Fix training

5 files changed, 1073 insertions, 15 deletions

diff --git a/pipelines/util.py b/pipelines/util.py
deleted file mode 100644
index 661dbee..0000000
--- a/pipelines/util.py
+++ /dev/null
@@ -1,9 +0,0 @@
-import torch
-
-
-def set_use_memory_efficient_attention_xformers(module: torch.nn.Module, valid: bool) -> None:
-    if hasattr(module, "set_use_memory_efficient_attention_xformers"):
-        module.set_use_memory_efficient_attention_xformers(valid)
-
-    for child in module.children():
-        set_use_memory_efficient_attention_xformers(child, valid)
diff --git a/train_dreambooth.py b/train_dreambooth.py
index 3eecf9c..0f8fece 100644
--- a/train_dreambooth.py
+++ b/train_dreambooth.py
@@ -23,7 +23,6 @@ from slugify import slugify
 
 from common import load_text_embeddings
 from pipelines.stable_diffusion.vlpn_stable_diffusion import VlpnStableDiffusion
-from pipelines.util import set_use_memory_efficient_attention_xformers
 from data.csv import CSVDataModule
 from training.optimization import get_one_cycle_schedule
 from models.clip.prompt import PromptProcessor
@@ -614,8 +613,8 @@ def main():
         args.pretrained_model_name_or_path, subfolder='scheduler')
 
     vae.enable_slicing()
-    set_use_memory_efficient_attention_xformers(unet, True)
-    set_use_memory_efficient_attention_xformers(vae, True)
+    vae.set_use_memory_efficient_attention_xformers(True)
+    unet.set_use_memory_efficient_attention_xformers(True)
 
     if args.gradient_checkpointing:
         unet.enable_gradient_checkpointing()
@@ -964,6 +963,7 @@ def main():
                     # 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)
+                    noisy_latents = noisy_latents.to(dtype=unet.dtype)
 
                     # Get the text embedding for conditioning
                     encoder_hidden_states = prompt_processor.get_embeddings(batch["input_ids"], batch["attention_mask"])
@@ -1065,6 +1065,7 @@ def main():
                     timesteps = timesteps.long()
 
                     noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+                    noisy_latents = noisy_latents.to(dtype=unet.dtype)
 
                     encoder_hidden_states = prompt_processor.get_embeddings(batch["input_ids"], batch["attention_mask"])
 
diff --git a/train_lora.py b/train_lora.py
new file mode 100644
index 0000000..ffc1d10
--- /dev/null
+++ b/train_lora.py
@@ -0,0 +1,1040 @@
+import argparse
+import itertools
+import math
+import datetime
+import logging
+import json
+from pathlib import Path
+
+import torch
+import torch.nn.functional as F
+import torch.utils.checkpoint
+
+from accelerate import Accelerator
+from accelerate.logging import get_logger
+from accelerate.utils import LoggerType, set_seed
+from diffusers import AutoencoderKL, DDPMScheduler, DPMSolverMultistepScheduler, UNet2DConditionModel
+from diffusers.optimization import get_scheduler, get_cosine_with_hard_restarts_schedule_with_warmup
+from diffusers.training_utils import EMAModel
+from PIL import Image
+from tqdm.auto import tqdm
+from transformers import CLIPTextModel, CLIPTokenizer
+from slugify import slugify
+
+from common import load_text_embeddings
+from pipelines.stable_diffusion.vlpn_stable_diffusion import VlpnStableDiffusion
+from data.csv import CSVDataModule
+from training.lora import LoraAttention
+from training.optimization import get_one_cycle_schedule
+from models.clip.prompt import PromptProcessor
+
+logger = get_logger(__name__)
+
+
+torch.backends.cuda.matmul.allow_tf32 = True
+torch.backends.cudnn.benchmark = True
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description="Simple example of a training script."
