 0)
+    current_step = Value("i", 0)
+    ds_for_collater = train_dataset_group if args.max_data_loader_n_workers == 0 else None
+    collater = train_util.collater_class(current_epoch, current_step, ds_for_collater)
 
-  # 学習に必要なクラスを準備する
-  print("prepare optimizer, data loader etc.")
-  _, _, optimizer = train_util.get_optimizer(args, trainable_params=params_to_optimize)
+    if args.debug_dataset:
+        train_util.debug_dataset(train_dataset_group)
+        return
+    if len(train_dataset_group) == 0:
+        print(
+            "No data found. Please verify the metadata file and train_data_dir option. / 画像がありません。メタデータおよびtrain_data_dirオプションを確認してください。"
+        )
+        return
 
-  # dataloaderを準備する
-  # DataLoaderのプロセス数：0はメインプロセスになる
-  n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)      # cpu_count-1 ただし最大で指定された数まで
-  train_dataloader = torch.utils.data.DataLoader(
-      train_dataset_group, batch_size=1, shuffle=True, collate_fn=collate_fn, num_workers=n_workers, persistent_workers=args.persistent_data_loader_workers)
+    if cache_latents:
+        assert (
+            train_dataset_group.is_latent_cacheable()
+        ), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
 
-  # 学習ステップ数を計算する
-  if args.max_train_epochs is not None:
-    args.max_train_steps = args.max_train_epochs * len(train_dataloader)
-    print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
+    # acceleratorを準備する
+    print("prepare accelerator")
+    accelerator, unwrap_model = train_util.prepare_accelerator(args)
 
-  # lr schedulerを用意する
-  lr_scheduler = train_util.get_scheduler_fix(args.lr_scheduler, optimizer, num_warmup_steps=args.lr_warmup_steps,
-                                              num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-                                              num_cycles=args.lr_scheduler_num_cycles, power=args.lr_scheduler_power)
+    # mixed precisionに対応した型を用意しておき適宜castする
+    weight_dtype, save_dtype = train_util.prepare_dtype(args)
 
-  # 実験的機能：勾配も含めたfp16学習を行う　モデル全体をfp16にする
-  if args.full_fp16:
-    assert args.mixed_precision == "fp16", "full_fp16 requires mixed precision='fp16' / full_fp16を使う場合はmixed_precision='fp16'を指定してください。"
-    print("enable full fp16 training.")
-    unet.to(weight_dtype)
-    text_encoder.to(weight_dtype)
+    # モデルを読み込む
+    text_encoder, vae, unet, load_stable_diffusion_format = train_util.load_target_model(args, weight_dtype)
 
-  # acceleratorがなんかよろしくやってくれるらしい
-  if args.train_text_encoder:
-    unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        unet, text_encoder, optimizer, train_dataloader, lr_scheduler)
-  else:
-    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(unet, optimizer, train_dataloader, lr_scheduler)
+    # verify load/save model formats
+    if load_stable_diffusion_format:
+        src_stable_diffusion_ckpt = args.pretrained_model_name_or_path
+        src_diffusers_model_path = None
+    else:
+        src_stable_diffusion_ckpt = None
+        src_diffusers_model_path = args.pretrained_model_name_or_path
 
-  # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
-  if args.full_fp16:
-    train_util.patch_accelerator_for_fp16_training(accelerator)
+    if args.save_model_as is None:
+        save_stable_diffusion_format = load_stable_diffusion_format
+        use_safetensors = args.use_safetensors
+    else:
+        save_stable_diffusion_format = args.save_model_as.lower() == "ckpt" or args.save_model_as.lower() == "safetensors"
+        use_safetensors = args.use_safetensors or ("safetensors" in args.save_model_as.lower())
 
-  # resumeする
-  if args.resume is not None:
-    print(f"resume training from state: {args.resume}")
-    accelerator.load_state(args.resume)
+    # Diffusers版のxformers使用フラグを設定する関数
+    def set_diffusers_xformers_flag(model, valid):
+        #   model.set_use_memory_efficient_attention_xformers(valid)            # 次のリリースでなくなりそう
+        # pipeが自動で再帰的にset_use_memory_efficient_attention_xformersを探すんだって(;´Д｀)
+        # U-Netだけ使う時にはどうすればいいのか……仕方ないからコピって使うか
+        # 0.10.2でなんか巻き戻って個別に指定するようになった(;^ω^)
 
-  # epoch数を計算する
-  num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-  num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-  if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
-    args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
+        # Recursively walk through all the children.
+        # Any children which exposes the set_use_memory_efficient_attention_xformers method
+        # gets the message
+        def fn_recursive_set_mem_eff(module: torch.nn.Module):
+            if hasattr(module, "set_use_memory_efficient_attention_xformers"):
+                module.set_use_memory_efficient_attention_xformers(valid)
 
-  # 学習する
-  total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-  print("running training / 学習開始")
-  print(f"  num examples / サンプル数: {train_dataset_group.num_train_images}")
-  print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
-  print(f"  num epochs / epoch数: {num_train_epochs}")
-  print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
-  print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
-  print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
-  print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
+            for child in module.children():
+                fn_recursive_set_mem_eff(child)
 
-  progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
-  global_step = 0
+        fn_recursive_set_mem_eff(model)
 
-  noise_scheduler = DDPMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear",
-                                  num_train_timesteps=1000, clip_sample=False)
+    # モデルに xformers とか memory efficient attention を組み込む
+    if args.diffusers_xformers:
+        print("Use xformers by Diffusers")
+        set_diffusers_xformers_flag(unet, True)
+    else:
+        # Windows版のxformersはfloatで学習できないのでxformersを使わない設定も可能にしておく必要がある
+        print("Disable Diffusers' xformers")
+        set_diffusers_xformers_flag(unet, False)
+        train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
 
-  if accelerator.is_main_process:
-    accelerator.init_trackers("finetuning")
+    # 学習を準備する
+    if cache_latents:
+        vae.to(accelerator.device, dtype=weight_dtype)
+        vae.requires_grad_(False)
+        vae.eval()
+        with torch.no_grad():
+            train_dataset_group.cache_latents(vae, args.vae_batch_size)
+        vae.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        gc.collect()
 
-  for epoch in range(num_train_epochs):
-    print(f"epoch {epoch+1}/{num_train_epochs}")
-    train_dataset_group.set_current_epoch(epoch + 1)
+    # 学習を準備する：モデルを適切な状態にする
+    training_models = []
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+    training_models.append(unet)
+
+    if args.train_text_encoder:
+        print("enable text encoder training")
+        if args.gradient_checkpointing:
+            text_encoder.gradient_checkpointing_enable()
+        training_models.append(text_encoder)
+    else:
+        text_encoder.to(accelerator.device, dtype=weight_dtype)
+        text_encoder.requires_grad_(False)  # text encoderは学習しない
+        if args.gradient_checkpointing:
+            text_encoder.gradient_checkpointing_enable()
+            text_encoder.train()  # required for gradient_checkpointing
+        else:
+            text_encoder.eval()
+
+    if not cache_latents:
+        vae.requires_grad_(False)
+        vae.eval()
+        vae.to(accelerator.device, dtype=weight_dtype)
 
     for m in training_models:
-      m.train()
+        m.requires_grad_(True)
+    params = []
+    for m in training_models:
+        params.extend(m.parameters())
+    params_to_optimize = params
 
-    loss_total = 0
-    for step, batch in enumerate(train_dataloader):
-      with accelerator.accumulate(training_models[0]):  # 複数モデルに対応していない模様だがとりあえずこうしておく
-        with torch.no_grad():
-          if "latents" in batch and batch["latents"] is not None:
-            latents = batch["latents"].to(accelerator.device)
-          else:
-            # latentに変換
-            latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
-          latents = latents * 0.18215
-        b_size = latents.shape[0]
+    # 学習に必要なクラスを準備する
+    print("prepare optimizer, data loader etc.")
+    _, _, optimizer = train_util.get_optimizer(args, trainable_params=params_to_optimize)
 
-        with torch.set_grad_enabled(args.train_text_encoder):
-          # Get the text embedding for conditioning
-          input_ids = batch["input_ids"].to(accelerator.device)
-          encoder_hidden_states = train_util.get_hidden_states(
-              args, input_ids, tokenizer, text_encoder, None if not args.full_fp16 else weight_dtype)
+    # dataloaderを準備する
+    # DataLoaderのプロセス数：0はメインプロセスになる
+    n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)  # cpu_count-1 ただし最大で指定された数まで
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset_group,
+        batch_size=1,
+        shuffle=True,
+        collate_fn=collater,
+        num_workers=n_workers,
+        persistent_workers=args.persistent_data_loader_workers,
+    )
 
-        # Sample noise that we'll add to the latents
-        noise = torch.randn_like(latents, device=latents.device)
-        if args.noise_offset:
-          # https://www.crosslabs.org//blog/diffusion-with-offset-noise
-          noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
+    # 学習ステップ数を計算する
+    if args.max_train_epochs is not None:
+        args.max_train_steps = args.max_train_epochs * math.ceil(
+            len(train_dataloader) / accelerator.num_processes / args.gradient_accumulation_steps
+        )
+        print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
 
-        # Sample a random timestep for each image
-        timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
-        timesteps = timesteps.long()
+    # データセット側にも学習ステップを送信
+    train_dataset_group.set_max_train_steps(args.max_train_steps)
 
-        # Add noise to the latents according to the noise magnitude at each timestep
-        # (this is the forward diffusion process)
-        noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+    # lr schedulerを用意する
+    lr_scheduler = train_util.get_scheduler_fix(args, optimizer, accelerator.num_processes)
 
-        # Predict the noise residual
-        noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+    # 実験的機能：勾配も含めたfp16学習を行う　モデル全体をfp16にする
+    if args.full_fp16:
+        assert (
+            args.mixed_precision == "fp16"
+        ), "full_fp16 requires mixed precision='fp16' / full_fp16を使う場合はmixed_precision='fp16'を指定してください。"
+        print("enable full fp16 training.")
+        unet.to(weight_dtype)
+        text_encoder.to(weight_dtype)
 
-        if args.v_parameterization:
-          # v-parameterization training
-          target = noise_scheduler.get_velocity(latents, noise, timesteps)
-        else:
-          target = noise
+    # acceleratorがなんかよろしくやってくれるらしい
+    if args.train_text_encoder:
+        unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, text_encoder, optimizer, train_dataloader, lr_scheduler
+        )
+    else:
+        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(unet, optimizer, train_dataloader, lr_scheduler)
 
-        loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="mean")
+    # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
+    if args.full_fp16:
+        train_util.patch_accelerator_for_fp16_training(accelerator)
 
-        accelerator.backward(loss)
-        if accelerator.sync_gradients and args.max_grad_norm != 0.0:
-          params_to_clip = []
-          for m in training_models:
-            params_to_clip.extend(m.parameters())
-          accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+    # resumeする
+    if args.resume is not None:
+        print(f"resume training from state: {args.resume}")
+        accelerator.load_state(args.resume)
 
-        optimizer.step()
-        lr_scheduler.step()
-        optimizer.zero_grad(set_to_none=True)
+    # epoch数を計算する
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+    if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
+        args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
 
-      # Checks if the accelerator has performed an optimization step behind the scenes
-      if accelerator.sync_gradients:
-        progress_bar.update(1)
-        global_step += 1
+    # 学習する
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+    print("running training / 学習開始")
+    print(f"  num examples / サンプル数: {train_dataset_group.num_train_images}")
+    print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
+    print(f"  num epochs / epoch数: {num_train_epochs}")
+    print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
+    print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
+    print(f"  gradient accumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
+    print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
 
-        train_util.sample_images(accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
+    progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
+    global_step = 0
 
-      current_loss = loss.detach().item()        # 平均なのでbatch sizeは関係ないはず
-      if args.logging_dir is not None:
-        logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
-        if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
-          logs["lr/d*lr"] = lr_scheduler.optimizers[0].param_groups[0]['d']*lr_scheduler.optimizers[0].param_groups[0]['lr']
-        accelerator.log(logs, step=global_step)
+    noise_scheduler = DDPMScheduler(
+        beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000, clip_sample=False
+    )
 
-      # TODO moving averageにする
-      loss_total += current_loss
-      avr_loss = loss_total / (step+1)
-      logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
-      progress_bar.set_postfix(**logs)
+    if accelerator.is_main_process:
+        accelerator.init_trackers("finetuning")
 
-      if global_step >= args.max_train_steps:
-        break
+    for epoch in range(num_train_epochs):
+        print(f"epoch {epoch+1}/{num_train_epochs}")
+        current_epoch.value = epoch + 1
 
-    if args.logging_dir is not None:
-      logs = {"loss/epoch": loss_total / len(train_dataloader)}
-      accelerator.log(logs, step=epoch+1)
+        for m in training_models:
+            m.train()
 
-    accelerator.wait_for_everyone()
+        loss_total = 0
+        for step, batch in enumerate(train_dataloader):
+            current_step.value = global_step
+            with accelerator.accumulate(training_models[0]):  # 複数モデルに対応していない模様だがとりあえずこうしておく
+                with torch.no_grad():
+                    if "latents" in batch and batch["latents"] is not None:
+                        latents = batch["latents"].to(accelerator.device)
+                    else:
+                        # latentに変換
+                        latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
+                    latents = latents * 0.18215
+                b_size = latents.shape[0]
 
-    if args.save_every_n_epochs is not None:
-      src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
-      train_util.save_sd_model_on_epoch_end(args, accelerator, src_path, save_stable_diffusion_format, use_safetensors,
-                                            save_dtype, epoch, num_train_epochs, global_step,  unwrap_model(text_encoder), unwrap_model(unet), vae)
+                with torch.set_grad_enabled(args.train_text_encoder):
+                    # Get the text embedding for conditioning
+                    input_ids = batch["input_ids"].to(accelerator.device)
+                    encoder_hidden_states = train_util.get_hidden_states(
+                        args, input_ids, tokenizer, text_encoder, None if not args.full_fp16 else weight_dtype
+                    )
 
-    train_util.sample_images(accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
-    
-  is_main_process = accelerator.is_main_process
-  if is_main_process:
-    unet = unwrap_model(unet)
-    text_encoder = unwrap_model(text_encoder)
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents, device=latents.device)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
 
-  accelerator.end_training()
+                # Sample a random timestep for each image
+                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
+                timesteps = timesteps.long()
 
-  if args.save_state:
-    train_util.save_state_on_train_end(args, accelerator)
+                # Add noise to the latents according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
 
-  del accelerator                         # この後メモリを使うのでこれは消す
+                # Predict the noise residual
+                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
 
-  if is_main_process:
-    src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
-    train_util.save_sd_model_on_train_end(args, src_path, save_stable_diffusion_format, use_safetensors,
-                                          save_dtype, epoch, global_step,  text_encoder, unet, vae)
-    print("model saved.")
+                if args.v_parameterization:
+                    # v-parameterization training
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    target = noise
+
+                if args.min_snr_gamma:
+                    # do not mean over batch dimension for snr weight
+                    loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
+                    loss = loss.mean([1, 2, 3])
+                    loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma)
+                    loss = loss.mean()  # mean over batch dimension
+                else:
+                    loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="mean")
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients and args.max_grad_norm != 0.0:
+                    params_to_clip = []
+                    for m in training_models:
+                        params_to_clip.extend(m.parameters())
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=True)
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+
+                train_util.sample_images(
+                    accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet
+                )
+
+            current_loss = loss.detach().item()  # 平均なのでbatch sizeは関係ないはず
+            if args.logging_dir is not None:
+                logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
+                if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
+                    logs["lr/d*lr"] = (
+                        lr_scheduler.optimizers[0].param_groups[0]["d"] * lr_scheduler.optimizers[0].param_groups[0]["lr"]
+                    )
+                accelerator.log(logs, step=global_step)
+
+            # TODO moving averageにする
+            loss_total += current_loss
+            avr_loss = loss_total / (step + 1)
+            logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if args.logging_dir is not None:
+            logs = {"loss/epoch": loss_total / len(train_dataloader)}
+            accelerator.log(logs, step=epoch + 1)
+
+        accelerator.wait_for_everyone()
+
+        if args.save_every_n_epochs is not None:
+            src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
+            train_util.save_sd_model_on_epoch_end(
+                args,
+                accelerator,
+                src_path,
+                save_stable_diffusion_format,
+                use_safetensors,
+                save_dtype,
+                epoch,
+                num_train_epochs,
+                global_step,
+                unwrap_model(text_encoder),
+                unwrap_model(unet),
+                vae,
+            )
+
+        train_util.sample_images(accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
+
+    is_main_process = accelerator.is_main_process
+    if is_main_process:
+        unet = unwrap_model(unet)
+        text_encoder = unwrap_model(text_encoder)
+
+    accelerator.end_training()
+
+    if args.save_state:
+        train_util.save_state_on_train_end(args, accelerator)
+
+    del accelerator  # この後メモリを使うのでこれは消す
+
+    if is_main_process:
+        src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
+        train_util.save_sd_model_on_train_end(
+            args, src_path, save_stable_diffusion_format, use_safetensors, save_dtype, epoch, global_step, text_encoder, unet, vae
+        )
+        print("model saved.")
 
 
-if __name__ == '__main__':
-  parser = argparse.ArgumentParser()
+def setup_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser()
 
-  train_util.add_sd_models_arguments(parser)
-  train_util.add_dataset_arguments(parser, False, True, True)
-  train_util.add_training_arguments(parser, False)
-  train_util.add_sd_saving_arguments(parser)
-  train_util.add_optimizer_arguments(parser)
-  config_util.add_config_arguments(parser)
+    train_util.add_sd_models_arguments(parser)
+    train_util.add_dataset_arguments(parser, False, True, True)
+    train_util.add_training_arguments(parser, False)
+    train_util.add_sd_saving_arguments(parser)
+    train_util.add_optimizer_arguments(parser)
+    config_util.add_config_arguments(parser)
+    custom_train_functions.add_custom_train_arguments(parser)
 
-  parser.add_argument("--diffusers_xformers", action='store_true',
-                      help='use xformers by diffusers / Diffusersでxformersを使用する')
-  parser.add_argument("--train_text_encoder", action="store_true", help="train text encoder / text encoderも学習する")
+    parser.add_argument("--diffusers_xformers", action="store_true", help="use xformers by diffusers / Diffusersでxformersを使用する")
+    parser.add_argument("--train_text_encoder", action="store_true", help="train text encoder / text encoderも学習する")
 
-  args = parser.parse_args()
-  train(args)
+    return parser
+
+
+if __name__ == "__main__":
+    parser = setup_parser()
+
+    args = parser.parse_args()
+    args = train_util.read_config_from_file(args, parser)
+
+    train(args)

finetune/clean_captions_and_tags.py

@@ -163,13 +163,19 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   # parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("in_json", type=str, help="metadata file to input / 読み込むメタデータファイル")
   parser.add_argument("out_json", type=str, help="metadata file to output / メタデータファイル書き出し先")
   parser.add_argument("--debug", action="store_true", help="debug mode")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args, unknown = parser.parse_known_args()
   if len(unknown) == 1:
     print("WARNING: train_data_dir argument is removed. This script will not work with three arguments in future. Please specify two arguments: in_json and out_json.")

finetune/make_captions.py

@@ -133,7 +133,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("--caption_weights", type=str, default="https://storage.googleapis.com/sfr-vision-language-research/BLIP/models/model_large_caption.pth",
@@ -153,6 +153,12 @@ if __name__ == '__main__':
   parser.add_argument('--seed', default=42, type=int, help='seed for reproducibility / 再現性を確保するための乱数seed')
   parser.add_argument("--debug", action="store_true", help="debug mode")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
 
   # スペルミスしていたオプションを復元する

finetune/make_captions_by_git.py

@@ -127,7 +127,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("--caption_extension", type=str, default=".caption", help="extension of caption file / 出力されるキャプションファイルの拡張子")
@@ -141,5 +141,11 @@ if __name__ == '__main__':
                       help="remove like `with the words xxx` from caption / `with the words xxx`のような部分をキャプションから削除する")
   parser.add_argument("--debug", action="store_true", help="debug mode")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   main(args)

finetune/merge_captions_to_metadata.py

@@ -4,7 +4,7 @@ from pathlib import Path
 from typing import List
 from tqdm import tqdm
 import library.train_util as train_util
-
+import os
 
 def main(args):
   assert not args.recursive or (args.recursive and args.full_path), "recursive requires full_path / recursiveはfull_pathと同時に指定してください"
@@ -29,6 +29,9 @@ def main(args):
     caption_path = image_path.with_suffix(args.caption_extension)
     caption = caption_path.read_text(encoding='utf-8').strip()
 
+    if not os.path.exists(caption_path):
+      caption_path = os.path.join(image_path, args.caption_extension)
+
     image_key = str(image_path) if args.full_path else image_path.stem
     if image_key not in metadata:
       metadata[image_key] = {}
@@ -43,7 +46,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("out_json", type=str, help="metadata file to output / メタデータファイル書き出し先")
@@ -58,6 +61,12 @@ if __name__ == '__main__':
                       help="recursively look for training tags in all child folders of train_data_dir / train_data_dirのすべての子フォルダにある学習タグを再帰的に探す")
   parser.add_argument("--debug", action="store_true", help="debug mode")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
 
   # スペルミスしていたオプションを復元する

finetune/merge_dd_tags_to_metadata.py

@@ -4,7 +4,7 @@ from pathlib import Path
 from typing import List
 from tqdm import tqdm
 import library.train_util as train_util
-
+import os
 
 def main(args):
   assert not args.recursive or (args.recursive and args.full_path), "recursive requires full_path / recursiveはfull_pathと同時に指定してください"
@@ -29,6 +29,9 @@ def main(args):
     tags_path = image_path.with_suffix(args.caption_extension)
     tags = tags_path.read_text(encoding='utf-8').strip()
 
