[SD3 dreambooth-lora training] small updates + bug fixes (#9682)

* add latent caching + smol updates

* update license

* replace with free_memory

* add --upcast_before_saving to allow saving transformer weights in lower precision

* fix models to accumulate

* fix mixed precision issue as proposed in https://github.com/huggingface/diffusers/pull/9565



* smol update to readme

* style

* fix caching latents

* style

* add tests for latent caching

* style

* fix latent caching

---------
Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>

examples/dreambooth/README_sd3.md

@@ -136,7 +136,7 @@ accelerate launch train_dreambooth_lora_sd3.py \
   --resolution=512 \
   --train_batch_size=1 \
   --gradient_accumulation_steps=4 \
-  --learning_rate=1e-5 \
+  --learning_rate=4e-4 \
   --report_to="wandb" \
   --lr_scheduler="constant" \
   --lr_warmup_steps=0 \

examples/dreambooth/test_dreambooth_lora_sd3.py

@@ -103,6 +103,39 @@ class DreamBoothLoRASD3(ExamplesTestsAccelerate):
             )
             self.assertTrue(starts_with_expected_prefix)
 
+    def test_dreambooth_lora_latent_caching(self):
+        with tempfile.TemporaryDirectory() as tmpdir:
+            test_args = f"""
+                {self.script_path}
+                --pretrained_model_name_or_path {self.pretrained_model_name_or_path}
+                --instance_data_dir {self.instance_data_dir}
+                --instance_prompt {self.instance_prompt}
+                --resolution 64
+                --train_batch_size 1
+                --gradient_accumulation_steps 1
+                --max_train_steps 2
+                --cache_latents
+                --learning_rate 5.0e-04
+                --scale_lr
+                --lr_scheduler constant
+                --lr_warmup_steps 0
+                --output_dir {tmpdir}
+                """.split()
+
+            run_command(self._launch_args + test_args)
+            # save_pretrained smoke test
+            self.assertTrue(os.path.isfile(os.path.join(tmpdir, "pytorch_lora_weights.safetensors")))
+
+            # make sure the state_dict has the correct naming in the parameters.
+            lora_state_dict = safetensors.torch.load_file(os.path.join(tmpdir, "pytorch_lora_weights.safetensors"))
+            is_lora = all("lora" in k for k in lora_state_dict.keys())
+            self.assertTrue(is_lora)
+
+            # when not training the text encoder, all the parameters in the state dict should start
+            # with `"transformer"` in their names.
+            starts_with_transformer = all(key.startswith("transformer") for key in lora_state_dict.keys())
+            self.assertTrue(starts_with_transformer)
+
     def test_dreambooth_lora_sd3_checkpointing_checkpoints_total_limit(self):
         with tempfile.TemporaryDirectory() as tmpdir:
             test_args = f"""

examples/dreambooth/train_dreambooth_lora_sd3.py

@@ -140,7 +140,7 @@ Please adhere to the licensing terms as described [here](https://huggingface.co/
     model_card = load_or_create_model_card(
         repo_id_or_path=repo_id,
         from_training=True,
-        license="openrail++",
+        license="other",
         base_model=base_model,
         prompt=instance_prompt,
         model_description=model_description,
@@ -186,7 +186,7 @@ def log_validation(
         f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
         f" {args.validation_prompt}."
     )
-    pipeline = pipeline.to(accelerator.device, dtype=torch_dtype)
+    pipeline = pipeline.to(accelerator.device)
     pipeline.set_progress_bar_config(disable=True)
 