+    )
+    parser.add_argument(
+        "--pretrained_model_name_or_path",
+        type=str,
+        default=None,
+        help="Path to pretrained model or model identifier from huggingface.co/models.",
+    )
+    parser.add_argument(
+        "--tokenizer_name",
+        type=str,
+        default=None,
+        help="Pretrained tokenizer name or path if not the same as model_name",
+    )
+    parser.add_argument(
+        "--train_data_file",
+        type=str,
+        default=None,
+        help="A folder containing the training data."
+    )
+    parser.add_argument(
+        "--train_data_template",
+        type=str,
+        default="template",
+    )
+    parser.add_argument(
+        "--instance_identifier",
+        type=str,
+        default=None,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--class_identifier",
+        type=str,
+        default=None,
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--placeholder_token",
+        type=str,
+        nargs='*',
+        default=[],
+        help="A token to use as a placeholder for the concept.",
+    )
+    parser.add_argument(
+        "--initializer_token",
+        type=str,
+        nargs='*',
+        default=[],
+        help="A token to use as initializer word."
+    )
+    parser.add_argument(
+        "--tag_dropout",
+        type=float,
+        default=0.1,
+        help="Tag dropout probability.",
+    )
+    parser.add_argument(
+        "--num_class_images",
+        type=int,
+        default=400,
+        help="How many class images to generate."
+    )
+    parser.add_argument(
+        "--repeats",
+        type=int,
+        default=1,
+        help="How many times to repeat the training data."
+    )
+    parser.add_argument(
+        "--output_dir",
+        type=str,
+        default="output/dreambooth",
+        help="The output directory where the model predictions and checkpoints will be written.",
+    )
+    parser.add_argument(
+        "--embeddings_dir",
+        type=str,
+        default=None,
+        help="The embeddings directory where Textual Inversion embeddings are stored.",
+    )
+    parser.add_argument(
+        "--mode",
+        type=str,
+        default=None,
+        help="A mode to filter the dataset.",
+    )
+    parser.add_argument(
+        "--seed",
+        type=int,
+        default=None,
+        help="A seed for reproducible training."
+    )
+    parser.add_argument(
+        "--resolution",
+        type=int,
+        default=768,
+        help=(
+            "The resolution for input images, all the images in the train/validation dataset will be resized to this"
+            " resolution"
+        ),
+    )
+    parser.add_argument(
+        "--center_crop",
+        action="store_true",
+        help="Whether to center crop images before resizing to resolution"
+    )
+    parser.add_argument(
+        "--dataloader_num_workers",
+        type=int,
+        default=0,
+        help=(
+            "The number of subprocesses to use for data loading. 0 means that the data will be loaded in the main"
+            " process."
+        ),
+    )
+    parser.add_argument(
+        "--num_train_epochs",
+        type=int,
+        default=100
+    )
+    parser.add_argument(
+        "--max_train_steps",
+        type=int,
+        default=None,
+        help="Total number of training steps to perform.  If provided, overrides num_train_epochs.",
+    )
+    parser.add_argument(
+        "--gradient_accumulation_steps",
+        type=int,
+        default=1,
+        help="Number of updates steps to accumulate before performing a backward/update pass.",
+    )
+    parser.add_argument(
+        "--gradient_checkpointing",
+        action="store_true",
+        help="Whether or not to use gradient checkpointing to save memory at the expense of slower backward pass.",
+    )
+    parser.add_argument(
+        "--learning_rate_unet",
+        type=float,
+        default=2e-6,
+        help="Initial learning rate (after the potential warmup period) to use.",
+    )
+    parser.add_argument(
+        "--scale_lr",
+        action="store_true",
+        default=True,
+        help="Scale the learning rate by the number of GPUs, gradient accumulation steps, and batch size.",
+    )
+    parser.add_argument(
+        "--lr_scheduler",
+        type=str,
+        default="one_cycle",
+        help=(
+            'The scheduler type to use. Choose between ["linear", "cosine", "cosine_with_restarts", "polynomial",'
+            ' "constant", "constant_with_warmup", "one_cycle"]'
+        ),
+    )
+    parser.add_argument(
+        "--lr_warmup_epochs",
+        type=int,
+        default=10,
+        help="Number of steps for the warmup in the lr scheduler."