+    if not os.path.exists(tags_path):
+      tags_path = os.path.join(image_path, args.caption_extension)
+
     image_key = str(image_path) if args.full_path else image_path.stem
     if image_key not in metadata:
       metadata[image_key] = {}
@@ -44,7 +47,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("out_json", type=str, help="metadata file to output / メタデータファイル書き出し先")
@@ -58,5 +61,11 @@ if __name__ == '__main__':
                       help="extension of caption (tag) file / 読み込むキャプション（タグ）ファイルの拡張子")
   parser.add_argument("--debug", action="store_true", help="debug mode, print tags")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   main(args)

finetune/prepare_buckets_latents.py

@@ -229,7 +229,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("in_json", type=str, help="metadata file to input / 読み込むメタデータファイル")
@@ -257,5 +257,11 @@ if __name__ == '__main__':
   parser.add_argument("--skip_existing", action="store_true",
                       help="skip images if npz already exists (both normal and flipped exists if flip_aug is enabled) / npzが既に存在する画像をスキップする（flip_aug有効時は通常、反転の両方が存在する画像をスキップ）")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   main(args)

finetune/tag_images_by_wd14_tagger.py

@@ -173,7 +173,7 @@ def main(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("train_data_dir", type=str, help="directory for train images / 学習画像データのディレクトリ")
   parser.add_argument("--repo_id", type=str, default=DEFAULT_WD14_TAGGER_REPO,
@@ -191,6 +191,12 @@ if __name__ == '__main__':
   parser.add_argument("--caption_extension", type=str, default=".txt", help="extension of caption file / 出力されるキャプションファイルの拡張子")
   parser.add_argument("--debug", action="store_true", help="debug mode")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
 
   # スペルミスしていたオプションを復元する

library/config_util.py

@@ -4,6 +4,7 @@ from dataclasses import (
   dataclass,
 )
 import functools
+import random
 from textwrap import dedent, indent
 import json
 from pathlib import Path
@@ -56,6 +57,8 @@ class BaseSubsetParams:
   caption_dropout_rate: float = 0.0
   caption_dropout_every_n_epochs: int = 0
   caption_tag_dropout_rate: float = 0.0
+  token_warmup_min: int = 1
+  token_warmup_step: float = 0
 
 @dataclass
 class DreamBoothSubsetParams(BaseSubsetParams):
@@ -137,6 +140,8 @@ class ConfigSanitizer:
     "random_crop": bool,
     "shuffle_caption": bool,
     "keep_tokens": int,
+    "token_warmup_min": int,
+    "token_warmup_step": Any(float,int),
   }
   # DO means DropOut
   DO_SUBSET_ASCENDABLE_SCHEMA = {
@@ -406,6 +411,8 @@ def generate_dataset_group_by_blueprint(dataset_group_blueprint: DatasetGroupBlu
           flip_aug: {subset.flip_aug}
           face_crop_aug_range: {subset.face_crop_aug_range}
           random_crop: {subset.random_crop}
+          token_warmup_min: {subset.token_warmup_min},
+          token_warmup_step: {subset.token_warmup_step},
       """), "  ")
 
       if is_dreambooth:
@@ -422,9 +429,12 @@ def generate_dataset_group_by_blueprint(dataset_group_blueprint: DatasetGroupBlu
   print(info)
 
   # make buckets first because it determines the length of dataset
+  # and set the same seed for all datasets
+  seed = random.randint(0, 2**31) # actual seed is seed + epoch_no
   for i, dataset in enumerate(datasets):
     print(f"[Dataset {i}]")
     dataset.make_buckets()
+    dataset.set_seed(seed)
 
   return DatasetGroup(datasets)
 
@@ -491,7 +501,6 @@ def load_user_config(file: str) -> dict:
 
   return config
 
-
 # for config test
 if __name__ == "__main__":
   parser = argparse.ArgumentParser()

library/custom_train_functions.py

@@ -0,0 +1,18 @@
+import torch
+import argparse
+
+def apply_snr_weight(loss, timesteps, noise_scheduler, gamma): 
+  alphas_cumprod = noise_scheduler.alphas_cumprod
+  sqrt_alphas_cumprod = torch.sqrt(alphas_cumprod)
+  sqrt_one_minus_alphas_cumprod = torch.sqrt(1.0 - alphas_cumprod)
+  alpha = sqrt_alphas_cumprod
+  sigma = sqrt_one_minus_alphas_cumprod
+  all_snr = (alpha / sigma) ** 2
+  snr = torch.stack([all_snr[t] for t in timesteps])
+  gamma_over_snr = torch.div(torch.ones_like(snr)*gamma,snr)
+  snr_weight = torch.minimum(gamma_over_snr,torch.ones_like(gamma_over_snr)).float() #from paper
+  loss = loss * snr_weight
+  return loss
+
+def add_custom_train_arguments(parser: argparse.ArgumentParser):
+  parser.add_argument("--min_snr_gamma", type=float, default=None, help="gamma for reducing the weight of high loss timesteps. Lower numbers have stronger effect. 5 is recommended by paper. / 低いタイムステップでの高いlossに対して重みを減らすためのgamma値、低いほど効果が強く、論文では5が推奨")

networks/check_lora_weights.py

@@ -24,9 +24,16 @@ def main(file):
     print(f"{key},{str(tuple(value.size())).replace(', ', '-')},{torch.mean(torch.abs(value))},{torch.min(torch.abs(value))}")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("file", type=str, help="model file to check / 重みを確認するモデルファイル")
+
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
 
   main(args.file)

networks/extract_lora_from_models.py

@@ -162,7 +162,7 @@ def svd(args):
   print(f"LoRA weights are saved to: {args.save_to}")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--v2", action='store_true',
                       help='load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む')
@@ -179,5 +179,11 @@ if __name__ == '__main__':
                       help="dimension (rank) of LoRA for Conv2d-3x3 (default None, disabled) / LoRAのConv2d-3x3の次元数（rank）（デフォルトNone、適用なし）")
   parser.add_argument("--device", type=str, default=None, help="device to use, cuda for GPU / 計算を行うデバイス、cuda でGPUを使う")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   svd(args)

networks/lora.py

@@ -13,386 +13,471 @@ from library import train_util
 
 
 class LoRAModule(torch.nn.Module):
-  """
-  replaces forward method of the original Linear, instead of replacing the original Linear module.
-  """
+    """
+    replaces forward method of the original Linear, instead of replacing the original Linear module.
+    """
 
-  def __init__(self, lora_name, org_module: torch.nn.Module, multiplier=1.0, lora_dim=4, alpha=1):
-    """ if alpha == 0 or None, alpha is rank (no scaling). """
-    super().__init__()
-    self.lora_name = lora_name
+    def __init__(self, lora_name, org_module: torch.nn.Module, multiplier=1.0, lora_dim=4, alpha=1):
+        """if alpha == 0 or None, alpha is rank (no scaling)."""
+        super().__init__()
+        self.lora_name = lora_name
 
-    if org_module.__class__.__name__ == 'Conv2d':
-      in_dim = org_module.in_channels
-      out_dim = org_module.out_channels
-    else:
-      in_dim = org_module.in_features
-      out_dim = org_module.out_features
+        if org_module.__class__.__name__ == "Conv2d":
+            in_dim = org_module.in_channels
+            out_dim = org_module.out_channels
+        else:
+            in_dim = org_module.in_features
+            out_dim = org_module.out_features
 
-    # if limit_rank:
-    #   self.lora_dim = min(lora_dim, in_dim, out_dim)
-    #   if self.lora_dim != lora_dim:
-    #     print(f"{lora_name} dim (rank) is changed to: {self.lora_dim}")
-    # else:
-    self.lora_dim = lora_dim
+        # if limit_rank:
+        #   self.lora_dim = min(lora_dim, in_dim, out_dim)
+        #   if self.lora_dim != lora_dim:
+        #     print(f"{lora_name} dim (rank) is changed to: {self.lora_dim}")
+        # else:
+        self.lora_dim = lora_dim
 
-    if org_module.__class__.__name__ == 'Conv2d':
-      kernel_size = org_module.kernel_size
-      stride = org_module.stride
-      padding = org_module.padding
-      self.lora_down = torch.nn.Conv2d(in_dim, self.lora_dim, kernel_size, stride, padding, bias=False)
-      self.lora_up = torch.nn.Conv2d(self.lora_dim, out_dim, (1, 1), (1, 1), bias=False)
-    else:
-      self.lora_down = torch.nn.Linear(in_dim, self.lora_dim, bias=False)
-      self.lora_up = torch.nn.Linear(self.lora_dim, out_dim, bias=False)
+        if org_module.__class__.__name__ == "Conv2d":
+            kernel_size = org_module.kernel_size
+            stride = org_module.stride
+            padding = org_module.padding
+            self.lora_down = torch.nn.Conv2d(in_dim, self.lora_dim, kernel_size, stride, padding, bias=False)
+            self.lora_up = torch.nn.Conv2d(self.lora_dim, out_dim, (1, 1), (1, 1), bias=False)
+        else:
+            self.lora_down = torch.nn.Linear(in_dim, self.lora_dim, bias=False)
+            self.lora_up = torch.nn.Linear(self.lora_dim, out_dim, bias=False)
 
-    if type(alpha) == torch.Tensor:
-      alpha = alpha.detach().float().numpy()                              # without casting, bf16 causes error
-    alpha = self.lora_dim if alpha is None or alpha == 0 else alpha
-    self.scale = alpha / self.lora_dim
-    self.register_buffer('alpha', torch.tensor(alpha))                    # 定数として扱える
+        if type(alpha) == torch.Tensor:
+            alpha = alpha.detach().float().numpy()  # without casting, bf16 causes error
+        alpha = self.lora_dim if alpha is None or alpha == 0 else alpha
+        self.scale = alpha / self.lora_dim
+        self.register_buffer("alpha", torch.tensor(alpha))  # 定数として扱える
 
-    # same as microsoft's
-    torch.nn.init.kaiming_uniform_(self.lora_down.weight, a=math.sqrt(5))
-    torch.nn.init.zeros_(self.lora_up.weight)
+        # same as microsoft's
+        torch.nn.init.kaiming_uniform_(self.lora_down.weight, a=math.sqrt(5))
+        torch.nn.init.zeros_(self.lora_up.weight)
 
-    self.multiplier = multiplier
-    self.org_module = org_module                  # remove in applying
-    self.region = None
-    self.region_mask = None
+        self.multiplier = multiplier
+        self.org_module = org_module  # remove in applying
+        self.region = None
+        self.region_mask = None
 
-  def apply_to(self):
-    self.org_forward = self.org_module.forward
-    self.org_module.forward = self.forward
-    del self.org_module
+    def apply_to(self):
+        self.org_forward = self.org_module.forward
+        self.org_module.forward = self.forward
+        del self.org_module
 
-  def set_region(self, region):
-    self.region = region
-    self.region_mask = None
+    def merge_to(self, sd, dtype, device):
+        # get up/down weight
+        up_weight = sd["lora_up.weight"].to(torch.float).to(device)
+        down_weight = sd["lora_down.weight"].to(torch.float).to(device)
 
-  def forward(self, x):
-    if self.region is None:
-      return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale
+        # extract weight from org_module
+        org_sd = self.org_module.state_dict()
+        weight = org_sd["weight"].to(torch.float)
 
-    # regional LoRA   FIXME same as additional-network extension
-    if x.size()[1] % 77 == 0:
-      # print(f"LoRA for context: {self.lora_name}")
-      self.region = None
-      return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale
+        # merge weight
+        if len(weight.size()) == 2:
+            # linear
+            weight = weight + self.multiplier * (up_weight @ down_weight) * self.scale
+        elif down_weight.size()[2:4] == (1, 1):
+            # conv2d 1x1
+            weight = (
+                weight
+                + self.multiplier
+                * (up_weight.squeeze(3).squeeze(2) @ down_weight.squeeze(3).squeeze(2)).unsqueeze(2).unsqueeze(3)
+                * self.scale
+            )
+        else:
+            # conv2d 3x3
+            conved = torch.nn.functional.conv2d(down_weight.permute(1, 0, 2, 3), up_weight).permute(1, 0, 2, 3)
+            # print(conved.size(), weight.size(), module.stride, module.padding)
+            weight = weight + self.multiplier * conved * self.scale
 
-    # calculate region mask first time
-    if self.region_mask is None:
-      if len(x.size()) == 4:
-        h, w = x.size()[2:4]
-      else:
-        seq_len = x.size()[1]
-        ratio = math.sqrt((self.region.size()[0] * self.region.size()[1]) / seq_len)
-        h = int(self.region.size()[0] / ratio + .5)
-        w = seq_len // h
+        # set weight to org_module
+        org_sd["weight"] = weight.to(dtype)
+        self.org_module.load_state_dict(org_sd)
 
-      r = self.region.to(x.device)
-      if r.dtype == torch.bfloat16:
-        r = r.to(torch.float)
-      r = r.unsqueeze(0).unsqueeze(1)
-      # print(self.lora_name, self.region.size(), x.size(), r.size(), h, w)
-      r = torch.nn.functional.interpolate(r, (h, w), mode='bilinear')
-      r = r.to(x.dtype)
+    def set_region(self, region):
+        self.region = region
+        self.region_mask = None
 
-      if len(x.size()) == 3:
-        r = torch.reshape(r, (1, x.size()[1], -1))
+    def forward(self, x):
+        if self.region is None:
+            return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale
 
-      self.region_mask = r
+        # regional LoRA   FIXME same as additional-network extension
+        if x.size()[1] % 77 == 0:
+            # print(f"LoRA for context: {self.lora_name}")
+            self.region = None
+            return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale
 
-    return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale * self.region_mask
+        # calculate region mask first time
+        if self.region_mask is None:
+            if len(x.size()) == 4:
+                h, w = x.size()[2:4]
+            else:
+                seq_len = x.size()[1]
+                ratio = math.sqrt((self.region.size()[0] * self.region.size()[1]) / seq_len)
+                h = int(self.region.size()[0] / ratio + 0.5)
+                w = seq_len // h
+
+            r = self.region.to(x.device)
+            if r.dtype == torch.bfloat16:
+                r = r.to(torch.float)
+            r = r.unsqueeze(0).unsqueeze(1)
+            # print(self.lora_name, self.region.size(), x.size(), r.size(), h, w)
+            r = torch.nn.functional.interpolate(r, (h, w), mode="bilinear")
+            r = r.to(x.dtype)
+
+            if len(x.size()) == 3:
+                r = torch.reshape(r, (1, x.size()[1], -1))
+
+            self.region_mask = r
+
+        return self.org_forward(x) + self.lora_up(self.lora_down(x)) * self.multiplier * self.scale * self.region_mask
 
 
 def create_network(multiplier, network_dim, network_alpha, vae, text_encoder, unet, **kwargs):
-  if network_dim is None:
-    network_dim = 4                     # default
+    if network_dim is None:
+        network_dim = 4  # default
 
-  # extract dim/alpha for conv2d, and block dim
-  conv_dim = kwargs.get('conv_dim', None)
-  conv_alpha = kwargs.get('conv_alpha', None)
-  if conv_dim is not None:
-    conv_dim = int(conv_dim)
-    if conv_alpha is None:
-      conv_alpha = 1.0
-    else:
-      conv_alpha = float(conv_alpha)
+    # extract dim/alpha for conv2d, and block dim
+    conv_dim = kwargs.get("conv_dim", None)
+    conv_alpha = kwargs.get("conv_alpha", None)
+    if conv_dim is not None:
+        conv_dim = int(conv_dim)
+        if conv_alpha is None:
+            conv_alpha = 1.0
+        else:
+            conv_alpha = float(conv_alpha)
 
-  """
-  block_dims = kwargs.get("block_dims")
-  block_alphas = None
+    """
+    block_dims = kwargs.get("block_dims")
+    block_alphas = None
 
-  if block_dims is not None:
+    if block_dims is not None:
     block_dims = [int(d) for d in block_dims.split(',')]
     assert len(block_dims) == NUM_BLOCKS, f"Number of block dimensions is not same to {NUM_BLOCKS}"
     block_alphas = kwargs.get("block_alphas")
     if block_alphas is None:
-      block_alphas = [1] * len(block_dims)
+        block_alphas = [1] * len(block_dims)
     else:
-      block_alphas = [int(a) for a in block_alphas(',')]
+        block_alphas = [int(a) for a in block_alphas(',')]
     assert len(block_alphas) == NUM_BLOCKS, f"Number of block alphas is not same to {NUM_BLOCKS}"
 
-  conv_block_dims = kwargs.get("conv_block_dims")
-  conv_block_alphas = None
+    conv_block_dims = kwargs.get("conv_block_dims")
+    conv_block_alphas = None
 
-  if conv_block_dims is not None:
+    if conv_block_dims is not None:
     conv_block_dims = [int(d) for d in conv_block_dims.split(',')]
     assert len(conv_block_dims) == NUM_BLOCKS, f"Number of block dimensions is not same to {NUM_BLOCKS}"
     conv_block_alphas = kwargs.get("conv_block_alphas")
     if conv_block_alphas is None:
-      conv_block_alphas = [1] * len(conv_block_dims)
+        conv_block_alphas = [1] * len(conv_block_dims)
     else:
-      conv_block_alphas = [int(a) for a in conv_block_alphas(',')]
+        conv_block_alphas = [int(a) for a in conv_block_alphas(',')]
     assert len(conv_block_alphas) == NUM_BLOCKS, f"Number of block alphas is not same to {NUM_BLOCKS}"
   """
 
-  network = LoRANetwork(text_encoder, unet, multiplier=multiplier, lora_dim=network_dim,
-                        alpha=network_alpha, conv_lora_dim=conv_dim, conv_alpha=conv_alpha)
-  return network
+    network = LoRANetwork(
+        text_encoder,
+        unet,
+        multiplier=multiplier,
+        lora_dim=network_dim,
+        alpha=network_alpha,
+        conv_lora_dim=conv_dim,
+        conv_alpha=conv_alpha,
+    )
+    return network
 
 
 def create_network_from_weights(multiplier, file, vae, text_encoder, unet, weights_sd=None, **kwargs):
-  if weights_sd is None:
-    if os.path.splitext(file)[1] == '.safetensors':
-      from safetensors.torch import load_file, safe_open
-      weights_sd = load_file(file)
-    else:
-      weights_sd = torch.load(file, map_location='cpu')
+    if weights_sd is None:
+        if os.path.splitext(file)[1] == ".safetensors":
+            from safetensors.torch import load_file, safe_open
 
-  # get dim/alpha mapping
-  modules_dim = {}
-  modules_alpha = {}
-  for key, value in weights_sd.items():
-    if '.' not in key:
-      continue
+            weights_sd = load_file(file)
+        else:
+            weights_sd = torch.load(file, map_location="cpu")
 
-    lora_name = key.split('.')[0]
-    if 'alpha' in key:
-      modules_alpha[lora_name] = value
-    elif 'lora_down' in key:
-      dim = value.size()[0]
-      modules_dim[lora_name] = dim
-      # print(lora_name, value.size(), dim)
+    # get dim/alpha mapping
+    modules_dim = {}
+    modules_alpha = {}
+    for key, value in weights_sd.items():
+        if "." not in key:
+            continue
 
-  # support old LoRA without alpha
-  for key in modules_dim.keys():
-    if key not in modules_alpha:
-      modules_alpha = modules_dim[key]
+        lora_name = key.split(".")[0]
+        if "alpha" in key:
+            modules_alpha[lora_name] = value
+        elif "lora_down" in key:
+            dim = value.size()[0]
+            modules_dim[lora_name] = dim
+            # print(lora_name, value.size(), dim)
 
-  network = LoRANetwork(text_encoder, unet, multiplier=multiplier, modules_dim=modules_dim, modules_alpha=modules_alpha)
-  network.weights_sd = weights_sd
-  return network
+    # support old LoRA without alpha
+    for key in modules_dim.keys():
+        if key not in modules_alpha:
+            modules_alpha = modules_dim[key]
+
+    network = LoRANetwork(text_encoder, unet, multiplier=multiplier, modules_dim=modules_dim, modules_alpha=modules_alpha)
+    network.weights_sd = weights_sd
+    return network
 
 
 class LoRANetwork(torch.nn.Module):
-  # is it possible to apply conv_in and conv_out?
-  UNET_TARGET_REPLACE_MODULE = ["Transformer2DModel", "Attention"]
-  UNET_TARGET_REPLACE_MODULE_CONV2D_3X3 = ["ResnetBlock2D", "Downsample2D", "Upsample2D"]
-  TEXT_ENCODER_TARGET_REPLACE_MODULE = ["CLIPAttention", "CLIPMLP"]
-  LORA_PREFIX_UNET = 'lora_unet'
-  LORA_PREFIX_TEXT_ENCODER = 'lora_te'
+    # is it possible to apply conv_in and conv_out?
+    UNET_TARGET_REPLACE_MODULE = ["Transformer2DModel", "Attention"]
+    UNET_TARGET_REPLACE_MODULE_CONV2D_3X3 = ["ResnetBlock2D", "Downsample2D", "Upsample2D"]
+    TEXT_ENCODER_TARGET_REPLACE_MODULE = ["CLIPAttention", "CLIPMLP"]
+    LORA_PREFIX_UNET = "lora_unet"
+    LORA_PREFIX_TEXT_ENCODER = "lora_te"
 
-  def __init__(self, text_encoder, unet, multiplier=1.0, lora_dim=4, alpha=1, conv_lora_dim=None, conv_alpha=None, modules_dim=None, modules_alpha=None) -> None:
-    super().__init__()
-    self.multiplier = multiplier
+    def __init__(
+        self,
+        text_encoder,
+        unet,
+        multiplier=1.0,
+        lora_dim=4,
+        alpha=1,
+        conv_lora_dim=None,
+        conv_alpha=None,
+        modules_dim=None,
+        modules_alpha=None,
+    ) -> None:
+        super().__init__()
+        self.multiplier = multiplier
 
-    self.lora_dim = lora_dim
-    self.alpha = alpha
-    self.conv_lora_dim = conv_lora_dim
-    self.conv_alpha = conv_alpha
+        self.lora_dim = lora_dim
+        self.alpha = alpha
+        self.conv_lora_dim = conv_lora_dim
+        self.conv_alpha = conv_alpha
 
