Merge pull request #1710 from kohya-ss/diff_output_prsv

Differential Output Preservation loss for LoRA

README.md

@@ -11,6 +11,23 @@ The command to install PyTorch is as follows:
 
 ### Recent Updates
 
+Oct 19, 2024:
+
+- Added an implementation of Differential Output Preservation (temporary name) for SDXL/FLUX.1 LoRA training. SD1/2 is not tested yet. This is an experimental feature. 
+  - A method to make the output of LoRA closer to the output when LoRA is not applied, with captions that do not contain trigger words.
+  - Define a Dataset subset for the regularization image (`is_reg = true`) with `.toml`. Add `custom_attributes.diff_output_preservation = true`.
+    - See [dataset configuration](docs/config_README-en.md) for the regularization dataset.
+  - Specify "number of training images x number of repeats >= number of regularization images x number of repeats".
+  - Specify a large value for `--prior_loss_weight` option (not dataset config). The appropriate value is unknown, but try around 10-100. Note that the default is 1.0.
+  - You may want to start with 2/3 to 3/4 of the loss value when DOP is not applied. If it is 1/2, DOP may not be working.
+```
+[[datasets.subsets]]
+image_dir = "path/to/image/dir"
+num_repeats = 1
+is_reg = true
+custom_attributes.diff_output_preservation = true # Add this
+```
+
 Oct 13, 2024:
 
 - Fixed an issue where it took a long time to load the image size when initializing the dataset, especially when the number of images in the dataset was large.

flux_train_network.py

@@ -373,33 +373,13 @@ def get_noise_pred_and_target(
         if not args.apply_t5_attn_mask:
             t5_attn_mask = None
 
-        if not args.split_mode:
-            # normal forward
-            with accelerator.autocast():
-                # YiYi notes: divide it by 1000 for now because we scale it by 1000 in the transformer model (we should not keep it but I want to keep the inputs same for the model for testing)
-                model_pred = unet(
-                    img=packed_noisy_model_input,
-                    img_ids=img_ids,
-                    txt=t5_out,
-                    txt_ids=txt_ids,
-                    y=l_pooled,
-                    timesteps=timesteps / 1000,
-                    guidance=guidance_vec,
-                    txt_attention_mask=t5_attn_mask,
-                )
-        else:
-            # split forward to reduce memory usage
-            assert network.train_blocks == "single", "train_blocks must be single for split mode"
-            with accelerator.autocast():
-                # move flux lower to cpu, and then move flux upper to gpu
-                unet.to("cpu")
-                clean_memory_on_device(accelerator.device)
-                self.flux_upper.to(accelerator.device)
-
-                # upper model does not require grad
-                with torch.no_grad():
-                    intermediate_img, intermediate_txt, vec, pe = self.flux_upper(
-                        img=packed_noisy_model_input,
+        def call_dit(img, img_ids, t5_out, txt_ids, l_pooled, timesteps, guidance_vec, t5_attn_mask):
+            if not args.split_mode:
+                # normal forward
+                with accelerator.autocast():
+                    # YiYi notes: divide it by 1000 for now because we scale it by 1000 in the transformer model (we should not keep it but I want to keep the inputs same for the model for testing)
+                    model_pred = unet(
+                        img=img,
                         img_ids=img_ids,
                         txt=t5_out,
                         txt_ids=txt_ids,
@@ -408,18 +388,52 @@ def get_noise_pred_and_target(
                         guidance=guidance_vec,
                         txt_attention_mask=t5_attn_mask,
                     )
-
-                # move flux upper back to cpu, and then move flux lower to gpu
-                self.flux_upper.to("cpu")
-                clean_memory_on_device(accelerator.device)
-                unet.to(accelerator.device)
-
-                # lower model requires grad
-                intermediate_img.requires_grad_(True)
-                intermediate_txt.requires_grad_(True)
-                vec.requires_grad_(True)
-                pe.requires_grad_(True)
-                model_pred = unet(img=intermediate_img, txt=intermediate_txt, vec=vec, pe=pe, txt_attention_mask=t5_attn_mask)
+            else:
+                # split forward to reduce memory usage
+                assert network.train_blocks == "single", "train_blocks must be single for split mode"
+                with accelerator.autocast():
+                    # move flux lower to cpu, and then move flux upper to gpu
+                    unet.to("cpu")
+                    clean_memory_on_device(accelerator.device)
+                    self.flux_upper.to(accelerator.device)
+
+                    # upper model does not require grad
+                    with torch.no_grad():
+                        intermediate_img, intermediate_txt, vec, pe = self.flux_upper(
+                            img=packed_noisy_model_input,
+                            img_ids=img_ids,
+                            txt=t5_out,
+                            txt_ids=txt_ids,
+                            y=l_pooled,
+                            timesteps=timesteps / 1000,
+                            guidance=guidance_vec,
+                            txt_attention_mask=t5_attn_mask,
+                        )
+
+                    # move flux upper back to cpu, and then move flux lower to gpu
+                    self.flux_upper.to("cpu")
+                    clean_memory_on_device(accelerator.device)
+                    unet.to(accelerator.device)
+
+                    # lower model requires grad
+                    intermediate_img.requires_grad_(True)
+                    intermediate_txt.requires_grad_(True)
+                    vec.requires_grad_(True)
+                    pe.requires_grad_(True)
+                    model_pred = unet(img=intermediate_img, txt=intermediate_txt, vec=vec, pe=pe, txt_attention_mask=t5_attn_mask)
+
+            return model_pred
+
+        model_pred = call_dit(
+            img=packed_noisy_model_input,
+            img_ids=img_ids,
+            t5_out=t5_out,
+            txt_ids=txt_ids,
+            l_pooled=l_pooled,
+            timesteps=timesteps,
+            guidance_vec=guidance_vec,
+            t5_attn_mask=t5_attn_mask,
+        )
 