+    )
+    parser.add_argument(
+        "--lr_cycles",
+        type=int,
+        default=None,
+        help="Number of restart cycles in the lr scheduler (if supported)."
+    )
+    parser.add_argument(
+        "--use_8bit_adam",
+        action="store_true",
+        default=True,
+        help="Whether or not to use 8-bit Adam from bitsandbytes."
+    )
+    parser.add_argument(
+        "--adam_beta1",
+        type=float,
+        default=0.9,
+        help="The beta1 parameter for the Adam optimizer."
+    )
+    parser.add_argument(
+        "--adam_beta2",
+        type=float,
+        default=0.999,
+        help="The beta2 parameter for the Adam optimizer."
+    )
+    parser.add_argument(
+        "--adam_weight_decay",
+        type=float,
+        default=1e-2,
+        help="Weight decay to use."
+    )
+    parser.add_argument(
+        "--adam_epsilon",
+        type=float,
+        default=1e-08,
+        help="Epsilon value for the Adam optimizer"
+    )
+    parser.add_argument(
+        "--mixed_precision",
+        type=str,
+        default="no",
+        choices=["no", "fp16", "bf16"],
+        help=(
+            "Whether to use mixed precision. Choose"
+            "between fp16 and bf16 (bfloat16). Bf16 requires PyTorch >= 1.10."
+            "and an Nvidia Ampere GPU."
+        ),
+    )
+    parser.add_argument(
+        "--sample_frequency",
+        type=int,
+        default=1,
+        help="How often to save a checkpoint and sample image",
+    )
+    parser.add_argument(
+        "--sample_image_size",
+        type=int,
+        default=768,
+        help="Size of sample images",
+    )
+    parser.add_argument(
+        "--sample_batches",
+        type=int,
+        default=1,
+        help="Number of sample batches to generate per checkpoint",
+    )
+    parser.add_argument(
+        "--sample_batch_size",
+        type=int,
+        default=1,
+        help="Number of samples to generate per batch",
+    )
+    parser.add_argument(
+        "--valid_set_size",
+        type=int,
+        default=None,
+        help="Number of images in the validation dataset."
+    )
+    parser.add_argument(
+        "--train_batch_size",
+        type=int,
+        default=1,
+        help="Batch size (per device) for the training dataloader."
+    )
+    parser.add_argument(
+        "--sample_steps",
+        type=int,
+        default=15,
+        help="Number of steps for sample generation. Higher values will result in more detailed samples, but longer runtimes.",
+    )
+    parser.add_argument(
+        "--prior_loss_weight",
+        type=float,
+        default=1.0,
+        help="The weight of prior preservation loss."
+    )
+    parser.add_argument(
+        "--max_grad_norm",
+        default=1.0,
+        type=float,
+        help="Max gradient norm."
+    )
+    parser.add_argument(
+        "--noise_timesteps",
+        type=int,
+        default=1000,
+    )
+    parser.add_argument(
+        "--config",
+        type=str,
+        default=None,
+        help="Path to a JSON configuration file containing arguments for invoking this script."