-    if modules_dim is not None:
-      print(f"create LoRA network from weights")
-    else:
-      print(f"create LoRA network. base dim (rank): {lora_dim}, alpha: {alpha}")
+        if modules_dim is not None:
+            print(f"create LoRA network from weights")
+        else:
+            print(f"create LoRA network. base dim (rank): {lora_dim}, alpha: {alpha}")
 
-    self.apply_to_conv2d_3x3 = self.conv_lora_dim is not None
-    if self.apply_to_conv2d_3x3:
-      if self.conv_alpha is None:
-        self.conv_alpha = self.alpha
-      print(f"apply LoRA to Conv2d with kernel size (3,3). dim (rank): {self.conv_lora_dim}, alpha: {self.conv_alpha}")
+        self.apply_to_conv2d_3x3 = self.conv_lora_dim is not None
+        if self.apply_to_conv2d_3x3:
+            if self.conv_alpha is None:
+                self.conv_alpha = self.alpha
+            print(f"apply LoRA to Conv2d with kernel size (3,3). dim (rank): {self.conv_lora_dim}, alpha: {self.conv_alpha}")
 
-    # create module instances
-    def create_modules(prefix, root_module: torch.nn.Module, target_replace_modules) -> List[LoRAModule]:
-      loras = []
-      for name, module in root_module.named_modules():
-        if module.__class__.__name__ in target_replace_modules:
-          # TODO get block index here
-          for child_name, child_module in module.named_modules():
-            is_linear = child_module.__class__.__name__ == "Linear"
-            is_conv2d = child_module.__class__.__name__ == "Conv2d"
-            is_conv2d_1x1 = is_conv2d and child_module.kernel_size == (1, 1)
-            if is_linear or is_conv2d:
-              lora_name = prefix + '.' + name + '.' + child_name
-              lora_name = lora_name.replace('.', '_')
+        # create module instances
+        def create_modules(prefix, root_module: torch.nn.Module, target_replace_modules) -> List[LoRAModule]:
+            loras = []
+            for name, module in root_module.named_modules():
+                if module.__class__.__name__ in target_replace_modules:
+                    # TODO get block index here
+                    for child_name, child_module in module.named_modules():
+                        is_linear = child_module.__class__.__name__ == "Linear"
+                        is_conv2d = child_module.__class__.__name__ == "Conv2d"
+                        is_conv2d_1x1 = is_conv2d and child_module.kernel_size == (1, 1)
+                        if is_linear or is_conv2d:
+                            lora_name = prefix + "." + name + "." + child_name
+                            lora_name = lora_name.replace(".", "_")
 
-              if modules_dim is not None:
-                if lora_name not in modules_dim:
-                  continue                                      # no LoRA module in this weights file
-                dim = modules_dim[lora_name]
-                alpha = modules_alpha[lora_name]
-              else:
-                if is_linear or is_conv2d_1x1:
-                  dim = self.lora_dim
-                  alpha = self.alpha
-                elif self.apply_to_conv2d_3x3:
-                  dim = self.conv_lora_dim
-                  alpha = self.conv_alpha
-                else:
-                  continue
+                            if modules_dim is not None:
+                                if lora_name not in modules_dim:
+                                    continue  # no LoRA module in this weights file
+                                dim = modules_dim[lora_name]
+                                alpha = modules_alpha[lora_name]
+                            else:
+                                if is_linear or is_conv2d_1x1:
+                                    dim = self.lora_dim
+                                    alpha = self.alpha
+                                elif self.apply_to_conv2d_3x3:
+                                    dim = self.conv_lora_dim
+                                    alpha = self.conv_alpha
+                                else:
+                                    continue
 
-              lora = LoRAModule(lora_name, child_module, self.multiplier, dim, alpha)
-              loras.append(lora)
-      return loras
+                            lora = LoRAModule(lora_name, child_module, self.multiplier, dim, alpha)
+                            loras.append(lora)
+            return loras
 
-    self.text_encoder_loras = create_modules(LoRANetwork.LORA_PREFIX_TEXT_ENCODER,
-                                             text_encoder, LoRANetwork.TEXT_ENCODER_TARGET_REPLACE_MODULE)
-    print(f"create LoRA for Text Encoder: {len(self.text_encoder_loras)} modules.")
+        self.text_encoder_loras = create_modules(
+            LoRANetwork.LORA_PREFIX_TEXT_ENCODER, text_encoder, LoRANetwork.TEXT_ENCODER_TARGET_REPLACE_MODULE
+        )
+        print(f"create LoRA for Text Encoder: {len(self.text_encoder_loras)} modules.")
 
-    # extend U-Net target modules if conv2d 3x3 is enabled, or load from weights
-    target_modules = LoRANetwork.UNET_TARGET_REPLACE_MODULE
-    if modules_dim is not None or self.conv_lora_dim is not None:
-      target_modules += LoRANetwork.UNET_TARGET_REPLACE_MODULE_CONV2D_3X3
+        # extend U-Net target modules if conv2d 3x3 is enabled, or load from weights
+        target_modules = LoRANetwork.UNET_TARGET_REPLACE_MODULE
+        if modules_dim is not None or self.conv_lora_dim is not None:
+            target_modules += LoRANetwork.UNET_TARGET_REPLACE_MODULE_CONV2D_3X3
 
-    self.unet_loras = create_modules(LoRANetwork.LORA_PREFIX_UNET, unet, target_modules)
-    print(f"create LoRA for U-Net: {len(self.unet_loras)} modules.")
+        self.unet_loras = create_modules(LoRANetwork.LORA_PREFIX_UNET, unet, target_modules)
+        print(f"create LoRA for U-Net: {len(self.unet_loras)} modules.")
 
-    self.weights_sd = None
+        self.weights_sd = None
 
-    # assertion
-    names = set()
-    for lora in self.text_encoder_loras + self.unet_loras:
-      assert lora.lora_name not in names, f"duplicated lora name: {lora.lora_name}"
-      names.add(lora.lora_name)
+        # assertion
+        names = set()
+        for lora in self.text_encoder_loras + self.unet_loras:
+            assert lora.lora_name not in names, f"duplicated lora name: {lora.lora_name}"
+            names.add(lora.lora_name)
 
-  def set_multiplier(self, multiplier):
-    self.multiplier = multiplier
-    for lora in self.text_encoder_loras + self.unet_loras:
-      lora.multiplier = self.multiplier
+    def set_multiplier(self, multiplier):
+        self.multiplier = multiplier
+        for lora in self.text_encoder_loras + self.unet_loras:
+            lora.multiplier = self.multiplier
 
-  def load_weights(self, file):
-    if os.path.splitext(file)[1] == '.safetensors':
-      from safetensors.torch import load_file, safe_open
-      self.weights_sd = load_file(file)
-    else:
-      self.weights_sd = torch.load(file, map_location='cpu')
+    def load_weights(self, file):
+        if os.path.splitext(file)[1] == ".safetensors":
+            from safetensors.torch import load_file, safe_open
 
-  def apply_to(self, text_encoder, unet, apply_text_encoder=None, apply_unet=None):
-    if self.weights_sd:
-      weights_has_text_encoder = weights_has_unet = False
-      for key in self.weights_sd.keys():
-        if key.startswith(LoRANetwork.LORA_PREFIX_TEXT_ENCODER):
-          weights_has_text_encoder = True
-        elif key.startswith(LoRANetwork.LORA_PREFIX_UNET):
-          weights_has_unet = True
+            self.weights_sd = load_file(file)
+        else:
+            self.weights_sd = torch.load(file, map_location="cpu")
 
-      if apply_text_encoder is None:
-        apply_text_encoder = weights_has_text_encoder
-      else:
-        assert apply_text_encoder == weights_has_text_encoder, f"text encoder weights: {weights_has_text_encoder} but text encoder flag: {apply_text_encoder} / 重みとText Encoderのフラグが矛盾しています"
+    def apply_to(self, text_encoder, unet, apply_text_encoder=None, apply_unet=None):
+        if self.weights_sd:
+            weights_has_text_encoder = weights_has_unet = False
+            for key in self.weights_sd.keys():
+                if key.startswith(LoRANetwork.LORA_PREFIX_TEXT_ENCODER):
+                    weights_has_text_encoder = True
+                elif key.startswith(LoRANetwork.LORA_PREFIX_UNET):
+                    weights_has_unet = True
 
-      if apply_unet is None:
-        apply_unet = weights_has_unet
-      else:
-        assert apply_unet == weights_has_unet, f"u-net weights: {weights_has_unet} but u-net flag: {apply_unet} / 重みとU-Netのフラグが矛盾しています"
-    else:
-      assert apply_text_encoder is not None and apply_unet is not None, f"internal error: flag not set"
+            if apply_text_encoder is None:
+                apply_text_encoder = weights_has_text_encoder
+            else:
+                assert (
+                    apply_text_encoder == weights_has_text_encoder
+                ), f"text encoder weights: {weights_has_text_encoder} but text encoder flag: {apply_text_encoder} / 重みとText Encoderのフラグが矛盾しています"
 
-    if apply_text_encoder:
-      print("enable LoRA for text encoder")
-    else:
-      self.text_encoder_loras = []
+            if apply_unet is None:
+                apply_unet = weights_has_unet
+            else:
+                assert (
+                    apply_unet == weights_has_unet
+                ), f"u-net weights: {weights_has_unet} but u-net flag: {apply_unet} / 重みとU-Netのフラグが矛盾しています"
+        else:
+            assert apply_text_encoder is not None and apply_unet is not None, f"internal error: flag not set"
 
-    if apply_unet:
-      print("enable LoRA for U-Net")
-    else:
-      self.unet_loras = []
+        if apply_text_encoder:
+            print("enable LoRA for text encoder")
+        else:
+            self.text_encoder_loras = []
 
-    for lora in self.text_encoder_loras + self.unet_loras:
-      lora.apply_to()
-      self.add_module(lora.lora_name, lora)
+        if apply_unet:
+            print("enable LoRA for U-Net")
+        else:
+            self.unet_loras = []
 
-    if self.weights_sd:
-      # if some weights are not in state dict, it is ok because initial LoRA does nothing (lora_up is initialized by zeros)
-      info = self.load_state_dict(self.weights_sd, False)
-      print(f"weights are loaded: {info}")
+        for lora in self.text_encoder_loras + self.unet_loras:
+            lora.apply_to()
+            self.add_module(lora.lora_name, lora)
 
-  def enable_gradient_checkpointing(self):
-    # not supported
-    pass
+        if self.weights_sd:
+            # if some weights are not in state dict, it is ok because initial LoRA does nothing (lora_up is initialized by zeros)
+            info = self.load_state_dict(self.weights_sd, False)
+            print(f"weights are loaded: {info}")
 
-  def prepare_optimizer_params(self, text_encoder_lr, unet_lr):
-    def enumerate_params(loras):
-      params = []
-      for lora in loras:
-        params.extend(lora.parameters())
-      return params
+    # TODO refactor to common function with apply_to
+    def merge_to(self, text_encoder, unet, dtype, device):
+        assert self.weights_sd is not None, "weights are not loaded"
 
-    self.requires_grad_(True)
-    all_params = []
+        apply_text_encoder = apply_unet = False
+        for key in self.weights_sd.keys():
+            if key.startswith(LoRANetwork.LORA_PREFIX_TEXT_ENCODER):
+                apply_text_encoder = True
+            elif key.startswith(LoRANetwork.LORA_PREFIX_UNET):
+                apply_unet = True
 
-    if self.text_encoder_loras:
-      param_data = {'params': enumerate_params(self.text_encoder_loras)}
-      if text_encoder_lr is not None:
-        param_data['lr'] = text_encoder_lr
-      all_params.append(param_data)
+        if apply_text_encoder:
+            print("enable LoRA for text encoder")
+        else:
+            self.text_encoder_loras = []
 
-    if self.unet_loras:
-      param_data = {'params': enumerate_params(self.unet_loras)}
-      if unet_lr is not None:
-        param_data['lr'] = unet_lr
-      all_params.append(param_data)
+        if apply_unet:
+            print("enable LoRA for U-Net")
+        else:
+            self.unet_loras = []
 
-    return all_params
+        for lora in self.text_encoder_loras + self.unet_loras:
+            sd_for_lora = {}
+            for key in self.weights_sd.keys():
+                if key.startswith(lora.lora_name):
+                    sd_for_lora[key[len(lora.lora_name) + 1 :]] = self.weights_sd[key]
+            lora.merge_to(sd_for_lora, dtype, device)
+        print(f"weights are merged")
 
-  def prepare_grad_etc(self, text_encoder, unet):
-    self.requires_grad_(True)
+    def enable_gradient_checkpointing(self):
+        # not supported
+        pass
 
-  def on_epoch_start(self, text_encoder, unet):
-    self.train()
+    def prepare_optimizer_params(self, text_encoder_lr, unet_lr):
+        def enumerate_params(loras):
+            params = []
+            for lora in loras:
+                params.extend(lora.parameters())
+            return params
 
-  def get_trainable_params(self):
-    return self.parameters()
+        self.requires_grad_(True)
+        all_params = []
 
-  def save_weights(self, file, dtype, metadata):
-    if metadata is not None and len(metadata) == 0:
-      metadata = None
+        if self.text_encoder_loras:
+            param_data = {"params": enumerate_params(self.text_encoder_loras)}
+            if text_encoder_lr is not None:
+                param_data["lr"] = text_encoder_lr
+            all_params.append(param_data)
 
-    state_dict = self.state_dict()
+        if self.unet_loras:
+            param_data = {"params": enumerate_params(self.unet_loras)}
+            if unet_lr is not None:
+                param_data["lr"] = unet_lr
+            all_params.append(param_data)
 
-    if dtype is not None:
-      for key in list(state_dict.keys()):
-        v = state_dict[key]
-        v = v.detach().clone().to("cpu").to(dtype)
-        state_dict[key] = v
+        return all_params
 
-    if os.path.splitext(file)[1] == '.safetensors':
-      from safetensors.torch import save_file
+    def prepare_grad_etc(self, text_encoder, unet):
+        self.requires_grad_(True)
 
-      # Precalculate model hashes to save time on indexing
-      if metadata is None:
-        metadata = {}
-      model_hash, legacy_hash = train_util.precalculate_safetensors_hashes(state_dict, metadata)
-      metadata["sshs_model_hash"] = model_hash
-      metadata["sshs_legacy_hash"] = legacy_hash
+    def on_epoch_start(self, text_encoder, unet):
+        self.train()
 
-      save_file(state_dict, file, metadata)
-    else:
-      torch.save(state_dict, file)
+    def get_trainable_params(self):
+        return self.parameters()
 
-  @ staticmethod
-  def set_regions(networks, image):
-    image = image.astype(np.float32) / 255.0
-    for i, network in enumerate(networks[:3]):
-      # NOTE: consider averaging overwrapping area
-      region = image[:, :, i]
-      if region.max() == 0:
-        continue
-      region = torch.tensor(region)
-      network.set_region(region)
+    def save_weights(self, file, dtype, metadata):
+        if metadata is not None and len(metadata) == 0:
+            metadata = None
 
-  def set_region(self, region):
-    for lora in self.unet_loras:
-      lora.set_region(region)
+        state_dict = self.state_dict()
+
+        if dtype is not None:
+            for key in list(state_dict.keys()):
+                v = state_dict[key]
+                v = v.detach().clone().to("cpu").to(dtype)
+                state_dict[key] = v
+
+        if os.path.splitext(file)[1] == ".safetensors":
+            from safetensors.torch import save_file
+
+            # Precalculate model hashes to save time on indexing
+            if metadata is None:
+                metadata = {}
+            model_hash, legacy_hash = train_util.precalculate_safetensors_hashes(state_dict, metadata)
+            metadata["sshs_model_hash"] = model_hash
+            metadata["sshs_legacy_hash"] = legacy_hash
+
+            save_file(state_dict, file, metadata)
+        else:
+            torch.save(state_dict, file)
+
+    @staticmethod
+    def set_regions(networks, image):
+        image = image.astype(np.float32) / 255.0
+        for i, network in enumerate(networks[:3]):
+            # NOTE: consider averaging overwrapping area
+            region = image[:, :, i]
+            if region.max() == 0:
+                continue
+            region = torch.tensor(region)
+            network.set_region(region)
+
+    def set_region(self, region):
+        for lora in self.unet_loras:
+            lora.set_region(region)

networks/lora_interrogator.py

@@ -105,7 +105,7 @@ def interrogate(args):
     print(f"[{i:3d}]: {token:5d} {string:<20s}: {diff:.5f}")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--v2", action='store_true',
                       help='load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む')
@@ -118,5 +118,11 @@ if __name__ == '__main__':
   parser.add_argument("--clip_skip", type=int, default=None,
                       help="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）")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   interrogate(args)

networks/merge_lora.py

@@ -1,4 +1,3 @@
-
 import math
 import argparse
 import os
@@ -9,210 +8,236 @@ import lora
 
 
 def load_state_dict(file_name, dtype):
-  if os.path.splitext(file_name)[1] == '.safetensors':
-    sd = load_file(file_name)
-  else:
-    sd = torch.load(file_name, map_location='cpu')
-  for key in list(sd.keys()):
-    if type(sd[key]) == torch.Tensor:
-      sd[key] = sd[key].to(dtype)
-  return sd
+    if os.path.splitext(file_name)[1] == ".safetensors":
+        sd = load_file(file_name)
+    else:
+        sd = torch.load(file_name, map_location="cpu")
+    for key in list(sd.keys()):
+        if type(sd[key]) == torch.Tensor:
+            sd[key] = sd[key].to(dtype)
+    return sd
 
 
 def save_to_file(file_name, model, state_dict, dtype):
-  if dtype is not None:
-    for key in list(state_dict.keys()):
-      if type(state_dict[key]) == torch.Tensor:
-        state_dict[key] = state_dict[key].to(dtype)
+    if dtype is not None:
+        for key in list(state_dict.keys()):
+            if type(state_dict[key]) == torch.Tensor:
+                state_dict[key] = state_dict[key].to(dtype)
 
-  if os.path.splitext(file_name)[1] == '.safetensors':
-    save_file(model, file_name)
-  else:
-    torch.save(model, file_name)
+    if os.path.splitext(file_name)[1] == ".safetensors":
+        save_file(model, file_name)
+    else:
+        torch.save(model, file_name)
 
 
 def merge_to_sd_model(text_encoder, unet, models, ratios, merge_dtype):
-  text_encoder.to(merge_dtype)
-  unet.to(merge_dtype)
+    text_encoder.to(merge_dtype)
+    unet.to(merge_dtype)
 
-  # create module map
-  name_to_module = {}
-  for i, root_module in enumerate([text_encoder, unet]):
-    if i == 0:
-      prefix = lora.LoRANetwork.LORA_PREFIX_TEXT_ENCODER
-      target_replace_modules = lora.LoRANetwork.TEXT_ENCODER_TARGET_REPLACE_MODULE
-    else:
-      prefix = lora.LoRANetwork.LORA_PREFIX_UNET
-      target_replace_modules = lora.LoRANetwork.UNET_TARGET_REPLACE_MODULE
-
-    for name, module in root_module.named_modules():
-      if module.__class__.__name__ in target_replace_modules:
-        for child_name, child_module in module.named_modules():
-          if child_module.__class__.__name__ == "Linear" or child_module.__class__.__name__ == "Conv2d":
-            lora_name = prefix + '.' + name + '.' + child_name
-            lora_name = lora_name.replace('.', '_')
-            name_to_module[lora_name] = child_module
-
-  for model, ratio in zip(models, ratios):
-    print(f"loading: {model}")
-    lora_sd = load_state_dict(model, merge_dtype)
-
-    print(f"merging...")
-    for key in lora_sd.keys():
-      if "lora_down" in key:
-        up_key = key.replace("lora_down", "lora_up")
-        alpha_key = key[:key.index("lora_down")] + 'alpha'
-
-        # find original module for this lora
-        module_name = '.'.join(key.split('.')[:-2])               # remove trailing ".lora_down.weight"
-        if module_name not in name_to_module:
-          print(f"no module found for LoRA weight: {key}")
-          continue
-        module = name_to_module[module_name]
-        # print(f"apply {key} to {module}")
-
-        down_weight = lora_sd[key]
-        up_weight = lora_sd[up_key]
-
-        dim = down_weight.size()[0]
-        alpha = lora_sd.get(alpha_key, dim)
-        scale = alpha / dim
-
-        # W <- W + U * D
-        weight = module.weight
-        # print(module_name, down_weight.size(), up_weight.size())
-        if len(weight.size()) == 2:
-          # linear
-          weight = weight + ratio * (up_weight @ down_weight) * scale
-        elif down_weight.size()[2:4] == (1, 1):
-          # conv2d 1x1
-          weight = weight + ratio * (up_weight.squeeze(3).squeeze(2) @ down_weight.squeeze(3).squeeze(2)
-                                     ).unsqueeze(2).unsqueeze(3) * scale
+    # create module map
+    name_to_module = {}
+    for i, root_module in enumerate([text_encoder, unet]):
+        if i == 0:
+            prefix = lora.LoRANetwork.LORA_PREFIX_TEXT_ENCODER
+            target_replace_modules = lora.LoRANetwork.TEXT_ENCODER_TARGET_REPLACE_MODULE
         else:
-          # conv2d 3x3
-          conved = torch.nn.functional.conv2d(down_weight.permute(1, 0, 2, 3), up_weight).permute(1, 0, 2, 3)
-          # print(conved.size(), weight.size(), module.stride, module.padding)
-          weight = weight + ratio * conved * scale
+            prefix = lora.LoRANetwork.LORA_PREFIX_UNET
+            target_replace_modules = (
+                lora.LoRANetwork.UNET_TARGET_REPLACE_MODULE + lora.LoRANetwork.UNET_TARGET_REPLACE_MODULE_CONV2D_3X3
+            )
 