     # run inference
@@ -608,6 +608,12 @@ def parse_args(input_args=None):
             " https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices"
         ),
     )
+    parser.add_argument(
+        "--cache_latents",
+        action="store_true",
+        default=False,
+        help="Cache the VAE latents",
+    )
     parser.add_argument(
         "--report_to",
         type=str,
@@ -628,6 +634,15 @@ def parse_args(input_args=None):
             " flag passed with the `accelerate.launch` command. Use this argument to override the accelerate config."
         ),
     )
+    parser.add_argument(
+        "--upcast_before_saving",
+        action="store_true",
+        default=False,
+        help=(
+            "Whether to upcast the trained transformer layers to float32 before saving (at the end of training). "
+            "Defaults to precision dtype used for training to save memory"
+        ),
+    )
     parser.add_argument(
         "--prior_generation_precision",
         type=str,
@@ -1394,6 +1409,16 @@ def main(args):
             logger.warning(
                 "Learning rate is too low. When using prodigy, it's generally better to set learning rate around 1.0"
             )
+        if args.train_text_encoder and args.text_encoder_lr:
+            logger.warning(
+                f"Learning rates were provided both for the transformer and the text encoder- e.g. text_encoder_lr:"
+                f" {args.text_encoder_lr} and learning_rate: {args.learning_rate}. "
+                f"When using prodigy only learning_rate is used as the initial learning rate."
+            )
+            # changes the learning rate of text_encoder_parameters_one and text_encoder_parameters_two to be
+            # --learning_rate
+            params_to_optimize[1]["lr"] = args.learning_rate
+            params_to_optimize[2]["lr"] = args.learning_rate
 
         optimizer = optimizer_class(
             params_to_optimize,
@@ -1440,6 +1465,9 @@ def main(args):
                 pooled_prompt_embeds = pooled_prompt_embeds.to(accelerator.device)
             return prompt_embeds, pooled_prompt_embeds
 
+    # If no type of tuning is done on the text_encoder and custom instance prompts are NOT
+    # provided (i.e. the --instance_prompt is used for all images), we encode the instance prompt once to avoid
+    # the redundant encoding.
     if not args.train_text_encoder and not train_dataset.custom_instance_prompts:
         instance_prompt_hidden_states, instance_pooled_prompt_embeds = compute_text_embeddings(
             args.instance_prompt, text_encoders, tokenizers
@@ -1484,6 +1512,21 @@ def main(args):
                 tokens_two = torch.cat([tokens_two, class_tokens_two], dim=0)
                 tokens_three = torch.cat([tokens_three, class_tokens_three], dim=0)
 
+    vae_config_shift_factor = vae.config.shift_factor
+    vae_config_scaling_factor = vae.config.scaling_factor
+    if args.cache_latents:
+        latents_cache = []
+        for batch in tqdm(train_dataloader, desc="Caching latents"):
+            with torch.no_grad():
+                batch["pixel_values"] = batch["pixel_values"].to(
+                    accelerator.device, non_blocking=True, dtype=weight_dtype
+                )
+                latents_cache.append(vae.encode(batch["pixel_values"]).latent_dist)
+
+        if args.validation_prompt is None:
+            del vae
+            free_memory()
+
     # Scheduler and math around the number of training steps.
     overrode_max_train_steps = False
     num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
@@ -1500,7 +1543,6 @@ def main(args):
         power=args.lr_power,
     )
 
-    # Prepare everything with our `accelerator`.
     # Prepare everything with our `accelerator`.
     if args.train_text_encoder:
         (
@@ -1607,8 +1649,9 @@ def main(args):
 
         for step, batch in enumerate(train_dataloader):
             models_to_accumulate = [transformer]
+            if args.train_text_encoder:
+                models_to_accumulate.extend([text_encoder_one, text_encoder_two])
             with accelerator.accumulate(models_to_accumulate):
-                pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
                 prompts = batch["prompts"]
 
                 # encode batch prompts when custom prompts are provided for each image -
@@ -1639,8 +1682,13 @@ def main(args):
                         )
 
                 # Convert images to latent space
-                model_input = vae.encode(pixel_values).latent_dist.sample()
-                model_input = (model_input - vae.config.shift_factor) * vae.config.scaling_factor
+                if args.cache_latents:
+                    model_input = latents_cache[step].sample()
+                else:
+                    pixel_values = batch["pixel_values"].to(dtype=vae.dtype)
+                    model_input = vae.encode(pixel_values).latent_dist.sample()
+
+                model_input = (model_input - vae_config_shift_factor) * vae_config_scaling_factor
                 model_input = model_input.to(dtype=weight_dtype)
 