+    )
+
+    args = parser.parse_args()
+    if args.config is not None:
+        with open(args.config, 'rt') as f:
+            args = parser.parse_args(
+                namespace=argparse.Namespace(**json.load(f)["args"]))
+
+    if args.train_data_file is None:
+        raise ValueError("You must specify --train_data_file")
+
+    if args.pretrained_model_name_or_path is None:
+        raise ValueError("You must specify --pretrained_model_name_or_path")
+
+    if args.instance_identifier is None:
+        raise ValueError("You must specify --instance_identifier")
+
+    if isinstance(args.initializer_token, str):
+        args.initializer_token = [args.initializer_token]
+
+    if isinstance(args.placeholder_token, str):
+        args.placeholder_token = [args.placeholder_token]
+
+    if len(args.placeholder_token) == 0:
+        args.placeholder_token = [f"<*{i}>" for i in range(len(args.initializer_token))]
+
+    if len(args.placeholder_token) != len(args.initializer_token):
+        raise ValueError("Number of items in --placeholder_token and --initializer_token must match")
+
+    if args.output_dir is None:
+        raise ValueError("You must specify --output_dir")
+
+    return args
+
+
+def save_args(basepath: Path, args, extra={}):
+    info = {"args": vars(args)}
+    info["args"].update(extra)
+    with open(basepath.joinpath("args.json"), "w") as f:
+        json.dump(info, f, indent=4)
+
+
+def freeze_params(params):
+    for param in params:
+        param.requires_grad = False
+
+
+def make_grid(images, rows, cols):
+    w, h = images[0].size
+    grid = Image.new('RGB', size=(cols*w, rows*h))
+    for i, image in enumerate(images):
+        grid.paste(image, box=(i % cols*w, i//cols*h))
+    return grid
+
+
+class AverageMeter:
+    def __init__(self, name=None):
+        self.name = name
+        self.reset()
+
+    def reset(self):
+        self.sum = self.count = self.avg = 0
+
+    def update(self, val, n=1):
+        self.sum += val * n
+        self.count += n
+        self.avg = self.sum / self.count
+
+
+class Checkpointer:
+    def __init__(
+        self,
+        datamodule,
+        accelerator,
+        vae,
+        unet,
+        unet_lora,
+        tokenizer,
+        text_encoder,
+        scheduler,
+        output_dir: Path,
+        instance_identifier,
+        placeholder_token,
+        placeholder_token_id,
+        sample_image_size,
+        sample_batches,
+        sample_batch_size,
+        seed
+    ):
+        self.datamodule = datamodule
+        self.accelerator = accelerator
+        self.vae = vae
+        self.unet = unet
+        self.unet_lora = unet_lora
+        self.tokenizer = tokenizer
+        self.text_encoder = text_encoder
+        self.scheduler = scheduler
+        self.output_dir = output_dir
+        self.instance_identifier = instance_identifier
+        self.placeholder_token = placeholder_token
+        self.placeholder_token_id = placeholder_token_id
+        self.sample_image_size = sample_image_size
+        self.seed = seed or torch.random.seed()
+        self.sample_batches = sample_batches
+        self.sample_batch_size = sample_batch_size
+
+    @torch.no_grad()
+    def save_model(self):
+        print("Saving model...")
+
+        unet_lora = self.accelerator.unwrap_model(self.unet_lora)
+        unet_lora.save_pretrained(self.output_dir.joinpath("model"))
+
+        del unet_lora
+
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+    @torch.no_grad()
+    def save_samples(self, step, num_inference_steps, guidance_scale=7.5, eta=0.0):
+        samples_path = Path(self.output_dir).joinpath("samples")
+
+        unet = self.accelerator.unwrap_model(self.unet)
+        text_encoder = self.accelerator.unwrap_model(self.text_encoder)
+
+        pipeline = VlpnStableDiffusion(
+            text_encoder=text_encoder,
+            vae=self.vae,
+            unet=unet,
+            tokenizer=self.tokenizer,