-        module.weight = torch.nn.Parameter(weight)
+        for name, module in root_module.named_modules():
+            if module.__class__.__name__ in target_replace_modules:
+                for child_name, child_module in module.named_modules():
+                    if child_module.__class__.__name__ == "Linear" or child_module.__class__.__name__ == "Conv2d":
+                        lora_name = prefix + "." + name + "." + child_name
+                        lora_name = lora_name.replace(".", "_")
+                        name_to_module[lora_name] = child_module
+
+    for model, ratio in zip(models, ratios):
+        print(f"loading: {model}")
+        lora_sd = load_state_dict(model, merge_dtype)
+
+        print(f"merging...")
+        for key in lora_sd.keys():
+            if "lora_down" in key:
+                up_key = key.replace("lora_down", "lora_up")
+                alpha_key = key[: key.index("lora_down")] + "alpha"
+
+                # find original module for this lora
+                module_name = ".".join(key.split(".")[:-2])  # remove trailing ".lora_down.weight"
+                if module_name not in name_to_module:
+                    print(f"no module found for LoRA weight: {key}")
+                    continue
+                module = name_to_module[module_name]
+                # print(f"apply {key} to {module}")
+
+                down_weight = lora_sd[key]
+                up_weight = lora_sd[up_key]
+
+                dim = down_weight.size()[0]
+                alpha = lora_sd.get(alpha_key, dim)
+                scale = alpha / dim
+
+                # W <- W + U * D
+                weight = module.weight
+                # print(module_name, down_weight.size(), up_weight.size())
+                if len(weight.size()) == 2:
+                    # linear
+                    weight = weight + ratio * (up_weight @ down_weight) * scale
+                elif down_weight.size()[2:4] == (1, 1):
+                    # conv2d 1x1
+                    weight = (
+                        weight
+                        + ratio
+                        * (up_weight.squeeze(3).squeeze(2) @ down_weight.squeeze(3).squeeze(2)).unsqueeze(2).unsqueeze(3)
+                        * scale
+                    )
+                else:
+                    # conv2d 3x3
+                    conved = torch.nn.functional.conv2d(down_weight.permute(1, 0, 2, 3), up_weight).permute(1, 0, 2, 3)
+                    # print(conved.size(), weight.size(), module.stride, module.padding)
+                    weight = weight + ratio * conved * scale
+
+                module.weight = torch.nn.Parameter(weight)
 
 
 def merge_lora_models(models, ratios, merge_dtype):
-  base_alphas = {}                          # alpha for merged model
-  base_dims = {}
+    base_alphas = {}  # alpha for merged model
+    base_dims = {}
 
-  merged_sd = {}
-  for model, ratio in zip(models, ratios):
-    print(f"loading: {model}")
-    lora_sd = load_state_dict(model, merge_dtype)
+    merged_sd = {}
+    for model, ratio in zip(models, ratios):
+        print(f"loading: {model}")
+        lora_sd = load_state_dict(model, merge_dtype)
 
-    # get alpha and dim
-    alphas = {}                             # alpha for current model
-    dims = {}                               # dims for current model
-    for key in lora_sd.keys():
-      if 'alpha' in key:
-        lora_module_name = key[:key.rfind(".alpha")]
-        alpha = float(lora_sd[key].detach().numpy())
-        alphas[lora_module_name] = alpha
-        if lora_module_name not in base_alphas:
-          base_alphas[lora_module_name] = alpha
-      elif "lora_down" in key:
-        lora_module_name = key[:key.rfind(".lora_down")]
-        dim = lora_sd[key].size()[0]
-        dims[lora_module_name] = dim
-        if lora_module_name not in base_dims:
-          base_dims[lora_module_name] = dim
+        # get alpha and dim
+        alphas = {}  # alpha for current model
+        dims = {}  # dims for current model
+        for key in lora_sd.keys():
+            if "alpha" in key:
+                lora_module_name = key[: key.rfind(".alpha")]
+                alpha = float(lora_sd[key].detach().numpy())
+                alphas[lora_module_name] = alpha
+                if lora_module_name not in base_alphas:
+                    base_alphas[lora_module_name] = alpha
+            elif "lora_down" in key:
+                lora_module_name = key[: key.rfind(".lora_down")]
+                dim = lora_sd[key].size()[0]
+                dims[lora_module_name] = dim
+                if lora_module_name not in base_dims:
+                    base_dims[lora_module_name] = dim
 
-    for lora_module_name in dims.keys():
-      if lora_module_name not in alphas:
-        alpha = dims[lora_module_name]
-        alphas[lora_module_name] = alpha
-        if lora_module_name not in base_alphas:
-          base_alphas[lora_module_name] = alpha
+        for lora_module_name in dims.keys():
+            if lora_module_name not in alphas:
+                alpha = dims[lora_module_name]
+                alphas[lora_module_name] = alpha
+                if lora_module_name not in base_alphas:
+                    base_alphas[lora_module_name] = alpha
 
-    print(f"dim: {list(set(dims.values()))}, alpha: {list(set(alphas.values()))}")
+        print(f"dim: {list(set(dims.values()))}, alpha: {list(set(alphas.values()))}")
 
-    # merge
-    print(f"merging...")
-    for key in lora_sd.keys():
-      if 'alpha' in key:
-        continue
+        # merge
+        print(f"merging...")
+        for key in lora_sd.keys():
+            if "alpha" in key:
+                continue
 
-      lora_module_name = key[:key.rfind(".lora_")]
+            lora_module_name = key[: key.rfind(".lora_")]
 
-      base_alpha = base_alphas[lora_module_name]
-      alpha = alphas[lora_module_name]
+            base_alpha = base_alphas[lora_module_name]
+            alpha = alphas[lora_module_name]
 
-      scale = math.sqrt(alpha / base_alpha) * ratio
+            scale = math.sqrt(alpha / base_alpha) * ratio
 
-      if key in merged_sd:
-        assert merged_sd[key].size() == lora_sd[key].size(
-        ), f"weights shape mismatch merging v1 and v2, different dims? / 重みのサイズが合いません。v1とv2、または次元数の異なるモデルはマージできません"
-        merged_sd[key] = merged_sd[key] + lora_sd[key] * scale
-      else:
-        merged_sd[key] = lora_sd[key] * scale
+            if key in merged_sd:
+                assert (
+                    merged_sd[key].size() == lora_sd[key].size()
+                ), f"weights shape mismatch merging v1 and v2, different dims? / 重みのサイズが合いません。v1とv2、または次元数の異なるモデルはマージできません"
+                merged_sd[key] = merged_sd[key] + lora_sd[key] * scale
+            else:
+                merged_sd[key] = lora_sd[key] * scale
 
-  # set alpha to sd
-  for lora_module_name, alpha in base_alphas.items():
-    key = lora_module_name + ".alpha"
-    merged_sd[key] = torch.tensor(alpha)
+    # set alpha to sd
+    for lora_module_name, alpha in base_alphas.items():
+        key = lora_module_name + ".alpha"
+        merged_sd[key] = torch.tensor(alpha)
 
-  print("merged model")
-  print(f"dim: {list(set(base_dims.values()))}, alpha: {list(set(base_alphas.values()))}")
+    print("merged model")
+    print(f"dim: {list(set(base_dims.values()))}, alpha: {list(set(base_alphas.values()))}")
 
-  return merged_sd
+    return merged_sd
 
 
 def merge(args):
-  assert len(args.models) == len(args.ratios), f"number of models must be equal to number of ratios / モデルの数と重みの数は合わせてください"
+    assert len(args.models) == len(args.ratios), f"number of models must be equal to number of ratios / モデルの数と重みの数は合わせてください"
 
-  def str_to_dtype(p):
-    if p == 'float':
-      return torch.float
-    if p == 'fp16':
-      return torch.float16
-    if p == 'bf16':
-      return torch.bfloat16
-    return None
+    def str_to_dtype(p):
+        if p == "float":
+            return torch.float
+        if p == "fp16":
+            return torch.float16
+        if p == "bf16":
+            return torch.bfloat16
+        return None
 
-  merge_dtype = str_to_dtype(args.precision)
-  save_dtype = str_to_dtype(args.save_precision)
-  if save_dtype is None:
-    save_dtype = merge_dtype
+    merge_dtype = str_to_dtype(args.precision)
+    save_dtype = str_to_dtype(args.save_precision)
+    if save_dtype is None:
+        save_dtype = merge_dtype
 
-  if args.sd_model is not None:
-    print(f"loading SD model: {args.sd_model}")
+    if args.sd_model is not None:
+        print(f"loading SD model: {args.sd_model}")
 
-    text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(args.v2, args.sd_model)
+        text_encoder, vae, unet = model_util.load_models_from_stable_diffusion_checkpoint(args.v2, args.sd_model)
 
-    merge_to_sd_model(text_encoder, unet, args.models, args.ratios, merge_dtype)
+        merge_to_sd_model(text_encoder, unet, args.models, args.ratios, merge_dtype)
 
-    print(f"saving SD model to: {args.save_to}")
-    model_util.save_stable_diffusion_checkpoint(args.v2, args.save_to, text_encoder, unet,
-                                                args.sd_model, 0, 0, save_dtype, vae)
-  else:
-    state_dict = merge_lora_models(args.models, args.ratios, merge_dtype)
+        print(f"saving SD model to: {args.save_to}")
+        model_util.save_stable_diffusion_checkpoint(args.v2, args.save_to, text_encoder, unet, args.sd_model, 0, 0, save_dtype, vae)
+    else:
+        state_dict = merge_lora_models(args.models, args.ratios, merge_dtype)
 
-    print(f"saving model to: {args.save_to}")
-    save_to_file(args.save_to, state_dict, state_dict, save_dtype)
+        print(f"saving model to: {args.save_to}")
+        save_to_file(args.save_to, state_dict, state_dict, save_dtype)
 
 
-if __name__ == '__main__':
-  parser = argparse.ArgumentParser()
-  parser.add_argument("--v2", action='store_true',
-                      help='load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む')
-  parser.add_argument("--save_precision", type=str, default=None,
-                      choices=[None, "float", "fp16", "bf16"], help="precision in saving, same to merging if omitted / 保存時に精度を変更して保存する、省略時はマージ時の精度と同じ")
-  parser.add_argument("--precision", type=str, default="float",
-                      choices=["float", "fp16", "bf16"], help="precision in merging (float is recommended) / マージの計算時の精度（floatを推奨）")
-  parser.add_argument("--sd_model", type=str, default=None,
-                      help="Stable Diffusion model to load: ckpt or safetensors file, merge LoRA models if omitted / 読み込むモデル、ckptまたはsafetensors。省略時はLoRAモデル同士をマージする")
-  parser.add_argument("--save_to", type=str, default=None,
-                      help="destination file name: ckpt or safetensors file / 保存先のファイル名、ckptまたはsafetensors")
-  parser.add_argument("--models", type=str, nargs='*',
-                      help="LoRA models to merge: ckpt or safetensors file / マージするLoRAモデル、ckptまたはsafetensors")
-  parser.add_argument("--ratios", type=float, nargs='*',
-                      help="ratios for each model / それぞれのLoRAモデルの比率")
+def setup_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--v2", action="store_true", help="load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む")
+    parser.add_argument(
+        "--save_precision",
+        type=str,
+        default=None,
+        choices=[None, "float", "fp16", "bf16"],
+        help="precision in saving, same to merging if omitted / 保存時に精度を変更して保存する、省略時はマージ時の精度と同じ",
+    )
+    parser.add_argument(
+        "--precision",
+        type=str,
+        default="float",
+        choices=["float", "fp16", "bf16"],
+        help="precision in merging (float is recommended) / マージの計算時の精度（floatを推奨）",
+    )
+    parser.add_argument(
+        "--sd_model",
+        type=str,
+        default=None,
+        help="Stable Diffusion model to load: ckpt or safetensors file, merge LoRA models if omitted / 読み込むモデル、ckptまたはsafetensors。省略時はLoRAモデル同士をマージする",
+    )
+    parser.add_argument(
+        "--save_to", type=str, default=None, help="destination file name: ckpt or safetensors file / 保存先のファイル名、ckptまたはsafetensors"
+    )
+    parser.add_argument(
+        "--models", type=str, nargs="*", help="LoRA models to merge: ckpt or safetensors file / マージするLoRAモデル、ckptまたはsafetensors"
+    )
+    parser.add_argument("--ratios", type=float, nargs="*", help="ratios for each model / それぞれのLoRAモデルの比率")
 
-  args = parser.parse_args()
-  merge(args)
+    return parser
+
+
+if __name__ == "__main__":
+    parser = setup_parser()
+
+    args = parser.parse_args()
+    merge(args)

networks/merge_lora_old.py

@@ -158,7 +158,7 @@ def merge(args):
     save_to_file(args.save_to, state_dict, state_dict, save_dtype)
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--v2", action='store_true',
                       help='load Stable Diffusion v2.x model / Stable Diffusion 2.xのモデルを読み込む')
@@ -175,5 +175,11 @@ if __name__ == '__main__':
   parser.add_argument("--ratios", type=float, nargs='*',
                       help="ratios for each model / それぞれのLoRAモデルの比率")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   merge(args)

networks/resize_lora.py

@@ -11,6 +11,8 @@ import numpy as np
 
 MIN_SV = 1e-6
 
+# Model save and load functions
+
 def load_state_dict(file_name, dtype):
   if model_util.is_safetensors(file_name):
     sd = load_file(file_name)
@@ -39,12 +41,13 @@ def save_to_file(file_name, model, state_dict, dtype, metadata):
     torch.save(model, file_name)
 
 
+# Indexing functions
+
 def index_sv_cumulative(S, target):
   original_sum = float(torch.sum(S))
   cumulative_sums = torch.cumsum(S, dim=0)/original_sum
   index = int(torch.searchsorted(cumulative_sums, target)) + 1
-  if index >= len(S):
-    index = len(S) - 1
+  index = max(1, min(index, len(S)-1))
 
   return index
 
@@ -54,8 +57,16 @@ def index_sv_fro(S, target):
   s_fro_sq = float(torch.sum(S_squared))
   sum_S_squared = torch.cumsum(S_squared, dim=0)/s_fro_sq
   index = int(torch.searchsorted(sum_S_squared, target**2)) + 1
-  if index >= len(S):
-    index = len(S) - 1
+  index = max(1, min(index, len(S)-1))
+
+  return index
+
+
+def index_sv_ratio(S, target):
+  max_sv = S[0]
+  min_sv = max_sv/target
+  index = int(torch.sum(S > min_sv).item())
+  index = max(1, min(index, len(S)-1))
 
   return index
 
@@ -125,26 +136,24 @@ def merge_linear(lora_down, lora_up, device):
     return weight
   
 
+# Calculate new rank
+
 def rank_resize(S, rank, dynamic_method, dynamic_param, scale=1):
     param_dict = {}
 
     if dynamic_method=="sv_ratio":
         # Calculate new dim and alpha based off ratio
-        max_sv = S[0]
-        min_sv = max_sv/dynamic_param
-        new_rank = max(torch.sum(S > min_sv).item(),1)
+        new_rank = index_sv_ratio(S, dynamic_param) + 1
         new_alpha = float(scale*new_rank)
 
     elif dynamic_method=="sv_cumulative":
         # Calculate new dim and alpha based off cumulative sum
-        new_rank = index_sv_cumulative(S, dynamic_param)
-        new_rank = max(new_rank, 1)
+        new_rank = index_sv_cumulative(S, dynamic_param) + 1
         new_alpha = float(scale*new_rank)
 
     elif dynamic_method=="sv_fro":
         # Calculate new dim and alpha based off sqrt sum of squares
-        new_rank = index_sv_fro(S, dynamic_param)
-        new_rank = min(max(new_rank, 1), len(S)-1)
+        new_rank = index_sv_fro(S, dynamic_param) + 1
         new_alpha = float(scale*new_rank)
     else:
         new_rank = rank
@@ -172,7 +181,7 @@ def rank_resize(S, rank, dynamic_method, dynamic_param, scale=1):
     param_dict["new_alpha"] = new_alpha
     param_dict["sum_retained"] = (s_rank)/s_sum
     param_dict["fro_retained"] = fro_percent
-    param_dict["max_ratio"] = S[0]/S[new_rank]
+    param_dict["max_ratio"] = S[0]/S[new_rank - 1]
 
     return param_dict
 
@@ -208,18 +217,28 @@ def resize_lora_model(lora_sd, new_rank, save_dtype, device, dynamic_method, dyn
 
   with torch.no_grad():
     for key, value in tqdm(lora_sd.items()):
+      weight_name = None
       if 'lora_down' in key:
         block_down_name = key.split(".")[0]
+        weight_name = key.split(".")[-1]
         lora_down_weight = value
-      if 'lora_up' in key:
-        block_up_name = key.split(".")[0]
-        lora_up_weight = value
+      else:
+        continue
+
+      # find corresponding lora_up and alpha
+      block_up_name = block_down_name
+      lora_up_weight = lora_sd.get(block_up_name + '.lora_up.' + weight_name, None)
+      lora_alpha = lora_sd.get(block_down_name + '.alpha', None)
 
       weights_loaded = (lora_down_weight is not None and lora_up_weight is not None)
 
-      if (block_down_name == block_up_name) and weights_loaded:
+      if weights_loaded:
 
         conv2d = (len(lora_down_weight.size()) == 4)
+        if lora_alpha is None:
+          scale = 1.0
+        else:
+          scale = lora_alpha/lora_down_weight.size()[0]
 
         if conv2d:
           full_weight_matrix = merge_conv(lora_down_weight, lora_up_weight, device)
@@ -311,7 +330,7 @@ def resize(args):
   save_to_file(args.save_to, state_dict, state_dict, save_dtype, metadata)
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
 
   parser.add_argument("--save_precision", type=str, default=None,
@@ -329,7 +348,12 @@ if __name__ == '__main__':
                       help="Specify dynamic resizing method, --new_rank is used as a hard limit for max rank")
   parser.add_argument("--dynamic_param", type=float, default=None,
                       help="Specify target for dynamic reduction")
-                                           
+       
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
 
   args = parser.parse_args()
   resize(args)

networks/svd_merge_lora.py

@@ -77,6 +77,10 @@ def merge_lora_models(models, ratios, new_rank, new_conv_rank, device, merge_dty
 
       # W <- W + U * D
       scale = (alpha / network_dim)
+
+      if device:                      # and isinstance(scale, torch.Tensor):
+        scale = scale.to(device)
+
       if not conv2d:        # linear
         weight = weight + ratio * (up_weight @ down_weight) * scale
       elif kernel_size == (1, 1):
@@ -160,7 +164,7 @@ def merge(args):
   save_to_file(args.save_to, state_dict, save_dtype)
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--save_precision", type=str, default=None,
                       choices=[None, "float", "fp16", "bf16"], help="precision in saving, same to merging if omitted / 保存時に精度を変更して保存する、省略時はマージ時の精度と同じ")
@@ -178,5 +182,11 @@ if __name__ == '__main__':
                       help="Specify rank of output LoRA for Conv2d 3x3, None for same as new_rank / 出力するConv2D 3x3 LoRAのrank (dim)、Noneでnew_rankと同じ")
   parser.add_argument("--device", type=str, default=None, help="device to use, cuda for GPU / 計算を行うデバイス、cuda でGPUを使う")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   merge(args)

tools/canny.py

@@ -13,12 +13,18 @@ def canny(args):
   print("done!")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--input", type=str, default=None, help="input path")
   parser.add_argument("--output", type=str, default=None, help="output path")
   parser.add_argument("--thres1", type=int, default=32, help="thres1")
   parser.add_argument("--thres2", type=int, default=224, help="thres2")
 
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
   canny(args)

tools/convert_diffusers20_original_sd.py

@@ -61,7 +61,7 @@ def convert(args):
     print(f"model saved.")
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--v1", action='store_true',
                       help='load v1.x model (v1 or v2 is required to load checkpoint) / 1.xのモデルを読み込む')
@@ -84,6 +84,11 @@ if __name__ == '__main__':
                       help="model to load: checkpoint file or Diffusers model's directory / 読み込むモデル、checkpointかDiffusers形式モデルのディレクトリ")
   parser.add_argument("model_to_save", type=str, default=None,
                       help="model to save: checkpoint (with extension) or Diffusers model's directory (without extension) / 変換後のモデル、拡張子がある場合はcheckpoint、ない場合はDiffusesモデルとして保存")
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
 
   args = parser.parse_args()
   convert(args)

tools/detect_face_rotate.py

@@ -214,7 +214,7 @@ def process(args):
         buf.tofile(f)
 
 
-if __name__ == '__main__':
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser()
   parser.add_argument("--src_dir", type=str, help="directory to load images / 画像を読み込むディレクトリ")
   parser.add_argument("--dst_dir", type=str, help="directory to save images / 画像を保存するディレクトリ")
@@ -234,6 +234,13 @@ if __name__ == '__main__':
   parser.add_argument("--multiple_faces", action="store_true",
                       help="output each faces / 複数の顔が見つかった場合、それぞれを切り出す")
   parser.add_argument("--debug", action="store_true", help="render rect for face / 処理後画像の顔位置に矩形を描画します")
+
+  return parser
+
+
+if __name__ == '__main__':
+  parser = setup_parser()
+
   args = parser.parse_args()
 
   process(args)

tools/resize_images_to_resolution.py

@@ -98,7 +98,7 @@ def resize_images(src_img_folder, dst_img_folder, max_resolution="512x512", divi
           shutil.copy(os.path.join(src_img_folder, asoc_file), os.path.join(dst_img_folder, new_asoc_file))
 
 
-def main():
+def setup_parser() -> argparse.ArgumentParser:
   parser = argparse.ArgumentParser(
       description='Resize images in a folder to a specified max resolution(s) / 指定されたフォルダ内の画像を指定した最大画像サイズ（面積）以下にアスペクト比を維持したままリサイズします')
   parser.add_argument('src_img_folder', type=str, help='Source folder containing the images / 元画像のフォルダ')
@@ -113,6 +113,12 @@ def main():
   parser.add_argument('--copy_associated_files', action='store_true',
                       help='Copy files with same base name to images (captions etc) / 画像と同じファイル名（拡張子を除く）のファイルもコピーする')
 