                 # Sample noise that we'll add to the latents
@@ -1773,6 +1821,8 @@ def main(args):
                     text_encoder_one, text_encoder_two, text_encoder_three = load_text_encoders(
                         text_encoder_cls_one, text_encoder_cls_two, text_encoder_cls_three
                     )
+                    text_encoder_one.to(weight_dtype)
+                    text_encoder_two.to(weight_dtype)
                 pipeline = StableDiffusion3Pipeline.from_pretrained(
                     args.pretrained_model_name_or_path,
                     vae=vae,
@@ -1793,15 +1843,18 @@ def main(args):
                     epoch=epoch,
                     torch_dtype=weight_dtype,
                 )
-
-                del text_encoder_one, text_encoder_two, text_encoder_three
-                free_memory()
+                if not args.train_text_encoder:
+                    del text_encoder_one, text_encoder_two, text_encoder_three
+                    free_memory()
 
     # Save the lora layers
     accelerator.wait_for_everyone()
     if accelerator.is_main_process:
         transformer = unwrap_model(transformer)
-        transformer = transformer.to(torch.float32)
+        if args.upcast_before_saving:
+            transformer.to(torch.float32)
+        else:
+            transformer = transformer.to(weight_dtype)
         transformer_lora_layers = get_peft_model_state_dict(transformer)
 
         if args.train_text_encoder:

examples/dreambooth/train_dreambooth_sd3.py

@@ -15,7 +15,6 @@
 
 import argparse
 import copy
-import gc
 import itertools
 import logging
 import math
@@ -51,7 +50,7 @@ from diffusers import (
     StableDiffusion3Pipeline,
 )
 from diffusers.optimization import get_scheduler
-from diffusers.training_utils import compute_density_for_timestep_sampling, compute_loss_weighting_for_sd3
+from diffusers.training_utils import compute_density_for_timestep_sampling, compute_loss_weighting_for_sd3, free_memory
 from diffusers.utils import (
     check_min_version,
     is_wandb_available,
@@ -119,7 +118,7 @@ Please adhere to the licensing terms as described `[here](https://huggingface.co
     model_card = load_or_create_model_card(
         repo_id_or_path=repo_id,
         from_training=True,
-        license="openrail++",
+        license="other",
         base_model=base_model,
         prompt=instance_prompt,
         model_description=model_description,
@@ -164,7 +163,7 @@ def log_validation(
         f"Running validation... \n Generating {args.num_validation_images} images with prompt:"
         f" {args.validation_prompt}."
     )
-    pipeline = pipeline.to(accelerator.device, dtype=torch_dtype)
+    pipeline = pipeline.to(accelerator.device)
     pipeline.set_progress_bar_config(disable=True)
 
     # run inference
@@ -190,8 +189,7 @@ def log_validation(
             )
 
     del pipeline
-    if torch.cuda.is_available():
-        torch.cuda.empty_cache()
+    free_memory()
 
     return images
 
@@ -1065,8 +1063,7 @@ def main(args):
                     image.save(image_filename)
 
             del pipeline
-            if torch.cuda.is_available():
-                torch.cuda.empty_cache()
+            free_memory()
 
     # Handle the repository creation
     if accelerator.is_main_process:
@@ -1386,9 +1383,7 @@ def main(args):
         del tokenizers, text_encoders
         # Explicitly delete the objects as well, otherwise only the lists are deleted and the original references remain, preventing garbage collection
         del text_encoder_one, text_encoder_two, text_encoder_three
-        gc.collect()
-        if torch.cuda.is_available():
-            torch.cuda.empty_cache()
+        free_memory()
 
     # If custom instance prompts are NOT provided (i.e. the instance prompt is used for all images),
     # pack the statically computed variables appropriately here. This is so that we don't
@@ -1708,6 +1703,9 @@ def main(args):
                     text_encoder_one, text_encoder_two, text_encoder_three = load_text_encoders(
                         text_encoder_cls_one, text_encoder_cls_two, text_encoder_cls_three
                     )
+                    text_encoder_one.to(weight_dtype)
+                    text_encoder_two.to(weight_dtype)
+                    text_encoder_three.to(weight_dtype)
                 pipeline = StableDiffusion3Pipeline.from_pretrained(
                     args.pretrained_model_name_or_path,
                     vae=vae,
@@ -1730,8 +1728,7 @@ def main(args):
                 )
                 if not args.train_text_encoder:
                     del text_encoder_one, text_encoder_two, text_encoder_three
-                    torch.cuda.empty_cache()
-                    gc.collect()
+                    free_memory()
 
     # Save the lora layers
     accelerator.wait_for_everyone()