+            scheduler=self.scheduler,
+        ).to(self.accelerator.device)
+        pipeline.set_progress_bar_config(dynamic_ncols=True)
+
+        train_data = self.datamodule.train_dataloader()
+        val_data = self.datamodule.val_dataloader()
+
+        generator = torch.Generator(device=pipeline.device).manual_seed(self.seed)
+        stable_latents = torch.randn(
+            (self.sample_batch_size, pipeline.unet.in_channels, self.sample_image_size // 8, self.sample_image_size // 8),
+            device=pipeline.device,
+            generator=generator,
+        )
+
+        with torch.autocast("cuda"), torch.inference_mode():
+            for pool, data, latents in [("stable", val_data, stable_latents), ("val", val_data, None), ("train", train_data, None)]:
+                all_samples = []
+                file_path = samples_path.joinpath(pool, f"step_{step}.jpg")
+                file_path.parent.mkdir(parents=True, exist_ok=True)
+
+                data_enum = enumerate(data)
+
+                batches = [
+                    batch
+                    for j, batch in data_enum
+                    if j * data.batch_size < self.sample_batch_size * self.sample_batches
+                ]
+                prompts = [
+                    prompt.format(identifier=self.instance_identifier)
+                    for batch in batches
+                    for prompt in batch["prompts"]
+                ]
+                nprompts = [
+                    prompt
+                    for batch in batches
+                    for prompt in batch["nprompts"]
+                ]
+
+                for i in range(self.sample_batches):
+                    prompt = prompts[i * self.sample_batch_size:(i + 1) * self.sample_batch_size]
+                    nprompt = nprompts[i * self.sample_batch_size:(i + 1) * self.sample_batch_size]
+
+                    samples = pipeline(
+                        prompt=prompt,
+                        negative_prompt=nprompt,
+                        height=self.sample_image_size,
+                        width=self.sample_image_size,
+                        image=latents[:len(prompt)] if latents is not None else None,
+                        generator=generator if latents is not None else None,
+                        guidance_scale=guidance_scale,
+                        eta=eta,
+                        num_inference_steps=num_inference_steps,
+                        output_type='pil'
+                    ).images
+
+                    all_samples += samples
+
+                    del samples
+
+                image_grid = make_grid(all_samples, self.sample_batches, self.sample_batch_size)
+                image_grid.save(file_path, quality=85)
+
+                del all_samples
+                del image_grid
+
+        del unet
+        del text_encoder
+        del pipeline
+        del generator
+        del stable_latents
+
+        if torch.cuda.is_available():
+            torch.cuda.empty_cache()
+
+
+def main():
+    args = parse_args()
+
+    instance_identifier = args.instance_identifier
+
+    if len(args.placeholder_token) != 0:
+        instance_identifier = instance_identifier.format(args.placeholder_token[0])
+
+    now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
+    basepath = Path(args.output_dir).joinpath(slugify(instance_identifier), now)
+    basepath.mkdir(parents=True, exist_ok=True)
+
+    accelerator = Accelerator(
+        log_with=LoggerType.TENSORBOARD,
+        logging_dir=f"{basepath}",
+        gradient_accumulation_steps=args.gradient_accumulation_steps,
+        mixed_precision=args.mixed_precision
+    )
+
+    logging.basicConfig(filename=basepath.joinpath("log.txt"), level=logging.DEBUG)
+
+    args.seed = args.seed or (torch.random.seed() >> 32)
+    set_seed(args.seed)
+
+    save_args(basepath, args)
+
+    # Load the tokenizer and add the placeholder token as a additional special token