+  return parser
+
+
+def main():
+  parser = setup_parser()
+
   args = parser.parse_args()
   resize_images(args.src_img_folder, args.dst_img_folder, args.max_resolution,
                 args.divisible_by, args.interpolation, args.save_as_png, args.copy_associated_files)

train_db.py

@@ -7,6 +7,8 @@ import argparse
 import itertools
 import math
 import os
+import toml
+from multiprocessing import Value
 
 from tqdm import tqdm
 import torch
@@ -17,348 +19,411 @@ from diffusers import DDPMScheduler
 import library.train_util as train_util
 import library.config_util as config_util
 from library.config_util import (
-  ConfigSanitizer,
-  BlueprintGenerator,
+    ConfigSanitizer,
+    BlueprintGenerator,
 )
-
-
-def collate_fn(examples):
-  return examples[0]
+import library.custom_train_functions as custom_train_functions
+from library.custom_train_functions import apply_snr_weight
 
 
 def train(args):
-  train_util.verify_training_args(args)
-  train_util.prepare_dataset_args(args, False)
+    train_util.verify_training_args(args)
+    train_util.prepare_dataset_args(args, False)
 
-  cache_latents = args.cache_latents
+    cache_latents = args.cache_latents
 
-  if args.seed is not None:
-    set_seed(args.seed)                           # 乱数系列を初期化する
+    if args.seed is not None:
+        set_seed(args.seed)  # 乱数系列を初期化する
 
-  tokenizer = train_util.load_tokenizer(args)
+    tokenizer = train_util.load_tokenizer(args)
 
-  blueprint_generator = BlueprintGenerator(ConfigSanitizer(True, False, True))
-  if args.dataset_config is not None:
-    print(f"Load dataset config from {args.dataset_config}")
-    user_config = config_util.load_user_config(args.dataset_config)
-    ignored = ["train_data_dir", "reg_data_dir"]
-    if any(getattr(args, attr) is not None for attr in ignored):
-      print("ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(', '.join(ignored)))
-  else:
-    user_config = {
-      "datasets": [{
-        "subsets": config_util.generate_dreambooth_subsets_config_by_subdirs(args.train_data_dir, args.reg_data_dir)
-      }]
-    }
+    blueprint_generator = BlueprintGenerator(ConfigSanitizer(True, False, True))
+    if args.dataset_config is not None:
+        print(f"Load dataset config from {args.dataset_config}")
+        user_config = config_util.load_user_config(args.dataset_config)
+        ignored = ["train_data_dir", "reg_data_dir"]
+        if any(getattr(args, attr) is not None for attr in ignored):
+            print(
+                "ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(
+                    ", ".join(ignored)
+                )
+            )
+    else:
+        user_config = {
+            "datasets": [
+                {"subsets": config_util.generate_dreambooth_subsets_config_by_subdirs(args.train_data_dir, args.reg_data_dir)}
+            ]
+        }
 
-  blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
-  train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
+    blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
+    train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
 
-  if args.no_token_padding:
-    train_dataset_group.disable_token_padding()
+    current_epoch = Value("i", 0)
+    current_step = Value("i", 0)
+    ds_for_collater = train_dataset_group if args.max_data_loader_n_workers == 0 else None
+    collater = train_util.collater_class(current_epoch, current_step, ds_for_collater)
 
-  if args.debug_dataset:
-    train_util.debug_dataset(train_dataset_group)
-    return
+    if args.no_token_padding:
+        train_dataset_group.disable_token_padding()
 
-  if cache_latents:
-    assert train_dataset_group.is_latent_cacheable(), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
+    if args.debug_dataset:
+        train_util.debug_dataset(train_dataset_group)
+        return
 
-  # acceleratorを準備する
-  print("prepare accelerator")
+    if cache_latents:
+        assert (
+            train_dataset_group.is_latent_cacheable()
+        ), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
 
-  if args.gradient_accumulation_steps > 1:
-    print(f"gradient_accumulation_steps is {args.gradient_accumulation_steps}. accelerate does not support gradient_accumulation_steps when training multiple models (U-Net and Text Encoder), so something might be wrong")
-    print(
-        f"gradient_accumulation_stepsが{args.gradient_accumulation_steps}に設定されています。accelerateは複数モデル（U-NetおよびText Encoder）の学習時にgradient_accumulation_stepsをサポートしていないため結果は未知数です")
+    # acceleratorを準備する
+    print("prepare accelerator")
 
-  accelerator, unwrap_model = train_util.prepare_accelerator(args)
+    if args.gradient_accumulation_steps > 1:
+        print(
+            f"gradient_accumulation_steps is {args.gradient_accumulation_steps}. accelerate does not support gradient_accumulation_steps when training multiple models (U-Net and Text Encoder), so something might be wrong"
+        )
+        print(
+            f"gradient_accumulation_stepsが{args.gradient_accumulation_steps}に設定されています。accelerateは複数モデル（U-NetおよびText Encoder）の学習時にgradient_accumulation_stepsをサポートしていないため結果は未知数です"
+        )
 
-  # mixed precisionに対応した型を用意しておき適宜castする
-  weight_dtype, save_dtype = train_util.prepare_dtype(args)
+    accelerator, unwrap_model = train_util.prepare_accelerator(args)
 
-  # モデルを読み込む
-  text_encoder, vae, unet, load_stable_diffusion_format = train_util.load_target_model(args, weight_dtype)
+    # mixed precisionに対応した型を用意しておき適宜castする
+    weight_dtype, save_dtype = train_util.prepare_dtype(args)
 
-  # verify load/save model formats
-  if load_stable_diffusion_format:
-    src_stable_diffusion_ckpt = args.pretrained_model_name_or_path
-    src_diffusers_model_path = None
-  else:
-    src_stable_diffusion_ckpt = None
-    src_diffusers_model_path = args.pretrained_model_name_or_path
+    # モデルを読み込む
+    text_encoder, vae, unet, load_stable_diffusion_format = train_util.load_target_model(args, weight_dtype)
 
-  if args.save_model_as is None:
-    save_stable_diffusion_format = load_stable_diffusion_format
-    use_safetensors = args.use_safetensors
-  else:
-    save_stable_diffusion_format = args.save_model_as.lower() == 'ckpt' or args.save_model_as.lower() == 'safetensors'
-    use_safetensors = args.use_safetensors or ("safetensors" in args.save_model_as.lower())
+    # verify load/save model formats
+    if load_stable_diffusion_format:
+        src_stable_diffusion_ckpt = args.pretrained_model_name_or_path
+        src_diffusers_model_path = None
+    else:
+        src_stable_diffusion_ckpt = None
+        src_diffusers_model_path = args.pretrained_model_name_or_path
 
-  # モデルに xformers とか memory efficient attention を組み込む
-  train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
+    if args.save_model_as is None:
+        save_stable_diffusion_format = load_stable_diffusion_format
+        use_safetensors = args.use_safetensors
+    else:
+        save_stable_diffusion_format = args.save_model_as.lower() == "ckpt" or args.save_model_as.lower() == "safetensors"
+        use_safetensors = args.use_safetensors or ("safetensors" in args.save_model_as.lower())
 
-  # 学習を準備する
-  if cache_latents:
-    vae.to(accelerator.device, dtype=weight_dtype)
-    vae.requires_grad_(False)
-    vae.eval()
-    with torch.no_grad():
-      train_dataset_group.cache_latents(vae)
-    vae.to("cpu")
-    if torch.cuda.is_available():
-      torch.cuda.empty_cache()
-    gc.collect()
+    # モデルに xformers とか memory efficient attention を組み込む
+    train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
 
-  # 学習を準備する：モデルを適切な状態にする
-  train_text_encoder = args.stop_text_encoder_training is None or args.stop_text_encoder_training >= 0
-  unet.requires_grad_(True)                   # 念のため追加
-  text_encoder.requires_grad_(train_text_encoder)
-  if not train_text_encoder:
-    print("Text Encoder is not trained.")
-
-  if args.gradient_checkpointing:
-    unet.enable_gradient_checkpointing()
-    text_encoder.gradient_checkpointing_enable()
-
-  if not cache_latents:
-    vae.requires_grad_(False)
-    vae.eval()
-    vae.to(accelerator.device, dtype=weight_dtype)
-
-  # 学習に必要なクラスを準備する
-  print("prepare optimizer, data loader etc.")
-  if train_text_encoder:
-    trainable_params = (itertools.chain(unet.parameters(), text_encoder.parameters()))
-  else:
-    trainable_params = unet.parameters()
-
-  _, _, optimizer = train_util.get_optimizer(args, trainable_params)
-
-  # dataloaderを準備する
-  # DataLoaderのプロセス数：0はメインプロセスになる
-  n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)      # cpu_count-1 ただし最大で指定された数まで
-  train_dataloader = torch.utils.data.DataLoader(
-      train_dataset_group, batch_size=1, shuffle=True, collate_fn=collate_fn, num_workers=n_workers, persistent_workers=args.persistent_data_loader_workers)
-
-  # 学習ステップ数を計算する
-  if args.max_train_epochs is not None:
-    args.max_train_steps = args.max_train_epochs * len(train_dataloader)
-    print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
-
-  if args.stop_text_encoder_training is None:
-    args.stop_text_encoder_training = args.max_train_steps + 1                # do not stop until end
-
-  # lr schedulerを用意する TODO gradient_accumulation_stepsの扱いが何かおかしいかもしれない。後で確認する
-  lr_scheduler = train_util.get_scheduler_fix(args.lr_scheduler, optimizer, num_warmup_steps=args.lr_warmup_steps,
-                                              num_training_steps=args.max_train_steps,
-                                              num_cycles=args.lr_scheduler_num_cycles, power=args.lr_scheduler_power)
-
-  # 実験的機能：勾配も含めたfp16学習を行う　モデル全体をfp16にする
-  if args.full_fp16:
-    assert args.mixed_precision == "fp16", "full_fp16 requires mixed precision='fp16' / full_fp16を使う場合はmixed_precision='fp16'を指定してください。"
-    print("enable full fp16 training.")
-    unet.to(weight_dtype)
-    text_encoder.to(weight_dtype)
-
-  # acceleratorがなんかよろしくやってくれるらしい
-  if train_text_encoder:
-    unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-        unet, text_encoder, optimizer, train_dataloader, lr_scheduler)
-  else:
-    unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(unet, optimizer, train_dataloader, lr_scheduler)
-
-  if not train_text_encoder:
-    text_encoder.to(accelerator.device, dtype=weight_dtype)             # to avoid 'cpu' vs 'cuda' error
-
-  # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
-  if args.full_fp16:
-    train_util.patch_accelerator_for_fp16_training(accelerator)
-
-  # resumeする
-  if args.resume is not None:
-    print(f"resume training from state: {args.resume}")
-    accelerator.load_state(args.resume)
-
-  # epoch数を計算する
-  num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-  num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-  if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
-    args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
-
-  # 学習する
-  total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-  print("running training / 学習開始")
-  print(f"  num train images * repeats / 学習画像の数×繰り返し回数: {train_dataset_group.num_train_images}")
-  print(f"  num reg images / 正則化画像の数: {train_dataset_group.num_reg_images}")
-  print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
-  print(f"  num epochs / epoch数: {num_train_epochs}")
-  print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
-  print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
-  print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
-  print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
-
-  progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
-  global_step = 0
-
-  noise_scheduler = DDPMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear",
-                                  num_train_timesteps=1000, clip_sample=False)
-
-  if accelerator.is_main_process:
-    accelerator.init_trackers("dreambooth")
-
-  loss_list = []
-  loss_total = 0.0
-  for epoch in range(num_train_epochs):
-    print(f"epoch {epoch+1}/{num_train_epochs}")
-    train_dataset_group.set_current_epoch(epoch + 1)
-
-    # 指定したステップ数までText Encoderを学習する：epoch最初の状態
-    unet.train()
-    # train==True is required to enable gradient_checkpointing
-    if args.gradient_checkpointing or global_step < args.stop_text_encoder_training:
-      text_encoder.train()
-
-    for step, batch in enumerate(train_dataloader):
-      # 指定したステップ数でText Encoderの学習を止める
-      if global_step == args.stop_text_encoder_training:
-        print(f"stop text encoder training at step {global_step}")
-        if not args.gradient_checkpointing:
-          text_encoder.train(False)
-        text_encoder.requires_grad_(False)
-
-      with accelerator.accumulate(unet):
+    # 学習を準備する
+    if cache_latents:
+        vae.to(accelerator.device, dtype=weight_dtype)
+        vae.requires_grad_(False)
+        vae.eval()
         with torch.no_grad():
-          # latentに変換
-          if cache_latents:
-            latents = batch["latents"].to(accelerator.device)
-          else:
-            latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
-          latents = latents * 0.18215
-        b_size = latents.shape[0]
+            train_dataset_group.cache_latents(vae, args.vae_batch_size)
+        vae.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        gc.collect()
 
-        # Sample noise that we'll add to the latents
-        noise = torch.randn_like(latents, device=latents.device)
-        if args.noise_offset:
-          # https://www.crosslabs.org//blog/diffusion-with-offset-noise
-          noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
+    # 学習を準備する：モデルを適切な状態にする
+    train_text_encoder = args.stop_text_encoder_training is None or args.stop_text_encoder_training >= 0
+    unet.requires_grad_(True)  # 念のため追加
+    text_encoder.requires_grad_(train_text_encoder)
+    if not train_text_encoder:
+        print("Text Encoder is not trained.")
 
-        # Get the text embedding for conditioning
-        with torch.set_grad_enabled(global_step < args.stop_text_encoder_training):
-          input_ids = batch["input_ids"].to(accelerator.device)
-          encoder_hidden_states = train_util.get_hidden_states(
-              args, input_ids, tokenizer, text_encoder, None if not args.full_fp16 else weight_dtype)
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
 
-        # Sample a random timestep for each image
-        timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
-        timesteps = timesteps.long()
+    if not cache_latents:
+        vae.requires_grad_(False)
+        vae.eval()
+        vae.to(accelerator.device, dtype=weight_dtype)
 
-        # Add noise to the latents according to the noise magnitude at each timestep
-        # (this is the forward diffusion process)
-        noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+    # 学習に必要なクラスを準備する
+    print("prepare optimizer, data loader etc.")
+    if train_text_encoder:
+        trainable_params = itertools.chain(unet.parameters(), text_encoder.parameters())
+    else:
+        trainable_params = unet.parameters()
 
-        # Predict the noise residual
-        noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+    _, _, optimizer = train_util.get_optimizer(args, trainable_params)
 
-        if args.v_parameterization:
-          # v-parameterization training
-          target = noise_scheduler.get_velocity(latents, noise, timesteps)
-        else:
-          target = noise
+    # dataloaderを準備する
+    # DataLoaderのプロセス数：0はメインプロセスになる
+    n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)  # cpu_count-1 ただし最大で指定された数まで
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset_group,
+        batch_size=1,
+        shuffle=True,
+        collate_fn=collater,
+        num_workers=n_workers,
+        persistent_workers=args.persistent_data_loader_workers,
+    )
 
-        loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
-        loss = loss.mean([1, 2, 3])
+    # 学習ステップ数を計算する
+    if args.max_train_epochs is not None:
+        args.max_train_steps = args.max_train_epochs * math.ceil(
+            len(train_dataloader) / accelerator.num_processes / args.gradient_accumulation_steps
+        )
+        print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
 
-        loss_weights = batch["loss_weights"]                      # 各sampleごとのweight
-        loss = loss * loss_weights
+    # データセット側にも学習ステップを送信
+    train_dataset_group.set_max_train_steps(args.max_train_steps)
 
-        loss = loss.mean()                # 平均なのでbatch_sizeで割る必要なし
+    if args.stop_text_encoder_training is None:
+        args.stop_text_encoder_training = args.max_train_steps + 1  # do not stop until end
 
-        accelerator.backward(loss)
-        if accelerator.sync_gradients and args.max_grad_norm != 0.0:
-          if train_text_encoder:
-            params_to_clip = (itertools.chain(unet.parameters(), text_encoder.parameters()))
-          else:
-            params_to_clip = unet.parameters()
-          accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+    # lr schedulerを用意する TODO gradient_accumulation_stepsの扱いが何かおかしいかもしれない。後で確認する
+    lr_scheduler = train_util.get_scheduler_fix(args, optimizer, accelerator.num_processes)
 
-        optimizer.step()
-        lr_scheduler.step()
-        optimizer.zero_grad(set_to_none=True)
+    # 実験的機能：勾配も含めたfp16学習を行う　モデル全体をfp16にする
+    if args.full_fp16:
+        assert (
+            args.mixed_precision == "fp16"
+        ), "full_fp16 requires mixed precision='fp16' / full_fp16を使う場合はmixed_precision='fp16'を指定してください。"
+        print("enable full fp16 training.")
+        unet.to(weight_dtype)
+        text_encoder.to(weight_dtype)
 
-      # Checks if the accelerator has performed an optimization step behind the scenes
-      if accelerator.sync_gradients:
-        progress_bar.update(1)
-        global_step += 1
+    # acceleratorがなんかよろしくやってくれるらしい
+    if train_text_encoder:
+        unet, text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+            unet, text_encoder, optimizer, train_dataloader, lr_scheduler
+        )
+    else:
+        unet, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(unet, optimizer, train_dataloader, lr_scheduler)
 
-        train_util.sample_images(accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
+    if not train_text_encoder:
+        text_encoder.to(accelerator.device, dtype=weight_dtype)  # to avoid 'cpu' vs 'cuda' error
 
-      current_loss = loss.detach().item()
-      if args.logging_dir is not None:
-        logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
-        if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
-          logs["lr/d*lr"] = lr_scheduler.optimizers[0].param_groups[0]['d']*lr_scheduler.optimizers[0].param_groups[0]['lr']
-        accelerator.log(logs, step=global_step)
+    # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
+    if args.full_fp16:
+        train_util.patch_accelerator_for_fp16_training(accelerator)
 
-      if epoch == 0:
-        loss_list.append(current_loss)
-      else:
-        loss_total -= loss_list[step]
-        loss_list[step] = current_loss
-      loss_total += current_loss
-      avr_loss = loss_total / len(loss_list)
-      logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
-      progress_bar.set_postfix(**logs)
+    # resumeする
+    if args.resume is not None:
+        print(f"resume training from state: {args.resume}")
+        accelerator.load_state(args.resume)
 
-      if global_step >= args.max_train_steps:
-        break
+    # epoch数を計算する
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+    if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
+        args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
 
-    if args.logging_dir is not None:
-      logs = {"loss/epoch": loss_total / len(loss_list)}
-      accelerator.log(logs, step=epoch+1)
+    # 学習する
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+    print("running training / 学習開始")
+    print(f"  num train images * repeats / 学習画像の数×繰り返し回数: {train_dataset_group.num_train_images}")
+    print(f"  num reg images / 正則化画像の数: {train_dataset_group.num_reg_images}")
+    print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
+    print(f"  num epochs / epoch数: {num_train_epochs}")
+    print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
+    print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
+    print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
+    print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
 
-    accelerator.wait_for_everyone()
+    progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
+    global_step = 0
 
-    if args.save_every_n_epochs is not None:
-      src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
-      train_util.save_sd_model_on_epoch_end(args, accelerator, src_path, save_stable_diffusion_format, use_safetensors,
-                                            save_dtype, epoch, num_train_epochs, global_step,  unwrap_model(text_encoder), unwrap_model(unet), vae)
+    noise_scheduler = DDPMScheduler(
+        beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000, clip_sample=False
+    )
 
-    train_util.sample_images(accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
+    if accelerator.is_main_process:
+        accelerator.init_trackers("dreambooth")
 
-  is_main_process = accelerator.is_main_process
-  if is_main_process:
-    unet = unwrap_model(unet)
-    text_encoder = unwrap_model(text_encoder)
+    loss_list = []
+    loss_total = 0.0
+    for epoch in range(num_train_epochs):
+        print(f"epoch {epoch+1}/{num_train_epochs}")
+        current_epoch.value = epoch + 1
 
-  accelerator.end_training()
+        # 指定したステップ数までText Encoderを学習する：epoch最初の状態
+        unet.train()
+        # train==True is required to enable gradient_checkpointing
+        if args.gradient_checkpointing or global_step < args.stop_text_encoder_training:
+            text_encoder.train()
 
-  if args.save_state:
-    train_util.save_state_on_train_end(args, accelerator)
+        for step, batch in enumerate(train_dataloader):
+            current_step.value = global_step
+            # 指定したステップ数でText Encoderの学習を止める
+            if global_step == args.stop_text_encoder_training:
+                print(f"stop text encoder training at step {global_step}")
+                if not args.gradient_checkpointing:
+                    text_encoder.train(False)
+                text_encoder.requires_grad_(False)
 
-  del accelerator                         # この後メモリを使うのでこれは消す
+            with accelerator.accumulate(unet):
+                with torch.no_grad():
+                    # latentに変換
+                    if cache_latents:
+                        latents = batch["latents"].to(accelerator.device)
+                    else:
+                        latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
+                    latents = latents * 0.18215
+                b_size = latents.shape[0]
 
-  if is_main_process:
-    src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
-    train_util.save_sd_model_on_train_end(args, src_path, save_stable_diffusion_format, use_safetensors,
-                                          save_dtype, epoch, global_step,  text_encoder, unet, vae)
-    print("model saved.")
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents, device=latents.device)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
+
+                # Get the text embedding for conditioning
+                with torch.set_grad_enabled(global_step < args.stop_text_encoder_training):
+                    input_ids = batch["input_ids"].to(accelerator.device)
+                    encoder_hidden_states = train_util.get_hidden_states(
+                        args, input_ids, tokenizer, text_encoder, None if not args.full_fp16 else weight_dtype
+                    )
+
+                # Sample a random timestep for each image
+                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
+                timesteps = timesteps.long()
+
+                # Add noise to the latents according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                # Predict the noise residual
+                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                if args.v_parameterization:
+                    # v-parameterization training
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    target = noise
+
+                loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
+                loss = loss.mean([1, 2, 3])
+
+                loss_weights = batch["loss_weights"]  # 各sampleごとのweight
+                loss = loss * loss_weights
+
+                if args.min_snr_gamma:
+                    loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma)
+
+                loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients and args.max_grad_norm != 0.0:
+                    if train_text_encoder:
+                        params_to_clip = itertools.chain(unet.parameters(), text_encoder.parameters())
+                    else:
+                        params_to_clip = unet.parameters()
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=True)
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+
+                train_util.sample_images(
+                    accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet
+                )
+
+            current_loss = loss.detach().item()
+            if args.logging_dir is not None:
+                logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
+                if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
+                    logs["lr/d*lr"] = (
+                        lr_scheduler.optimizers[0].param_groups[0]["d"] * lr_scheduler.optimizers[0].param_groups[0]["lr"]
+                    )
+                accelerator.log(logs, step=global_step)
+
+            if epoch == 0:
+                loss_list.append(current_loss)
+            else:
+                loss_total -= loss_list[step]
+                loss_list[step] = current_loss
+            loss_total += current_loss
+            avr_loss = loss_total / len(loss_list)
+            logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if args.logging_dir is not None:
+            logs = {"loss/epoch": loss_total / len(loss_list)}
+            accelerator.log(logs, step=epoch + 1)
+
+        accelerator.wait_for_everyone()
+
+        if args.save_every_n_epochs is not None:
+            src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
+            train_util.save_sd_model_on_epoch_end(
+                args,
+                accelerator,
+                src_path,
+                save_stable_diffusion_format,
+                use_safetensors,
+                save_dtype,
+                epoch,
+                num_train_epochs,
+                global_step,
+                unwrap_model(text_encoder),
+                unwrap_model(unet),
+                vae,
+            )
+
+        train_util.sample_images(accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet)
+
+    is_main_process = accelerator.is_main_process
+    if is_main_process:
+        unet = unwrap_model(unet)
+        text_encoder = unwrap_model(text_encoder)
+
+    accelerator.end_training()
+
+    if args.save_state:
+        train_util.save_state_on_train_end(args, accelerator)
+
+    del accelerator  # この後メモリを使うのでこれは消す
+
+    if is_main_process:
+        src_path = src_stable_diffusion_ckpt if save_stable_diffusion_format else src_diffusers_model_path
+        train_util.save_sd_model_on_train_end(
+            args, src_path, save_stable_diffusion_format, use_safetensors, save_dtype, epoch, global_step, text_encoder, unet, vae
+        )
+        print("model saved.")
 