         # unpack latents
         model_pred = flux_utils.unpack_latents(model_pred, packed_latent_height, packed_latent_width)
@@ -430,6 +444,37 @@ def get_noise_pred_and_target(
         # flow matching loss: this is different from SD3
         target = noise - latents
 
+        # differential output preservation
+        if "custom_attributes" in batch:
+            diff_output_pr_indices = []
+            for i, custom_attributes in enumerate(batch["custom_attributes"]):
+                if "diff_output_preservation" in custom_attributes and custom_attributes["diff_output_preservation"]:
+                    diff_output_pr_indices.append(i)
+
+            if len(diff_output_pr_indices) > 0:
+                network.set_multiplier(0.0)
+                with torch.no_grad():
+                    model_pred_prior = call_dit(
+                        img=packed_noisy_model_input[diff_output_pr_indices],
+                        img_ids=img_ids[diff_output_pr_indices],
+                        t5_out=t5_out[diff_output_pr_indices],
+                        txt_ids=txt_ids[diff_output_pr_indices],
+                        l_pooled=l_pooled[diff_output_pr_indices],
+                        timesteps=timesteps[diff_output_pr_indices],
+                        guidance_vec=guidance_vec[diff_output_pr_indices] if guidance_vec is not None else None,
+                        t5_attn_mask=t5_attn_mask[diff_output_pr_indices] if t5_attn_mask is not None else None,
+                    )
+                network.set_multiplier(1.0)  # may be overwritten by "network_multipliers" in the next step
+
+                model_pred_prior = flux_utils.unpack_latents(model_pred_prior, packed_latent_height, packed_latent_width)
+                model_pred_prior, _ = flux_train_utils.apply_model_prediction_type(
+                    args,
+                    model_pred_prior,
+                    noisy_model_input[diff_output_pr_indices],
+                    sigmas[diff_output_pr_indices] if sigmas is not None else None,
+                )
+                target[diff_output_pr_indices] = model_pred_prior.to(target.dtype)
+
         return model_pred, target, timesteps, None, weighting
 
     def post_process_loss(self, loss, args, timesteps, noise_scheduler):

library/config_util.py

@@ -10,13 +10,7 @@
 from pathlib import Path
 
 # from toolz import curry
-from typing import (
-    List,
-    Optional,
-    Sequence,
-    Tuple,
-    Union,
-)
+from typing import Dict, List, Optional, Sequence, Tuple, Union
 
 import toml
 import voluptuous
@@ -78,6 +72,7 @@ class BaseSubsetParams:
     caption_tag_dropout_rate: float = 0.0
     token_warmup_min: int = 1
     token_warmup_step: float = 0
+    custom_attributes: Optional[Dict[str, Any]] = None
 
 
 @dataclass
@@ -197,6 +192,7 @@ def __validate_and_convert_scalar_or_twodim(klass, value: Union[float, Sequence]
         "token_warmup_step": Any(float, int),
         "caption_prefix": str,
         "caption_suffix": str,
+        "custom_attributes": dict,
     }
     # DO means DropOut
     DO_SUBSET_ASCENDABLE_SCHEMA = {
@@ -538,9 +534,10 @@ 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},
-          alpha_mask: {subset.alpha_mask},
+          token_warmup_min: {subset.token_warmup_min}
+          token_warmup_step: {subset.token_warmup_step}
+          alpha_mask: {subset.alpha_mask}
+          custom_attributes: {subset.custom_attributes}
       """
                 ),
                 "  ",

library/train_util.py

@@ -396,6 +396,7 @@ def __init__(
         caption_suffix: Optional[str],
         token_warmup_min: int,
         token_warmup_step: Union[float, int],
+        custom_attributes: Optional[Dict[str, Any]] = None,
     ) -> None:
         self.image_dir = image_dir
         self.alpha_mask = alpha_mask if alpha_mask is not None else False
@@ -419,6 +420,8 @@ def __init__(
         self.token_warmup_min = token_warmup_min  # step=0におけるタグの数
         self.token_warmup_step = token_warmup_step  # N（N<1ならN*max_train_steps）ステップ目でタグの数が最大になる
 