+    if args.tokenizer_name:
+        tokenizer = CLIPTokenizer.from_pretrained(args.tokenizer_name)
+    elif args.pretrained_model_name_or_path:
+        tokenizer = CLIPTokenizer.from_pretrained(args.pretrained_model_name_or_path, subfolder='tokenizer')
+
+    # Load models and create wrapper for stable diffusion
+    text_encoder = CLIPTextModel.from_pretrained(args.pretrained_model_name_or_path, subfolder='text_encoder')
+    vae = AutoencoderKL.from_pretrained(args.pretrained_model_name_or_path, subfolder='vae')
+    unet = UNet2DConditionModel.from_pretrained(args.pretrained_model_name_or_path, subfolder='unet')
+    noise_scheduler = DDPMScheduler.from_pretrained(args.pretrained_model_name_or_path, subfolder='scheduler')
+    checkpoint_scheduler = DPMSolverMultistepScheduler.from_pretrained(
+        args.pretrained_model_name_or_path, subfolder='scheduler')
+
+    unet_lora = LoraAttention()
+
+    vae.enable_slicing()
+    vae.set_use_memory_efficient_attention_xformers(True)
+    unet.set_use_memory_efficient_attention_xformers(True)
+    unet.set_attn_processor(unet_lora)
+
+    if args.gradient_checkpointing:
+        unet.enable_gradient_checkpointing()
+        text_encoder.gradient_checkpointing_enable()
+
+    # Freeze text_encoder and vae
+    vae.requires_grad_(False)
+    unet.requires_grad_(False)
+
+    if args.embeddings_dir is not None:
+        embeddings_dir = Path(args.embeddings_dir)
+        if not embeddings_dir.exists() or not embeddings_dir.is_dir():
+            raise ValueError("--embeddings_dir must point to an existing directory")
+        added_tokens = load_text_embeddings(tokenizer, text_encoder, embeddings_dir)
+        print(f"Added {len(added_tokens)} tokens from embeddings dir: {added_tokens}")
+
+    if len(args.placeholder_token) != 0:
+        # Convert the initializer_token, placeholder_token to ids
+        initializer_token_ids = torch.stack([
+            torch.tensor(tokenizer.encode(token, add_special_tokens=False)[:1])
+            for token in args.initializer_token
+        ])
+
+        num_added_tokens = tokenizer.add_tokens(args.placeholder_token)
+        print(f"Added {num_added_tokens} new tokens.")
+
+        placeholder_token_id = tokenizer.convert_tokens_to_ids(args.placeholder_token)
+
+        # Resize the token embeddings as we are adding new special tokens to the tokenizer
+        text_encoder.resize_token_embeddings(len(tokenizer))
+
+        token_embeds = text_encoder.get_input_embeddings().weight.data
+        original_token_embeds = token_embeds.clone().to(accelerator.device)
+        initializer_token_embeddings = text_encoder.get_input_embeddings()(initializer_token_ids)
+
+        for (token_id, embeddings) in zip(placeholder_token_id, initializer_token_embeddings):
+            token_embeds[token_id] = embeddings
+    else:
+        placeholder_token_id = []
+
+    print(f"Training added text embeddings")
+
+    freeze_params(itertools.chain(
+        text_encoder.text_model.encoder.parameters(),
+        text_encoder.text_model.final_layer_norm.parameters(),
+        text_encoder.text_model.embeddings.position_embedding.parameters(),
+    ))
+
+    index_fixed_tokens = torch.arange(len(tokenizer))
+    index_fixed_tokens = index_fixed_tokens[~torch.isin(index_fixed_tokens, torch.tensor(placeholder_token_id))]
+
+    prompt_processor = PromptProcessor(tokenizer, text_encoder)
+
+    if args.scale_lr:
+        args.learning_rate_unet = (
+            args.learning_rate_unet * args.gradient_accumulation_steps *
+            args.train_batch_size * accelerator.num_processes
+        )
+
+    # Use 8-bit Adam for lower memory usage or to fine-tune the model in 16GB GPUs
+    if args.use_8bit_adam:
+        try:
+            import bitsandbytes as bnb
+        except ImportError:
+            raise ImportError("To use 8-bit Adam, please install the bitsandbytes library: `pip install bitsandbytes`.")