 
-if __name__ == '__main__':
-  parser = argparse.ArgumentParser()
+def setup_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser()
 
-  train_util.add_sd_models_arguments(parser)
-  train_util.add_dataset_arguments(parser, True, False, True)
-  train_util.add_training_arguments(parser, True)
-  train_util.add_sd_saving_arguments(parser)
-  train_util.add_optimizer_arguments(parser)
-  config_util.add_config_arguments(parser)
+    train_util.add_sd_models_arguments(parser)
+    train_util.add_dataset_arguments(parser, True, False, True)
+    train_util.add_training_arguments(parser, True)
+    train_util.add_sd_saving_arguments(parser)
+    train_util.add_optimizer_arguments(parser)
+    config_util.add_config_arguments(parser)
+    custom_train_functions.add_custom_train_arguments(parser)
 
-  parser.add_argument("--no_token_padding", action="store_true",
-                      help="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）")
-  parser.add_argument("--stop_text_encoder_training", type=int, default=None,
-                      help="steps to stop text encoder training, -1 for no training / Text Encoderの学習を止めるステップ数、-1で最初から学習しない")
+    parser.add_argument(
+        "--no_token_padding",
+        action="store_true",
+        help="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）",
+    )
+    parser.add_argument(
+        "--stop_text_encoder_training",
+        type=int,
+        default=None,
+        help="steps to stop text encoder training, -1 for no training / Text Encoderの学習を止めるステップ数、-1で最初から学習しない",
+    )
 
-  args = parser.parse_args()
-  train(args)
+    return parser
+
+
+if __name__ == "__main__":
+    parser = setup_parser()
+
+    args = parser.parse_args()
+    args = train_util.read_config_from_file(args, parser)
+
+    train(args)

train_textual_inversion.py

@@ -3,6 +3,8 @@ import argparse
 import gc
 import math
 import os
+import toml
+from multiprocessing import Value
 
 from tqdm import tqdm
 import torch
@@ -13,9 +15,11 @@ from diffusers import DDPMScheduler
 import library.train_util as train_util
 import library.config_util as config_util
 from library.config_util import (
-  ConfigSanitizer,
-  BlueprintGenerator,
+    ConfigSanitizer,
+    BlueprintGenerator,
 )
+import library.custom_train_functions as custom_train_functions
+from library.custom_train_functions import apply_snr_weight
 
 imagenet_templates_small = [
     "a photo of a {}",
@@ -70,457 +74,517 @@ imagenet_style_templates_small = [
 ]
 
 
-def collate_fn(examples):
-  return examples[0]
-
-
 def train(args):
-  if args.output_name is None:
-    args.output_name = args.token_string
-  use_template = args.use_object_template or args.use_style_template
+    if args.output_name is None:
+        args.output_name = args.token_string
+    use_template = args.use_object_template or args.use_style_template
 
-  train_util.verify_training_args(args)
-  train_util.prepare_dataset_args(args, True)
+    train_util.verify_training_args(args)
+    train_util.prepare_dataset_args(args, True)
 
-  cache_latents = args.cache_latents
+    cache_latents = args.cache_latents
 
-  if args.seed is not None:
-    set_seed(args.seed)
+    if args.seed is not None:
+        set_seed(args.seed)
 
-  tokenizer = train_util.load_tokenizer(args)
+    tokenizer = train_util.load_tokenizer(args)
 
-  # acceleratorを準備する
-  print("prepare accelerator")
-  accelerator, unwrap_model = train_util.prepare_accelerator(args)
+    # acceleratorを準備する
+    print("prepare accelerator")
+    accelerator, unwrap_model = train_util.prepare_accelerator(args)
 
-  # mixed precisionに対応した型を用意しておき適宜castする
-  weight_dtype, save_dtype = train_util.prepare_dtype(args)
+    # mixed precisionに対応した型を用意しておき適宜castする
+    weight_dtype, save_dtype = train_util.prepare_dtype(args)
 
-  # モデルを読み込む
-  text_encoder, vae, unet, _ = train_util.load_target_model(args, weight_dtype)
+    # モデルを読み込む
+    text_encoder, vae, unet, _ = train_util.load_target_model(args, weight_dtype)
 
-  # Convert the init_word to token_id
-  if args.init_word is not None:
-    init_token_ids = tokenizer.encode(args.init_word, add_special_tokens=False)
-    if len(init_token_ids) > 1 and len(init_token_ids) != args.num_vectors_per_token:
-      print(
-          f"token length for init words is not same to num_vectors_per_token, init words is repeated or truncated / 初期化単語のトークン長がnum_vectors_per_tokenと合わないため、繰り返しまたは切り捨てが発生します: length {len(init_token_ids)}")
-  else:
-    init_token_ids = None
-
-  # add new word to tokenizer, count is num_vectors_per_token
-  token_strings = [args.token_string] + [f"{args.token_string}{i+1}" for i in range(args.num_vectors_per_token - 1)]
-  num_added_tokens = tokenizer.add_tokens(token_strings)
-  assert num_added_tokens == args.num_vectors_per_token, f"tokenizer has same word to token string. please use another one / 指定したargs.token_stringは既に存在します。別の単語を使ってください: {args.token_string}"
-
-  token_ids = tokenizer.convert_tokens_to_ids(token_strings)
-  print(f"tokens are added: {token_ids}")
-  assert min(token_ids) == token_ids[0] and token_ids[-1] == token_ids[0] + len(token_ids) - 1, f"token ids is not ordered"
-  assert len(tokenizer) - 1 == token_ids[-1], f"token ids is not end of tokenize: {len(tokenizer)}"
-
-  # Resize the token embeddings as we are adding new special tokens to the tokenizer
-  text_encoder.resize_token_embeddings(len(tokenizer))
-
-  # Initialise the newly added placeholder token with the embeddings of the initializer token
-  token_embeds = text_encoder.get_input_embeddings().weight.data
-  if init_token_ids is not None:
-    for i, token_id in enumerate(token_ids):
-      token_embeds[token_id] = token_embeds[init_token_ids[i % len(init_token_ids)]]
-      # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
-
-  # load weights
-  if args.weights is not None:
-    embeddings = load_weights(args.weights)
-    assert len(token_ids) == len(
-        embeddings), f"num_vectors_per_token is mismatch for weights / 指定した重みとnum_vectors_per_tokenの値が異なります: {len(embeddings)}"
-    # print(token_ids, embeddings.size())
-    for token_id, embedding in zip(token_ids, embeddings):
-      token_embeds[token_id] = embedding
-      # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
-    print(f"weighs loaded")
-
-  print(f"create embeddings for {args.num_vectors_per_token} tokens, for {args.token_string}")
-
-  # データセットを準備する
-  blueprint_generator = BlueprintGenerator(ConfigSanitizer(True, True, False))
-  if args.dataset_config is not None:
-    print(f"Load dataset config from {args.dataset_config}")
-    user_config = config_util.load_user_config(args.dataset_config)
-    ignored = ["train_data_dir", "reg_data_dir", "in_json"]
-    if any(getattr(args, attr) is not None for attr in ignored):
-      print("ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(', '.join(ignored)))
-  else:
-    use_dreambooth_method = args.in_json is None
-    if use_dreambooth_method:
-      print("Use DreamBooth method.")
-      user_config = {
-        "datasets": [{
-          "subsets": config_util.generate_dreambooth_subsets_config_by_subdirs(args.train_data_dir, args.reg_data_dir)
-        }]
-      }
+    # Convert the init_word to token_id
+    if args.init_word is not None:
+        init_token_ids = tokenizer.encode(args.init_word, add_special_tokens=False)
+        if len(init_token_ids) > 1 and len(init_token_ids) != args.num_vectors_per_token:
+            print(
+                f"token length for init words is not same to num_vectors_per_token, init words is repeated or truncated / 初期化単語のトークン長がnum_vectors_per_tokenと合わないため、繰り返しまたは切り捨てが発生します: length {len(init_token_ids)}"
+            )
     else:
-      print("Train with captions.")
-      user_config = {
-        "datasets": [{
-          "subsets": [{
-            "image_dir": args.train_data_dir,
-            "metadata_file": args.in_json,
-          }]
-        }]
-      }
+        init_token_ids = None
 
-  blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
-  train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
+    # add new word to tokenizer, count is num_vectors_per_token
+    token_strings = [args.token_string] + [f"{args.token_string}{i+1}" for i in range(args.num_vectors_per_token - 1)]
+    num_added_tokens = tokenizer.add_tokens(token_strings)
+    assert (
+        num_added_tokens == args.num_vectors_per_token
+    ), f"tokenizer has same word to token string. please use another one / 指定したargs.token_stringは既に存在します。別の単語を使ってください: {args.token_string}"
 
-  # make captions: tokenstring tokenstring1 tokenstring2 ...tokenstringn という文字列に書き換える超乱暴な実装
-  if use_template:
-    print("use template for training captions. is object: {args.use_object_template}")
-    templates = imagenet_templates_small if args.use_object_template else imagenet_style_templates_small
-    replace_to = " ".join(token_strings)
-    captions = []
-    for tmpl in templates:
-      captions.append(tmpl.format(replace_to))
-    train_dataset_group.add_replacement("", captions)
+    token_ids = tokenizer.convert_tokens_to_ids(token_strings)
+    print(f"tokens are added: {token_ids}")
+    assert min(token_ids) == token_ids[0] and token_ids[-1] == token_ids[0] + len(token_ids) - 1, f"token ids is not ordered"
+    assert len(tokenizer) - 1 == token_ids[-1], f"token ids is not end of tokenize: {len(tokenizer)}"
 
-    if args.num_vectors_per_token > 1:
-      prompt_replacement = (args.token_string, replace_to)
+    # Resize the token embeddings as we are adding new special tokens to the tokenizer
+    text_encoder.resize_token_embeddings(len(tokenizer))
+
+    # Initialise the newly added placeholder token with the embeddings of the initializer token
+    token_embeds = text_encoder.get_input_embeddings().weight.data
+    if init_token_ids is not None:
+        for i, token_id in enumerate(token_ids):
+            token_embeds[token_id] = token_embeds[init_token_ids[i % len(init_token_ids)]]
+            # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
+
+    # load weights
+    if args.weights is not None:
+        embeddings = load_weights(args.weights)
+        assert len(token_ids) == len(
+            embeddings
+        ), f"num_vectors_per_token is mismatch for weights / 指定した重みとnum_vectors_per_tokenの値が異なります: {len(embeddings)}"
+        # print(token_ids, embeddings.size())
+        for token_id, embedding in zip(token_ids, embeddings):
+            token_embeds[token_id] = embedding
+            # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
+        print(f"weighs loaded")
+
+    print(f"create embeddings for {args.num_vectors_per_token} tokens, for {args.token_string}")
+
+    # データセットを準備する
+    blueprint_generator = BlueprintGenerator(ConfigSanitizer(True, True, False))
+    if args.dataset_config is not None:
+        print(f"Load dataset config from {args.dataset_config}")
+        user_config = config_util.load_user_config(args.dataset_config)
+        ignored = ["train_data_dir", "reg_data_dir", "in_json"]
+        if any(getattr(args, attr) is not None for attr in ignored):
+            print(
+                "ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(
+                    ", ".join(ignored)
+                )
+            )
     else:
-      prompt_replacement = None
-  else:
-    if args.num_vectors_per_token > 1:
-      replace_to = " ".join(token_strings)
-      train_dataset_group.add_replacement(args.token_string, replace_to)
-      prompt_replacement = (args.token_string, replace_to)
-    else:
-      prompt_replacement = None
-
-  if args.debug_dataset:
-    train_util.debug_dataset(train_dataset_group, show_input_ids=True)
-    return
-  if len(train_dataset_group) == 0:
-    print("No data found. Please verify arguments / 画像がありません。引数指定を確認してください")
-    return
-
-  if cache_latents:
-    assert train_dataset_group.is_latent_cacheable(), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
-
-  # モデルに xformers とか memory efficient attention を組み込む
-  train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
-
-  # 学習を準備する
-  if cache_latents:
-    vae.to(accelerator.device, dtype=weight_dtype)
-    vae.requires_grad_(False)
-    vae.eval()
-    with torch.no_grad():
-      train_dataset_group.cache_latents(vae)
-    vae.to("cpu")
-    if torch.cuda.is_available():
-      torch.cuda.empty_cache()
-    gc.collect()
-
-  if args.gradient_checkpointing:
-    unet.enable_gradient_checkpointing()
-    text_encoder.gradient_checkpointing_enable()
-
-  # 学習に必要なクラスを準備する
-  print("prepare optimizer, data loader etc.")
-  trainable_params = text_encoder.get_input_embeddings().parameters()
-  _, _, optimizer = train_util.get_optimizer(args, trainable_params)
-
-  # dataloaderを準備する
-  # DataLoaderのプロセス数：0はメインプロセスになる
-  n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)      # cpu_count-1 ただし最大で指定された数まで
-  train_dataloader = torch.utils.data.DataLoader(
-      train_dataset_group, batch_size=1, shuffle=True, collate_fn=collate_fn, num_workers=n_workers, persistent_workers=args.persistent_data_loader_workers)
-
-  # 学習ステップ数を計算する
-  if args.max_train_epochs is not None:
-    args.max_train_steps = args.max_train_epochs * len(train_dataloader)
-    print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
-
-  # lr schedulerを用意する
-  lr_scheduler = train_util.get_scheduler_fix(args.lr_scheduler, optimizer, num_warmup_steps=args.lr_warmup_steps,
-                                              num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
-                                              num_cycles=args.lr_scheduler_num_cycles, power=args.lr_scheduler_power)
-
-  # acceleratorがなんかよろしくやってくれるらしい
-  text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
-      text_encoder, optimizer, train_dataloader, lr_scheduler)
-
-  index_no_updates = torch.arange(len(tokenizer)) < token_ids[0]
-  # print(len(index_no_updates), torch.sum(index_no_updates))
-  orig_embeds_params = unwrap_model(text_encoder).get_input_embeddings().weight.data.detach().clone()
-
-  # Freeze all parameters except for the token embeddings in text encoder
-  text_encoder.requires_grad_(True)
-  text_encoder.text_model.encoder.requires_grad_(False)
-  text_encoder.text_model.final_layer_norm.requires_grad_(False)
-  text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
-  # text_encoder.text_model.embeddings.token_embedding.requires_grad_(True)
-
-  unet.requires_grad_(False)
-  unet.to(accelerator.device, dtype=weight_dtype)
-  if args.gradient_checkpointing:                       # according to TI example in Diffusers, train is required
-    unet.train()
-  else:
-    unet.eval()
-
-  if not cache_latents:
-    vae.requires_grad_(False)
-    vae.eval()
-    vae.to(accelerator.device, dtype=weight_dtype)
-
-  # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
-  if args.full_fp16:
-    train_util.patch_accelerator_for_fp16_training(accelerator)
-    text_encoder.to(weight_dtype)
-
-  # resumeする
-  if args.resume is not None:
-    print(f"resume training from state: {args.resume}")
-    accelerator.load_state(args.resume)
-
-  # epoch数を計算する
-  num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
-  num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
-  if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
-    args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
-
-  # 学習する
-  total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
-  print("running training / 学習開始")
-  print(f"  num train images * repeats / 学習画像の数×繰り返し回数: {train_dataset_group.num_train_images}")
-  print(f"  num reg images / 正則化画像の数: {train_dataset_group.num_reg_images}")
-  print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
-  print(f"  num epochs / epoch数: {num_train_epochs}")
-  print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
-  print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
-  print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
-  print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
-
-  progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
-  global_step = 0
-
-  noise_scheduler = DDPMScheduler(beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear",
-                                  num_train_timesteps=1000, clip_sample=False)
-
-  if accelerator.is_main_process:
-    accelerator.init_trackers("textual_inversion")
-
-  for epoch in range(num_train_epochs):
-    print(f"epoch {epoch+1}/{num_train_epochs}")
-    train_dataset_group.set_current_epoch(epoch + 1)
-
-    text_encoder.train()
-
-    loss_total = 0
-    for step, batch in enumerate(train_dataloader):
-      with accelerator.accumulate(text_encoder):
-        with torch.no_grad():
-          if "latents" in batch and batch["latents"] is not None:
-            latents = batch["latents"].to(accelerator.device)
-          else:
-            # latentに変換
-            latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
-          latents = latents * 0.18215
-        b_size = latents.shape[0]
-
-        # Get the text embedding for conditioning
-        input_ids = batch["input_ids"].to(accelerator.device)
-        # weight_dtype) use float instead of fp16/bf16 because text encoder is float
-        encoder_hidden_states = train_util.get_hidden_states(args, input_ids, tokenizer, text_encoder, torch.float)
-
-        # Sample noise that we'll add to the latents
-        noise = torch.randn_like(latents, device=latents.device)
-        if args.noise_offset:
-          # https://www.crosslabs.org//blog/diffusion-with-offset-noise
-          noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
-
-        # Sample a random timestep for each image
-        timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
-        timesteps = timesteps.long()
-
-        # Add noise to the latents according to the noise magnitude at each timestep
-        # (this is the forward diffusion process)
-        noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
-
-        # Predict the noise residual
-        noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
-
-        if args.v_parameterization:
-          # v-parameterization training
-          target = noise_scheduler.get_velocity(latents, noise, timesteps)
+        use_dreambooth_method = args.in_json is None
+        if use_dreambooth_method:
+            print("Use DreamBooth method.")
+            user_config = {
+                "datasets": [
+                    {"subsets": config_util.generate_dreambooth_subsets_config_by_subdirs(args.train_data_dir, args.reg_data_dir)}
+                ]
+            }
         else:
-          target = noise
+            print("Train with captions.")
+            user_config = {
+                "datasets": [
+                    {
+                        "subsets": [
+                            {
+                                "image_dir": args.train_data_dir,
+                                "metadata_file": args.in_json,
+                            }
+                        ]
+                    }
+                ]
+            }
 
-        loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
-        loss = loss.mean([1, 2, 3])
+    blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
+    train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
 
-        loss_weights = batch["loss_weights"]                      # 各sampleごとのweight
-        loss = loss * loss_weights
+    current_epoch = Value('i',0)
+    current_step = Value('i',0)
+    ds_for_collater = train_dataset_group if args.max_data_loader_n_workers == 0 else None
+    collater = train_util.collater_class(current_epoch,current_step, ds_for_collater)
 
-        loss = loss.mean()                # 平均なのでbatch_sizeで割る必要なし
+    # make captions: tokenstring tokenstring1 tokenstring2 ...tokenstringn という文字列に書き換える超乱暴な実装
+    if use_template:
+        print("use template for training captions. is object: {args.use_object_template}")
+        templates = imagenet_templates_small if args.use_object_template else imagenet_style_templates_small
+        replace_to = " ".join(token_strings)
+        captions = []
+        for tmpl in templates:
+            captions.append(tmpl.format(replace_to))
+        train_dataset_group.add_replacement("", captions)
 
-        accelerator.backward(loss)
-        if accelerator.sync_gradients and args.max_grad_norm != 0.0:
-          params_to_clip = text_encoder.get_input_embeddings().parameters()
-          accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+        if args.num_vectors_per_token > 1:
+            prompt_replacement = (args.token_string, replace_to)
+        else:
+            prompt_replacement = None
+    else:
+        if args.num_vectors_per_token > 1:
+            replace_to = " ".join(token_strings)
+            train_dataset_group.add_replacement(args.token_string, replace_to)
+            prompt_replacement = (args.token_string, replace_to)
+        else:
+            prompt_replacement = None
 