+        self.custom_attributes = custom_attributes if custom_attributes is not None else {}
+
         self.img_count = 0
 
 
@@ -449,6 +452,7 @@ def __init__(
         caption_suffix,
         token_warmup_min,
         token_warmup_step,
+        custom_attributes: Optional[Dict[str, Any]] = None,
     ) -> None:
         assert image_dir is not None, "image_dir must be specified / image_dirは指定が必須です"
 
@@ -473,6 +477,7 @@ def __init__(
             caption_suffix,
             token_warmup_min,
             token_warmup_step,
+            custom_attributes=custom_attributes,
         )
 
         self.is_reg = is_reg
@@ -512,6 +517,7 @@ def __init__(
         caption_suffix,
         token_warmup_min,
         token_warmup_step,
+        custom_attributes: Optional[Dict[str, Any]] = None,
     ) -> None:
         assert metadata_file is not None, "metadata_file must be specified / metadata_fileは指定が必須です"
 
@@ -536,6 +542,7 @@ def __init__(
             caption_suffix,
             token_warmup_min,
             token_warmup_step,
+            custom_attributes=custom_attributes,
         )
 
         self.metadata_file = metadata_file
@@ -1474,11 +1481,14 @@ def __getitem__(self, index):
         target_sizes_hw = []
         flippeds = []  # 変数名が微妙
         text_encoder_outputs_list = []
+        custom_attributes = []
 
         for image_key in bucket[image_index : image_index + bucket_batch_size]:
             image_info = self.image_data[image_key]
             subset = self.image_to_subset[image_key]
 
+            custom_attributes.append(subset.custom_attributes)
+
             # in case of fine tuning, is_reg is always False
             loss_weights.append(self.prior_loss_weight if image_info.is_reg else 1.0)
 
@@ -1646,7 +1656,9 @@ def none_or_stack_elements(tensors_list, converter):
                 return None
             return [torch.stack([converter(x[i]) for x in tensors_list]) for i in range(len(tensors_list[0]))]
 
+        # set example
         example = {}
+        example["custom_attributes"] = custom_attributes  # may be list of empty dict
         example["loss_weights"] = torch.FloatTensor(loss_weights)
         example["text_encoder_outputs_list"] = none_or_stack_elements(text_encoder_outputs_list, torch.FloatTensor)
         example["input_ids_list"] = none_or_stack_elements(input_ids_list, lambda x: x)
@@ -2630,7 +2642,9 @@ def debug_dataset(train_dataset, show_input_ids=False):
                     f'{ik}, size: {train_dataset.image_data[ik].image_size}, loss weight: {lw}, caption: "{cap}", original size: {orgsz}, crop top left: {crptl}, target size: {trgsz}, flipped: {flpdz}'
                 )
                 if "network_multipliers" in example:
-                    print(f"network multiplier: {example['network_multipliers'][j]}")
+                    logger.info(f"network multiplier: {example['network_multipliers'][j]}")
+                if "custom_attributes" in example:
+                    logger.info(f"custom attributes: {example['custom_attributes'][j]}")
 
                 # if show_input_ids:
                 #     logger.info(f"input ids: {iid}")
@@ -4091,6 +4105,7 @@ def enable_high_vram(args: argparse.Namespace):
         global HIGH_VRAM
         HIGH_VRAM = True
 
+
 def verify_training_args(args: argparse.Namespace):
     r"""
     Verify training arguments. Also reflect highvram option to global variable

sdxl_train_network.py

@@ -1,4 +1,5 @@
 import argparse
+from typing import List, Optional
 
 import torch
 from accelerate import Accelerator
@@ -172,7 +173,18 @@ def get_text_cond(self, args, accelerator, batch, tokenizers, text_encoders, wei
 
         return encoder_hidden_states1, encoder_hidden_states2, pool2
 
-    def call_unet(self, args, accelerator, unet, noisy_latents, timesteps, text_conds, batch, weight_dtype):
+    def call_unet(
+        self,
+        args,
+        accelerator,
+        unet,
+        noisy_latents,
+        timesteps,
+        text_conds,
+        batch,
+        weight_dtype,
+        indices: Optional[List[int]] = None,
+    ):
         noisy_latents = noisy_latents.to(weight_dtype)  # TODO check why noisy_latents is not weight_dtype
 
         # get size embeddings
@@ -186,6 +198,12 @@ def call_unet(self, args, accelerator, unet, noisy_latents, timesteps, text_cond
         vector_embedding = torch.cat([pool2, embs], dim=1).to(weight_dtype)
         text_embedding = torch.cat([encoder_hidden_states1, encoder_hidden_states2], dim=2).to(weight_dtype)
 
+        if indices is not None and len(indices) > 0:
+            noisy_latents = noisy_latents[indices]
+            timesteps = timesteps[indices]
+            text_embedding = text_embedding[indices]
+            vector_embedding = vector_embedding[indices]
+
         noise_pred = unet(noisy_latents, timesteps, text_embedding, vector_embedding)
         return noise_pred