+
+        optimizer_class = bnb.optim.AdamW8bit
+    else:
+        optimizer_class = torch.optim.AdamW
+
+    # Initialize the optimizer
+    optimizer = optimizer_class(
+        [
+            {
+                'params': unet_lora.parameters(),
+                'lr': args.learning_rate_unet,
+            },
+        ],
+        betas=(args.adam_beta1, args.adam_beta2),
+        weight_decay=args.adam_weight_decay,
+        eps=args.adam_epsilon,
+    )
+
+    weight_dtype = torch.float32
+    if args.mixed_precision == "fp16":
+        weight_dtype = torch.float16
+    elif args.mixed_precision == "bf16":
+        weight_dtype = torch.bfloat16
+
+    def collate_fn(examples):
+        prompts = [example["prompts"] for example in examples]
+        nprompts = [example["nprompts"] for example in examples]
+        input_ids = [example["instance_prompt_ids"] for example in examples]
+        pixel_values = [example["instance_images"] for example in examples]
+
+        # concat class and instance examples for prior preservation
+        if args.num_class_images != 0 and "class_prompt_ids" in examples[0]:
+            input_ids += [example["class_prompt_ids"] for example in examples]
+            pixel_values += [example["class_images"] for example in examples]
+
+        pixel_values = torch.stack(pixel_values)
+        pixel_values = pixel_values.to(dtype=weight_dtype, memory_format=torch.contiguous_format)
+
+        inputs = prompt_processor.unify_input_ids(input_ids)
+
+        batch = {
+            "prompts": prompts,
+            "nprompts": nprompts,
+            "input_ids": inputs.input_ids,
+            "pixel_values": pixel_values,
+            "attention_mask": inputs.attention_mask,
+        }
+        return batch
+
+    datamodule = CSVDataModule(
+        data_file=args.train_data_file,
+        batch_size=args.train_batch_size,
+        prompt_processor=prompt_processor,
+        instance_identifier=instance_identifier,
+        class_identifier=args.class_identifier,
+        class_subdir="cls",
+        num_class_images=args.num_class_images,
+        size=args.resolution,
+        repeats=args.repeats,
+        mode=args.mode,
+        dropout=args.tag_dropout,
+        center_crop=args.center_crop,
+        template_key=args.train_data_template,
+        valid_set_size=args.valid_set_size,
+        num_workers=args.dataloader_num_workers,
+        collate_fn=collate_fn
+    )
+
+    datamodule.prepare_data()
+    datamodule.setup()
+
+    if args.num_class_images != 0:
+        missing_data = [item for item in datamodule.data_train if not item.class_image_path.exists()]
+
+        if len(missing_data) != 0:
+            batched_data = [
+                missing_data[i:i+args.sample_batch_size]
+                for i in range(0, len(missing_data), args.sample_batch_size)
+            ]
+
+            pipeline = VlpnStableDiffusion(
+                text_encoder=text_encoder,
+                vae=vae,
+                unet=unet,
+                tokenizer=tokenizer,
+                scheduler=checkpoint_scheduler,
+            ).to(accelerator.device)
+            pipeline.set_progress_bar_config(dynamic_ncols=True)
+
+            with torch.autocast("cuda"), torch.inference_mode():
+                for batch in batched_data:
+                    image_name = [item.class_image_path for item in batch]
+                    prompt = [item.prompt.format(identifier=args.class_identifier) for item in batch]
+                    nprompt = [item.nprompt for item in batch]
+
+                    images = pipeline(
+                        prompt=prompt,
+                        negative_prompt=nprompt,
+                        num_inference_steps=args.sample_steps
+                    ).images
+
+                    for i, image in enumerate(images):
+                        image.save(image_name[i])
+
+            del pipeline
+
+            if torch.cuda.is_available():
+                torch.cuda.empty_cache()
+
+    train_dataloader = datamodule.train_dataloader()
+    val_dataloader = datamodule.val_dataloader()
+
+    # 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)