-        optimizer.step()
-        lr_scheduler.step()
-        optimizer.zero_grad(set_to_none=True)
+    if args.debug_dataset:
+        train_util.debug_dataset(train_dataset_group, show_input_ids=True)
+        return
+    if len(train_dataset_group) == 0:
+        print("No data found. Please verify arguments / 画像がありません。引数指定を確認してください")
+        return
 
-        # Let's make sure we don't update any embedding weights besides the newly added token
+    if cache_latents:
+        assert (
+            train_dataset_group.is_latent_cacheable()
+        ), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
+
+    # モデルに xformers とか memory efficient attention を組み込む
+    train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
+
+    # 学習を準備する
+    if cache_latents:
+        vae.to(accelerator.device, dtype=weight_dtype)
+        vae.requires_grad_(False)
+        vae.eval()
         with torch.no_grad():
-          unwrap_model(text_encoder).get_input_embeddings().weight[index_no_updates] = orig_embeds_params[index_no_updates]
+            train_dataset_group.cache_latents(vae, args.vae_batch_size)
+        vae.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        gc.collect()
 
-      # Checks if the accelerator has performed an optimization step behind the scenes
-      if accelerator.sync_gradients:
-        progress_bar.update(1)
-        global_step += 1
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
 
-        train_util.sample_images(accelerator, args, None, global_step, accelerator.device,
-                                 vae, tokenizer, text_encoder, unet, prompt_replacement)
+    # 学習に必要なクラスを準備する
+    print("prepare optimizer, data loader etc.")
+    trainable_params = text_encoder.get_input_embeddings().parameters()
+    _, _, optimizer = train_util.get_optimizer(args, trainable_params)
 
-      current_loss = loss.detach().item()
-      if args.logging_dir is not None:
-        logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
-        if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
-          logs["lr/d*lr"] = lr_scheduler.optimizers[0].param_groups[0]['d']*lr_scheduler.optimizers[0].param_groups[0]['lr']
-        accelerator.log(logs, step=global_step)
+    # dataloaderを準備する
+    # DataLoaderのプロセス数：0はメインプロセスになる
+    n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)  # cpu_count-1 ただし最大で指定された数まで
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset_group,
+        batch_size=1,
+        shuffle=True,
+        collate_fn=collater,
+        num_workers=n_workers,
+        persistent_workers=args.persistent_data_loader_workers,
+    )
 
-      loss_total += current_loss
-      avr_loss = loss_total / (step+1)
-      logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
-      progress_bar.set_postfix(**logs)
+    # 学習ステップ数を計算する
+    if args.max_train_epochs is not None:
+        args.max_train_steps = args.max_train_epochs * math.ceil(len(train_dataloader) / accelerator.num_processes / args.gradient_accumulation_steps)
+        print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
 
-      if global_step >= args.max_train_steps:
-        break
+    # データセット側にも学習ステップを送信
+    train_dataset_group.set_max_train_steps(args.max_train_steps)
 
-    if args.logging_dir is not None:
-      logs = {"loss/epoch": loss_total / len(train_dataloader)}
-      accelerator.log(logs, step=epoch+1)
+    # lr schedulerを用意する
+    lr_scheduler = train_util.get_scheduler_fix(args, optimizer, accelerator.num_processes)
 
-    accelerator.wait_for_everyone()
+    # acceleratorがなんかよろしくやってくれるらしい
+    text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder, optimizer, train_dataloader, lr_scheduler
+    )
 
-    updated_embs = unwrap_model(text_encoder).get_input_embeddings().weight[token_ids].data.detach().clone()
+    index_no_updates = torch.arange(len(tokenizer)) < token_ids[0]
+    # print(len(index_no_updates), torch.sum(index_no_updates))
+    orig_embeds_params = unwrap_model(text_encoder).get_input_embeddings().weight.data.detach().clone()
 
-    if args.save_every_n_epochs is not None:
-      model_name = train_util.DEFAULT_EPOCH_NAME if args.output_name is None else args.output_name
+    # Freeze all parameters except for the token embeddings in text encoder
+    text_encoder.requires_grad_(True)
+    text_encoder.text_model.encoder.requires_grad_(False)
+    text_encoder.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
+    # text_encoder.text_model.embeddings.token_embedding.requires_grad_(True)
 
-      def save_func():
-        ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, epoch + 1) + '.' + args.save_model_as
+    unet.requires_grad_(False)
+    unet.to(accelerator.device, dtype=weight_dtype)
+    if args.gradient_checkpointing:  # according to TI example in Diffusers, train is required
+        unet.train()
+    else:
+        unet.eval()
+
+    if not cache_latents:
+        vae.requires_grad_(False)
+        vae.eval()
+        vae.to(accelerator.device, dtype=weight_dtype)
+
+    # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
+    if args.full_fp16:
+        train_util.patch_accelerator_for_fp16_training(accelerator)
+        text_encoder.to(weight_dtype)
+
+    # resumeする
+    if args.resume is not None:
+        print(f"resume training from state: {args.resume}")
+        accelerator.load_state(args.resume)
+
+    # epoch数を計算する
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+    if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
+        args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
+
+    # 学習する
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+    print("running training / 学習開始")
+    print(f"  num train images * repeats / 学習画像の数×繰り返し回数: {train_dataset_group.num_train_images}")
+    print(f"  num reg images / 正則化画像の数: {train_dataset_group.num_reg_images}")
+    print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
+    print(f"  num epochs / epoch数: {num_train_epochs}")
+    print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
+    print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
+    print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
+    print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
+
+    progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
+    global_step = 0
+
+    noise_scheduler = DDPMScheduler(
+        beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000, clip_sample=False
+    )
+
+    if accelerator.is_main_process:
+        accelerator.init_trackers("textual_inversion")
+
+    for epoch in range(num_train_epochs):
+        print(f"epoch {epoch+1}/{num_train_epochs}")
+        current_epoch.value = epoch+1
+
+        text_encoder.train()
+
+        loss_total = 0
+
+        for step, batch in enumerate(train_dataloader):
+            current_step.value = global_step
+            with accelerator.accumulate(text_encoder):
+                with torch.no_grad():
+                    if "latents" in batch and batch["latents"] is not None:
+                        latents = batch["latents"].to(accelerator.device)
+                    else:
+                        # latentに変換
+                        latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
+                    latents = latents * 0.18215
+                b_size = latents.shape[0]
+
+                # Get the text embedding for conditioning
+                input_ids = batch["input_ids"].to(accelerator.device)
+                # weight_dtype) use float instead of fp16/bf16 because text encoder is float
+                encoder_hidden_states = train_util.get_hidden_states(args, input_ids, tokenizer, text_encoder, torch.float)
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents, device=latents.device)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
+
+                # Sample a random timestep for each image
+                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
+                timesteps = timesteps.long()
+
+                # Add noise to the latents according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                # Predict the noise residual
+                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                if args.v_parameterization:
+                    # v-parameterization training
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    target = noise
+
+                loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
+                loss = loss.mean([1, 2, 3])
+                
+                if args.min_snr_gamma:
+                  loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma)
+
+                loss_weights = batch["loss_weights"]  # 各sampleごとのweight
+                loss = loss * loss_weights
+
+                loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients and args.max_grad_norm != 0.0:
+                    params_to_clip = text_encoder.get_input_embeddings().parameters()
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=True)
+
+                # Let's make sure we don't update any embedding weights besides the newly added token
+                with torch.no_grad():
+                    unwrap_model(text_encoder).get_input_embeddings().weight[index_no_updates] = orig_embeds_params[
+                        index_no_updates
+                    ]
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+
+                train_util.sample_images(
+                    accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet, prompt_replacement
+                )
+
+            current_loss = loss.detach().item()
+            if args.logging_dir is not None:
+                logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
+                if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
+                    logs["lr/d*lr"] = (
+                        lr_scheduler.optimizers[0].param_groups[0]["d"] * lr_scheduler.optimizers[0].param_groups[0]["lr"]
+                    )
+                accelerator.log(logs, step=global_step)
+
+            loss_total += current_loss
+            avr_loss = loss_total / (step + 1)
+            logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if args.logging_dir is not None:
+            logs = {"loss/epoch": loss_total / len(train_dataloader)}
+            accelerator.log(logs, step=epoch + 1)
+
+        accelerator.wait_for_everyone()
+
+        updated_embs = unwrap_model(text_encoder).get_input_embeddings().weight[token_ids].data.detach().clone()
+
+        if args.save_every_n_epochs is not None:
+            model_name = train_util.DEFAULT_EPOCH_NAME if args.output_name is None else args.output_name
+
+            def save_func():
+                ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, epoch + 1) + "." + args.save_model_as
+                ckpt_file = os.path.join(args.output_dir, ckpt_name)
+                print(f"saving checkpoint: {ckpt_file}")
+                save_weights(ckpt_file, updated_embs, save_dtype)
+
+            def remove_old_func(old_epoch_no):
+                old_ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, old_epoch_no) + "." + args.save_model_as
+                old_ckpt_file = os.path.join(args.output_dir, old_ckpt_name)
+                if os.path.exists(old_ckpt_file):
+                    print(f"removing old checkpoint: {old_ckpt_file}")
+                    os.remove(old_ckpt_file)
+
+            saving = train_util.save_on_epoch_end(args, save_func, remove_old_func, epoch + 1, num_train_epochs)
+            if saving and args.save_state:
+                train_util.save_state_on_epoch_end(args, accelerator, model_name, epoch + 1)
+
+        train_util.sample_images(
+            accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet, prompt_replacement
+        )
+
+        # end of epoch
+
+    is_main_process = accelerator.is_main_process
+    if is_main_process:
+        text_encoder = unwrap_model(text_encoder)
+
+    accelerator.end_training()
+
+    if args.save_state:
+        train_util.save_state_on_train_end(args, accelerator)
+
+    updated_embs = text_encoder.get_input_embeddings().weight[token_ids].data.detach().clone()
+
+    del accelerator  # この後メモリを使うのでこれは消す
+
+    if is_main_process:
+        os.makedirs(args.output_dir, exist_ok=True)
+
+        model_name = train_util.DEFAULT_LAST_OUTPUT_NAME if args.output_name is None else args.output_name
+        ckpt_name = model_name + "." + args.save_model_as
         ckpt_file = os.path.join(args.output_dir, ckpt_name)
-        print(f"saving checkpoint: {ckpt_file}")
+
+        print(f"save trained model to {ckpt_file}")
         save_weights(ckpt_file, updated_embs, save_dtype)
-
-      def remove_old_func(old_epoch_no):
-        old_ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, old_epoch_no) + '.' + args.save_model_as
-        old_ckpt_file = os.path.join(args.output_dir, old_ckpt_name)
-        if os.path.exists(old_ckpt_file):
-          print(f"removing old checkpoint: {old_ckpt_file}")
-          os.remove(old_ckpt_file)
-
-      saving = train_util.save_on_epoch_end(args, save_func, remove_old_func, epoch + 1, num_train_epochs)
-      if saving and args.save_state:
-        train_util.save_state_on_epoch_end(args, accelerator, model_name, epoch + 1)
-
-    train_util.sample_images(accelerator, args, epoch + 1, global_step, accelerator.device,
-                             vae, tokenizer, text_encoder, unet, prompt_replacement)
-
-    # end of epoch
-
-  is_main_process = accelerator.is_main_process
-  if is_main_process:
-    text_encoder = unwrap_model(text_encoder)
-
-  accelerator.end_training()
-
-  if args.save_state:
-    train_util.save_state_on_train_end(args, accelerator)
-
-  updated_embs = text_encoder.get_input_embeddings().weight[token_ids].data.detach().clone()
-
-  del accelerator                         # この後メモリを使うのでこれは消す
-
-  if is_main_process:
-    os.makedirs(args.output_dir, exist_ok=True)
-
-    model_name = train_util.DEFAULT_LAST_OUTPUT_NAME if args.output_name is None else args.output_name
-    ckpt_name = model_name + '.' + args.save_model_as
-    ckpt_file = os.path.join(args.output_dir, ckpt_name)
-
-    print(f"save trained model to {ckpt_file}")
-    save_weights(ckpt_file, updated_embs, save_dtype)
-    print("model saved.")
+        print("model saved.")
 
 
 def save_weights(file, updated_embs, save_dtype):
-  state_dict = {"emb_params": updated_embs}
+    state_dict = {"emb_params": updated_embs}
 
-  if save_dtype is not None:
-    for key in list(state_dict.keys()):
-      v = state_dict[key]
-      v = v.detach().clone().to("cpu").to(save_dtype)
-      state_dict[key] = v
+    if save_dtype is not None:
+        for key in list(state_dict.keys()):
+            v = state_dict[key]
+            v = v.detach().clone().to("cpu").to(save_dtype)
+            state_dict[key] = v
 
-  if os.path.splitext(file)[1] == '.safetensors':
-    from safetensors.torch import save_file
-    save_file(state_dict, file)
-  else:
-    torch.save(state_dict, file)                    # can be loaded in Web UI
+    if os.path.splitext(file)[1] == ".safetensors":
+        from safetensors.torch import save_file
+
+        save_file(state_dict, file)
+    else:
+        torch.save(state_dict, file)  # can be loaded in Web UI
 
 
 def load_weights(file):
-  if os.path.splitext(file)[1] == '.safetensors':
-    from safetensors.torch import load_file
-    data = load_file(file)
-  else:
-    # compatible to Web UI's file format
-    data = torch.load(file, map_location='cpu')
-    if type(data) != dict:
-      raise ValueError(f"weight file is not dict / 重みファイルがdict形式ではありません: {file}")
+    if os.path.splitext(file)[1] == ".safetensors":
+        from safetensors.torch import load_file
 
-    if 'string_to_param' in data:                           # textual inversion embeddings
-      data = data['string_to_param']
-      if hasattr(data, '_parameters'):                      # support old PyTorch?
-        data = getattr(data, '_parameters')
+        data = load_file(file)
+    else:
+        # compatible to Web UI's file format
+        data = torch.load(file, map_location="cpu")
+        if type(data) != dict:
+            raise ValueError(f"weight file is not dict / 重みファイルがdict形式ではありません: {file}")
 
-  emb = next(iter(data.values()))
-  if type(emb) != torch.Tensor:
-    raise ValueError(f"weight file does not contains Tensor / 重みファイルのデータがTensorではありません: {file}")
+        if "string_to_param" in data:  # textual inversion embeddings
+            data = data["string_to_param"]
+            if hasattr(data, "_parameters"):  # support old PyTorch?
+                data = getattr(data, "_parameters")
 
-  if len(emb.size()) == 1:
-    emb = emb.unsqueeze(0)
+    emb = next(iter(data.values()))
+    if type(emb) != torch.Tensor:
+        raise ValueError(f"weight file does not contains Tensor / 重みファイルのデータがTensorではありません: {file}")
 
-  return emb
+    if len(emb.size()) == 1:
+        emb = emb.unsqueeze(0)
+
+    return emb
 
 
-if __name__ == '__main__':
-  parser = argparse.ArgumentParser()
+def setup_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser()
 
-  train_util.add_sd_models_arguments(parser)
-  train_util.add_dataset_arguments(parser, True, True, False)
-  train_util.add_training_arguments(parser, True)
-  train_util.add_optimizer_arguments(parser)
-  config_util.add_config_arguments(parser)
+    train_util.add_sd_models_arguments(parser)
+    train_util.add_dataset_arguments(parser, True, True, False)
+    train_util.add_training_arguments(parser, True)
+    train_util.add_optimizer_arguments(parser)
+    config_util.add_config_arguments(parser)
+    custom_train_functions.add_custom_train_arguments(parser)
 
-  parser.add_argument("--save_model_as", type=str, default="pt", choices=[None, "ckpt", "pt", "safetensors"],
-                      help="format to save the model (default is .pt) / モデル保存時の形式（デフォルトはpt）")
+    parser.add_argument(
+        "--save_model_as",
+        type=str,
+        default="pt",
+        choices=[None, "ckpt", "pt", "safetensors"],
+        help="format to save the model (default is .pt) / モデル保存時の形式（デフォルトはpt）",
+    )
 
-  parser.add_argument("--weights", type=str, default=None,
-                      help="embedding weights to initialize / 学習するネットワークの初期重み")
-  parser.add_argument("--num_vectors_per_token", type=int, default=1,
-                      help='number of vectors per token / トークンに割り当てるembeddingsの要素数')
-  parser.add_argument("--token_string", type=str, default=None,
-                      help="token string used in training, must not exist in tokenizer / 学習時に使用されるトークン文字列、tokenizerに存在しない文字であること")
-  parser.add_argument("--init_word", type=str, default=None,
-                      help="words to initialize vector / ベクトルを初期化に使用する単語、複数可")
-  parser.add_argument("--use_object_template", action='store_true',
-                      help="ignore caption and use default templates for object / キャプションは使わずデフォルトの物体用テンプレートで学習する")
-  parser.add_argument("--use_style_template", action='store_true',
-                      help="ignore caption and use default templates for stype / キャプションは使わずデフォルトのスタイル用テンプレートで学習する")
+    parser.add_argument("--weights", type=str, default=None, help="embedding weights to initialize / 学習するネットワークの初期重み")
+    parser.add_argument(
+        "--num_vectors_per_token", type=int, default=1, help="number of vectors per token / トークンに割り当てるembeddingsの要素数"
+    )
+    parser.add_argument(
+        "--token_string",
+        type=str,
+        default=None,
+        help="token string used in training, must not exist in tokenizer / 学習時に使用されるトークン文字列、tokenizerに存在しない文字であること",
+    )
+    parser.add_argument("--init_word", type=str, default=None, help="words to initialize vector / ベクトルを初期化に使用する単語、複数可")
+    parser.add_argument(
+        "--use_object_template",
+        action="store_true",
+        help="ignore caption and use default templates for object / キャプションは使わずデフォルトの物体用テンプレートで学習する",
+    )
+    parser.add_argument(
+        "--use_style_template",
+        action="store_true",
+        help="ignore caption and use default templates for stype / キャプションは使わずデフォルトのスタイル用テンプレートで学習する",
+    )
 
-  args = parser.parse_args()
-  train(args)
+    return parser
+
+
+if __name__ == "__main__":
+    parser = setup_parser()
+
+    args = parser.parse_args()
+    args = train_util.read_config_from_file(args, parser)
+
+    train(args)

train_textual_inversion_XTI.py

@@ -0,0 +1,644 @@
+import importlib
+import argparse
+import gc
+import math
+import os
+import toml
+from multiprocessing import Value
+
+from tqdm import tqdm
+import torch
+from accelerate.utils import set_seed
+import diffusers
+from diffusers import DDPMScheduler
+
+import library.train_util as train_util
+import library.config_util as config_util
+from library.config_util import (
+    ConfigSanitizer,
+    BlueprintGenerator,
+)
+import library.custom_train_functions as custom_train_functions
+from library.custom_train_functions import apply_snr_weight
+from XTI_hijack import unet_forward_XTI, downblock_forward_XTI, upblock_forward_XTI
+
+imagenet_templates_small = [
+    "a photo of a {}",
+    "a rendering of a {}",
+    "a cropped photo of the {}",
+    "the photo of a {}",
+    "a photo of a clean {}",
+    "a photo of a dirty {}",
+    "a dark photo of the {}",
+    "a photo of my {}",
+    "a photo of the cool {}",
+    "a close-up photo of a {}",
+    "a bright photo of the {}",
+    "a cropped photo of a {}",
+    "a photo of the {}",
+    "a good photo of the {}",
+    "a photo of one {}",
+    "a close-up photo of the {}",
+    "a rendition of the {}",
+    "a photo of the clean {}",
+    "a rendition of a {}",
+    "a photo of a nice {}",
+    "a good photo of a {}",
+    "a photo of the nice {}",
+    "a photo of the small {}",
+    "a photo of the weird {}",
+    "a photo of the large {}",
+    "a photo of a cool {}",
+    "a photo of a small {}",
+]
+
+imagenet_style_templates_small = [
+    "a painting in the style of {}",
+    "a rendering in the style of {}",
+    "a cropped painting in the style of {}",
+    "the painting in the style of {}",
+    "a clean painting in the style of {}",
+    "a dirty painting in the style of {}",
+    "a dark painting in the style of {}",
+    "a picture in the style of {}",
+    "a cool painting in the style of {}",
+    "a close-up painting in the style of {}",
+    "a bright painting in the style of {}",
+    "a cropped painting in the style of {}",
+    "a good painting in the style of {}",
+    "a close-up painting in the style of {}",
+    "a rendition in the style of {}",
+    "a nice painting in the style of {}",
+    "a small painting in the style of {}",
+    "a weird painting in the style of {}",
+    "a large painting in the style of {}",
+]
+
+
+def train(args):
+    if args.output_name is None:
+        args.output_name = args.token_string
+    use_template = args.use_object_template or args.use_style_template
+
+    train_util.verify_training_args(args)
+    train_util.prepare_dataset_args(args, True)
+
+    if args.sample_every_n_steps is not None or args.sample_every_n_epochs is not None:
+        print(
+            "sample_every_n_steps and sample_every_n_epochs are not supported in this script currently / sample_every_n_stepsとsample_every_n_epochsは現在このスクリプトではサポートされていません"
+        )
+
+    cache_latents = args.cache_latents
+
+    if args.seed is not None:
+        set_seed(args.seed)
+
+    tokenizer = train_util.load_tokenizer(args)
+
+    # acceleratorを準備する
+    print("prepare accelerator")
+    accelerator, unwrap_model = train_util.prepare_accelerator(args)
+
+    # mixed precisionに対応した型を用意しておき適宜castする
+    weight_dtype, save_dtype = train_util.prepare_dtype(args)
+
+    # モデルを読み込む
+    text_encoder, vae, unet, _ = train_util.load_target_model(args, weight_dtype)
+
+    # Convert the init_word to token_id
+    if args.init_word is not None:
+        init_token_ids = tokenizer.encode(args.init_word, add_special_tokens=False)
+        if len(init_token_ids) > 1 and len(init_token_ids) != args.num_vectors_per_token:
+            print(
+                f"token length for init words is not same to num_vectors_per_token, init words is repeated or truncated / 初期化単語のトークン長がnum_vectors_per_tokenと合わないため、繰り返しまたは切り捨てが発生します: length {len(init_token_ids)}"
+            )
+    else:
+        init_token_ids = None
+
+    # add new word to tokenizer, count is num_vectors_per_token
+    token_strings = [args.token_string] + [f"{args.token_string}{i+1}" for i in range(args.num_vectors_per_token - 1)]
+    num_added_tokens = tokenizer.add_tokens(token_strings)
+    assert (
+        num_added_tokens == args.num_vectors_per_token
+    ), f"tokenizer has same word to token string. please use another one / 指定したargs.token_stringは既に存在します。別の単語を使ってください: {args.token_string}"
+
+    token_ids = tokenizer.convert_tokens_to_ids(token_strings)
+    print(f"tokens are added: {token_ids}")
+    assert min(token_ids) == token_ids[0] and token_ids[-1] == token_ids[0] + len(token_ids) - 1, f"token ids is not ordered"
+    assert len(tokenizer) - 1 == token_ids[-1], f"token ids is not end of tokenize: {len(tokenizer)}"
+
+    token_strings_XTI = []
+    XTI_layers = [
+        "IN01",
+        "IN02",
+        "IN04",
+        "IN05",
+        "IN07",
+        "IN08",
+        "MID",
+        "OUT03",
+        "OUT04",
+        "OUT05",
+        "OUT06",
+        "OUT07",
+        "OUT08",
+        "OUT09",
+        "OUT10",
+        "OUT11",
+    ]
+    for layer_name in XTI_layers:
+        token_strings_XTI += [f"{t}_{layer_name}" for t in token_strings]
+
+    tokenizer.add_tokens(token_strings_XTI)
+    token_ids_XTI = tokenizer.convert_tokens_to_ids(token_strings_XTI)
+    print(f"tokens are added (XTI): {token_ids_XTI}")
+    # Resize the token embeddings as we are adding new special tokens to the tokenizer
+    text_encoder.resize_token_embeddings(len(tokenizer))
+
+    # Initialise the newly added placeholder token with the embeddings of the initializer token
+    token_embeds = text_encoder.get_input_embeddings().weight.data
+    if init_token_ids is not None:
+        for i, token_id in enumerate(token_ids_XTI):
+            token_embeds[token_id] = token_embeds[init_token_ids[(i // 16) % len(init_token_ids)]]
+            # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
+
+    # load weights
+    if args.weights is not None:
+        embeddings = load_weights(args.weights)
+        assert len(token_ids) == len(
+            embeddings
+        ), f"num_vectors_per_token is mismatch for weights / 指定した重みとnum_vectors_per_tokenの値が異なります: {len(embeddings)}"
+        # print(token_ids, embeddings.size())
+        for token_id, embedding in zip(token_ids_XTI, embeddings):
+            token_embeds[token_id] = embedding
+            # print(token_id, token_embeds[token_id].mean(), token_embeds[token_id].min())
+        print(f"weighs loaded")
+
+    print(f"create embeddings for {args.num_vectors_per_token} tokens, for {args.token_string}")
+
+    # データセットを準備する
+    blueprint_generator = BlueprintGenerator(ConfigSanitizer(True, True, False))
+    if args.dataset_config is not None:
+        print(f"Load dataset config from {args.dataset_config}")
+        user_config = config_util.load_user_config(args.dataset_config)
+        ignored = ["train_data_dir", "reg_data_dir", "in_json"]
+        if any(getattr(args, attr) is not None for attr in ignored):
+            print(
+                "ignore following options because config file is found: {0} / 設定ファイルが利用されるため以下のオプションは無視されます: {0}".format(
+                    ", ".join(ignored)
+                )
+            )
+    else:
+        use_dreambooth_method = args.in_json is None
+        if use_dreambooth_method:
+            print("Use DreamBooth method.")
+            user_config = {
+                "datasets": [
+                    {"subsets": config_util.generate_dreambooth_subsets_config_by_subdirs(args.train_data_dir, args.reg_data_dir)}
+                ]
+            }
+        else:
+            print("Train with captions.")
+            user_config = {
+                "datasets": [
+                    {
+                        "subsets": [
+                            {
+                                "image_dir": args.train_data_dir,
+                                "metadata_file": args.in_json,
+                            }
+                        ]
+                    }
+                ]
+            }
+
+    blueprint = blueprint_generator.generate(user_config, args, tokenizer=tokenizer)
+    train_dataset_group = config_util.generate_dataset_group_by_blueprint(blueprint.dataset_group)
+    train_dataset_group.enable_XTI(XTI_layers, token_strings=token_strings)
+    current_epoch = Value("i", 0)
+    current_step = Value("i", 0)
+    ds_for_collater = train_dataset_group if args.max_data_loader_n_workers == 0 else None
+    collater = train_util.collater_class(current_epoch, current_step, ds_for_collater)
+
+    # make captions: tokenstring tokenstring1 tokenstring2 ...tokenstringn という文字列に書き換える超乱暴な実装
+    if use_template:
+        print("use template for training captions. is object: {args.use_object_template}")
+        templates = imagenet_templates_small if args.use_object_template else imagenet_style_templates_small
+        replace_to = " ".join(token_strings)
+        captions = []
+        for tmpl in templates:
+            captions.append(tmpl.format(replace_to))
+        train_dataset_group.add_replacement("", captions)
+
+        if args.num_vectors_per_token > 1:
+            prompt_replacement = (args.token_string, replace_to)
+        else:
+            prompt_replacement = None
+    else:
+        if args.num_vectors_per_token > 1:
+            replace_to = " ".join(token_strings)
+            train_dataset_group.add_replacement(args.token_string, replace_to)
+            prompt_replacement = (args.token_string, replace_to)
+        else:
+            prompt_replacement = None
+
+    if args.debug_dataset:
+        train_util.debug_dataset(train_dataset_group, show_input_ids=True)
+        return
+    if len(train_dataset_group) == 0:
+        print("No data found. Please verify arguments / 画像がありません。引数指定を確認してください")
+        return
+
+    if cache_latents:
+        assert (
+            train_dataset_group.is_latent_cacheable()
+        ), "when caching latents, either color_aug or random_crop cannot be used / latentをキャッシュするときはcolor_augとrandom_cropは使えません"
+
+    # モデルに xformers とか memory efficient attention を組み込む
+    train_util.replace_unet_modules(unet, args.mem_eff_attn, args.xformers)
+    diffusers.models.UNet2DConditionModel.forward = unet_forward_XTI
+    diffusers.models.unet_2d_blocks.CrossAttnDownBlock2D.forward = downblock_forward_XTI
+    diffusers.models.unet_2d_blocks.CrossAttnUpBlock2D.forward = upblock_forward_XTI
+
+    # 学習を準備する
+    if cache_latents:
+        vae.to(accelerator.device, dtype=weight_dtype)
+        vae.requires_grad_(False)
+        vae.eval()
+        with torch.no_grad():
+            train_dataset_group.cache_latents(vae, args.vae_batch_size)
+        vae.to("cpu")
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+        gc.collect()
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
+
+    # 学習に必要なクラスを準備する
+    print("prepare optimizer, data loader etc.")
+    trainable_params = text_encoder.get_input_embeddings().parameters()
+    _, _, optimizer = train_util.get_optimizer(args, trainable_params)
+
+    # dataloaderを準備する
+    # DataLoaderのプロセス数：0はメインプロセスになる
+    n_workers = min(args.max_data_loader_n_workers, os.cpu_count() - 1)  # cpu_count-1 ただし最大で指定された数まで
+    train_dataloader = torch.utils.data.DataLoader(
+        train_dataset_group,
+        batch_size=1,
+        shuffle=True,
+        collate_fn=collater,
+        num_workers=n_workers,
+        persistent_workers=args.persistent_data_loader_workers,
+    )
+
+    # 学習ステップ数を計算する
+    if args.max_train_epochs is not None:
+        args.max_train_steps = args.max_train_epochs * math.ceil(
+            len(train_dataloader) / accelerator.num_processes / args.gradient_accumulation_steps
+        )
+        print(f"override steps. steps for {args.max_train_epochs} epochs is / 指定エポックまでのステップ数: {args.max_train_steps}")
+
+    # データセット側にも学習ステップを送信
+    train_dataset_group.set_max_train_steps(args.max_train_steps)
+
+    # lr schedulerを用意する
+    lr_scheduler = train_util.get_scheduler_fix(args, optimizer, accelerator.num_processes)
+
+    # acceleratorがなんかよろしくやってくれるらしい
+    text_encoder, optimizer, train_dataloader, lr_scheduler = accelerator.prepare(
+        text_encoder, optimizer, train_dataloader, lr_scheduler
+    )
+
+    index_no_updates = torch.arange(len(tokenizer)) < token_ids_XTI[0]
+    # print(len(index_no_updates), torch.sum(index_no_updates))
+    orig_embeds_params = unwrap_model(text_encoder).get_input_embeddings().weight.data.detach().clone()
+
+    # Freeze all parameters except for the token embeddings in text encoder
+    text_encoder.requires_grad_(True)
+    text_encoder.text_model.encoder.requires_grad_(False)
+    text_encoder.text_model.final_layer_norm.requires_grad_(False)
+    text_encoder.text_model.embeddings.position_embedding.requires_grad_(False)
+    # text_encoder.text_model.embeddings.token_embedding.requires_grad_(True)
+
+    unet.requires_grad_(False)
+    unet.to(accelerator.device, dtype=weight_dtype)
+    if args.gradient_checkpointing:  # according to TI example in Diffusers, train is required
+        unet.train()
+    else:
+        unet.eval()
+
+    if not cache_latents:
+        vae.requires_grad_(False)
+        vae.eval()
+        vae.to(accelerator.device, dtype=weight_dtype)
+
+    # 実験的機能：勾配も含めたfp16学習を行う　PyTorchにパッチを当ててfp16でのgrad scaleを有効にする
+    if args.full_fp16:
+        train_util.patch_accelerator_for_fp16_training(accelerator)
+        text_encoder.to(weight_dtype)
+
+    # resumeする
+    if args.resume is not None:
+        print(f"resume training from state: {args.resume}")
+        accelerator.load_state(args.resume)
+
+    # epoch数を計算する
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    num_train_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+    if (args.save_n_epoch_ratio is not None) and (args.save_n_epoch_ratio > 0):
+        args.save_every_n_epochs = math.floor(num_train_epochs / args.save_n_epoch_ratio) or 1
+
+    # 学習する
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+    print("running training / 学習開始")
+    print(f"  num train images * repeats / 学習画像の数×繰り返し回数: {train_dataset_group.num_train_images}")
+    print(f"  num reg images / 正則化画像の数: {train_dataset_group.num_reg_images}")
+    print(f"  num batches per epoch / 1epochのバッチ数: {len(train_dataloader)}")
+    print(f"  num epochs / epoch数: {num_train_epochs}")
+    print(f"  batch size per device / バッチサイズ: {args.train_batch_size}")
+    print(f"  total train batch size (with parallel & distributed & accumulation) / 総バッチサイズ（並列学習、勾配合計含む）: {total_batch_size}")
+    print(f"  gradient ccumulation steps / 勾配を合計するステップ数 = {args.gradient_accumulation_steps}")
+    print(f"  total optimization steps / 学習ステップ数: {args.max_train_steps}")
+
+    progress_bar = tqdm(range(args.max_train_steps), smoothing=0, disable=not accelerator.is_local_main_process, desc="steps")
+    global_step = 0
+
+    noise_scheduler = DDPMScheduler(
+        beta_start=0.00085, beta_end=0.012, beta_schedule="scaled_linear", num_train_timesteps=1000, clip_sample=False
+    )
+
+    if accelerator.is_main_process:
+        accelerator.init_trackers("textual_inversion")
+
+    for epoch in range(num_train_epochs):
+        print(f"epoch {epoch+1}/{num_train_epochs}")
+        current_epoch.value = epoch + 1
+
+        text_encoder.train()
+
+        loss_total = 0
+
+        for step, batch in enumerate(train_dataloader):
+            current_step.value = global_step
+            with accelerator.accumulate(text_encoder):
+                with torch.no_grad():
+                    if "latents" in batch and batch["latents"] is not None:
+                        latents = batch["latents"].to(accelerator.device)
+                    else:
+                        # latentに変換
+                        latents = vae.encode(batch["images"].to(dtype=weight_dtype)).latent_dist.sample()
+                    latents = latents * 0.18215
+                b_size = latents.shape[0]
+
+                # Get the text embedding for conditioning
+                input_ids = batch["input_ids"].to(accelerator.device)
+                # weight_dtype) use float instead of fp16/bf16 because text encoder is float
+                encoder_hidden_states = torch.stack(
+                    [
+                        train_util.get_hidden_states(args, s, tokenizer, text_encoder, weight_dtype)
+                        for s in torch.split(input_ids, 1, dim=1)
+                    ]
+                )
+
+                # Sample noise that we'll add to the latents
+                noise = torch.randn_like(latents, device=latents.device)
+                if args.noise_offset:
+                    # https://www.crosslabs.org//blog/diffusion-with-offset-noise
+                    noise += args.noise_offset * torch.randn((latents.shape[0], latents.shape[1], 1, 1), device=latents.device)
+
+                # Sample a random timestep for each image
+                timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps, (b_size,), device=latents.device)
+                timesteps = timesteps.long()
+
+                # Add noise to the latents according to the noise magnitude at each timestep
+                # (this is the forward diffusion process)
+                noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                # Predict the noise residual
+                noise_pred = unet(noisy_latents, timesteps, encoder_hidden_states=encoder_hidden_states).sample
+
+                if args.v_parameterization:
+                    # v-parameterization training
+                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                else:
+                    target = noise
+
+                loss = torch.nn.functional.mse_loss(noise_pred.float(), target.float(), reduction="none")
+                loss = loss.mean([1, 2, 3])
+
+                if args.min_snr_gamma:
+                    loss = apply_snr_weight(loss, timesteps, noise_scheduler, args.min_snr_gamma)
+
+                loss_weights = batch["loss_weights"]  # 各sampleごとのweight
+                loss = loss * loss_weights
+
+                loss = loss.mean()  # 平均なのでbatch_sizeで割る必要なし
+
+                accelerator.backward(loss)
+                if accelerator.sync_gradients and args.max_grad_norm != 0.0:
+                    params_to_clip = text_encoder.get_input_embeddings().parameters()
+                    accelerator.clip_grad_norm_(params_to_clip, args.max_grad_norm)
+
+                optimizer.step()
+                lr_scheduler.step()
+                optimizer.zero_grad(set_to_none=True)
+
+                # Let's make sure we don't update any embedding weights besides the newly added token
+                with torch.no_grad():
+                    unwrap_model(text_encoder).get_input_embeddings().weight[index_no_updates] = orig_embeds_params[
+                        index_no_updates
+                    ]
+
+            # Checks if the accelerator has performed an optimization step behind the scenes
+            if accelerator.sync_gradients:
+                progress_bar.update(1)
+                global_step += 1
+                # TODO: fix sample_images
+                # train_util.sample_images(
+                #     accelerator, args, None, global_step, accelerator.device, vae, tokenizer, text_encoder, unet, prompt_replacement
+                # )
+
+            current_loss = loss.detach().item()
+            if args.logging_dir is not None:
+                logs = {"loss": current_loss, "lr": float(lr_scheduler.get_last_lr()[0])}
+                if args.optimizer_type.lower() == "DAdaptation".lower():  # tracking d*lr value
+                    logs["lr/d*lr"] = (
+                        lr_scheduler.optimizers[0].param_groups[0]["d"] * lr_scheduler.optimizers[0].param_groups[0]["lr"]
+                    )
+                accelerator.log(logs, step=global_step)
+
+            loss_total += current_loss
+            avr_loss = loss_total / (step + 1)
+            logs = {"loss": avr_loss}  # , "lr": lr_scheduler.get_last_lr()[0]}
+            progress_bar.set_postfix(**logs)
+
+            if global_step >= args.max_train_steps:
+                break
+
+        if args.logging_dir is not None:
+            logs = {"loss/epoch": loss_total / len(train_dataloader)}
+            accelerator.log(logs, step=epoch + 1)
+
+        accelerator.wait_for_everyone()
+
+        updated_embs = unwrap_model(text_encoder).get_input_embeddings().weight[token_ids_XTI].data.detach().clone()
+
+        if args.save_every_n_epochs is not None:
+            model_name = train_util.DEFAULT_EPOCH_NAME if args.output_name is None else args.output_name
+
+            def save_func():
+                ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, epoch + 1) + "." + args.save_model_as
+                ckpt_file = os.path.join(args.output_dir, ckpt_name)
+                print(f"saving checkpoint: {ckpt_file}")
+                save_weights(ckpt_file, updated_embs, save_dtype)
+
+            def remove_old_func(old_epoch_no):
+                old_ckpt_name = train_util.EPOCH_FILE_NAME.format(model_name, old_epoch_no) + "." + args.save_model_as
+                old_ckpt_file = os.path.join(args.output_dir, old_ckpt_name)
+                if os.path.exists(old_ckpt_file):
+                    print(f"removing old checkpoint: {old_ckpt_file}")
+                    os.remove(old_ckpt_file)
+
+            saving = train_util.save_on_epoch_end(args, save_func, remove_old_func, epoch + 1, num_train_epochs)
+            if saving and args.save_state:
+                train_util.save_state_on_epoch_end(args, accelerator, model_name, epoch + 1)
+
+        # TODO: fix sample_images
+        # train_util.sample_images(
+        #     accelerator, args, epoch + 1, global_step, accelerator.device, vae, tokenizer, text_encoder, unet, prompt_replacement
+        # )
+
+        # end of epoch
+
+    is_main_process = accelerator.is_main_process
+    if is_main_process:
+        text_encoder = unwrap_model(text_encoder)
+
+    accelerator.end_training()
+
+    if args.save_state:
+        train_util.save_state_on_train_end(args, accelerator)
+
+    updated_embs = text_encoder.get_input_embeddings().weight[token_ids_XTI].data.detach().clone()
+
+    del accelerator  # この後メモリを使うのでこれは消す
+
+    if is_main_process:
+        os.makedirs(args.output_dir, exist_ok=True)
+
+        model_name = train_util.DEFAULT_LAST_OUTPUT_NAME if args.output_name is None else args.output_name
+        ckpt_name = model_name + "." + args.save_model_as
+        ckpt_file = os.path.join(args.output_dir, ckpt_name)
+
+        print(f"save trained model to {ckpt_file}")
+        save_weights(ckpt_file, updated_embs, save_dtype)
+        print("model saved.")
+
+
+def save_weights(file, updated_embs, save_dtype):
+    updated_embs = updated_embs.reshape(16, -1, updated_embs.shape[-1])
+    updated_embs = updated_embs.chunk(16)
+    XTI_layers = [
+        "IN01",
+        "IN02",
+        "IN04",
+        "IN05",
+        "IN07",
+        "IN08",
+        "MID",
+        "OUT03",
+        "OUT04",
+        "OUT05",
+        "OUT06",
+        "OUT07",
+        "OUT08",
+        "OUT09",
+        "OUT10",
+        "OUT11",
+    ]
+    state_dict = {}
+    for i, layer_name in enumerate(XTI_layers):
+        state_dict[layer_name] = updated_embs[i].squeeze(0).detach().clone().to("cpu").to(save_dtype)
+
+    # if save_dtype is not None:
+    #     for key in list(state_dict.keys()):
+    #         v = state_dict[key]
+    #         v = v.detach().clone().to("cpu").to(save_dtype)
+    #         state_dict[key] = v
+
+    if os.path.splitext(file)[1] == ".safetensors":
+        from safetensors.torch import save_file
+
+        save_file(state_dict, file)
+    else:
+        torch.save(state_dict, file)  # can be loaded in Web UI
+
+
+def load_weights(file):
+    if os.path.splitext(file)[1] == ".safetensors":
+        from safetensors.torch import load_file
+
+        data = load_file(file)
+    else:
+        raise ValueError(f"NOT XTI: {file}")
+
+    if len(data.values()) != 16:
+        raise ValueError(f"NOT XTI: {file}")
+
+    emb = torch.concat([x for x in data.values()])
+
+    return emb
+
+
+def setup_parser() -> argparse.ArgumentParser:
+    parser = argparse.ArgumentParser()
+
+    train_util.add_sd_models_arguments(parser)
+    train_util.add_dataset_arguments(parser, True, True, False)
+    train_util.add_training_arguments(parser, True)
+    train_util.add_optimizer_arguments(parser)
+    config_util.add_config_arguments(parser)
+    custom_train_functions.add_custom_train_arguments(parser)
+
+    parser.add_argument(
+        "--save_model_as",
+        type=str,
+        default="pt",
+        choices=[None, "ckpt", "pt", "safetensors"],
+        help="format to save the model (default is .pt) / モデル保存時の形式（デフォルトはpt）",
+    )
+
+    parser.add_argument("--weights", type=str, default=None, help="embedding weights to initialize / 学習するネットワークの初期重み")
+    parser.add_argument(
+        "--num_vectors_per_token", type=int, default=1, help="number of vectors per token / トークンに割り当てるembeddingsの要素数"
+    )
+    parser.add_argument(
+        "--token_string",
+        type=str,
+        default=None,
+        help="token string used in training, must not exist in tokenizer / 学習時に使用されるトークン文字列、tokenizerに存在しない文字であること",
+    )
+    parser.add_argument("--init_word", type=str, default=None, help="words to initialize vector / ベクトルを初期化に使用する単語、複数可")
+    parser.add_argument(
+        "--use_object_template",
+        action="store_true",
+        help="ignore caption and use default templates for object / キャプションは使わずデフォルトの物体用テンプレートで学習する",
+    )
+    parser.add_argument(
+        "--use_style_template",
+        action="store_true",
+        help="ignore caption and use default templates for stype / キャプションは使わずデフォルトのスタイル用テンプレートで学習する",
+    )
+
+    return parser
+
+
+if __name__ == "__main__":
+    parser = setup_parser()
+
+    args = parser.parse_args()
+    args = train_util.read_config_from_file(args, parser)
+
+    train(args)