+    if args.max_train_steps is None:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+        overrode_max_train_steps = True
+    num_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+
+    warmup_steps = args.lr_warmup_epochs * num_update_steps_per_epoch * args.gradient_accumulation_steps
+
+    if args.lr_scheduler == "one_cycle":
+        lr_scheduler = get_one_cycle_schedule(
+            optimizer=optimizer,
+            num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        )
+    elif args.lr_scheduler == "cosine_with_restarts":
+        lr_scheduler = get_cosine_with_hard_restarts_schedule_with_warmup(
+            optimizer=optimizer,
+            num_warmup_steps=warmup_steps,
+            num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+            num_cycles=args.lr_cycles or math.ceil(math.sqrt(
+                ((args.max_train_steps - warmup_steps) / num_update_steps_per_epoch))),
+        )
+    else:
+        lr_scheduler = get_scheduler(
+            args.lr_scheduler,
+            optimizer=optimizer,
+            num_warmup_steps=warmup_steps,
+            num_training_steps=args.max_train_steps * args.gradient_accumulation_steps,
+        )
+
+    unet_lora, optimizer, train_dataloader, val_dataloader, lr_scheduler = accelerator.prepare(
+        unet_lora, optimizer, train_dataloader, val_dataloader, lr_scheduler
+    )
+
+    # Move text_encoder and vae to device
+    vae.to(accelerator.device, dtype=weight_dtype)
+    unet.to(accelerator.device, dtype=weight_dtype)
+    text_encoder.to(accelerator.device, dtype=weight_dtype)
+
+    # Keep text_encoder and vae in eval mode as we don't train these
+    vae.eval()
+    unet.eval()
+    text_encoder.eval()
+
+    # We need to recalculate our total training steps as the size of the training dataloader may have changed.
+    num_update_steps_per_epoch = math.ceil(len(train_dataloader) / args.gradient_accumulation_steps)
+    if overrode_max_train_steps:
+        args.max_train_steps = args.num_train_epochs * num_update_steps_per_epoch
+
+    num_val_steps_per_epoch = len(val_dataloader)
+    num_epochs = math.ceil(args.max_train_steps / num_update_steps_per_epoch)
+    val_steps = num_val_steps_per_epoch * num_epochs
+
+    # We need to initialize the trackers we use, and also store our configuration.
+    # The trackers initializes automatically on the main process.
+    if accelerator.is_main_process:
+        config = vars(args).copy()
+        config["initializer_token"] = " ".join(config["initializer_token"])
+        config["placeholder_token"] = " ".join(config["placeholder_token"])
+        accelerator.init_trackers("dreambooth", config=config)
+
+    # Train!
+    total_batch_size = args.train_batch_size * accelerator.num_processes * args.gradient_accumulation_steps
+
+    logger.info("***** Running training *****")
+    logger.info(f"  Num Epochs = {num_epochs}")
+    logger.info(f"  Instantaneous batch size per device = {args.train_batch_size}")
+    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
+    logger.info(f"  Gradient Accumulation steps = {args.gradient_accumulation_steps}")
+    logger.info(f"  Total optimization steps = {args.max_train_steps}")
+    # Only show the progress bar once on each machine.
+
+    global_step = 0
+
+    avg_loss = AverageMeter()
+    avg_acc = AverageMeter()
+
+    avg_loss_val = AverageMeter()
+    avg_acc_val = AverageMeter()
+
+    max_acc_val = 0.0
+
+    checkpointer = Checkpointer(
+        datamodule=datamodule,
+        accelerator=accelerator,
+        vae=vae,
+        unet=unet,
+        tokenizer=tokenizer,
+        text_encoder=text_encoder,
+        scheduler=checkpoint_scheduler,
+        output_dir=basepath,
+        instance_identifier=instance_identifier,
+        placeholder_token=args.placeholder_token,
+        placeholder_token_id=placeholder_token_id,
+        sample_image_size=args.sample_image_size,
+        sample_batch_size=args.sample_batch_size,
+        sample_batches=args.sample_batches,
+        seed=args.seed
+    )
+
+    if accelerator.is_main_process:
+        checkpointer.save_samples(0, args.sample_steps)
+
+    local_progress_bar = tqdm(
+        range(num_update_steps_per_epoch + num_val_steps_per_epoch),
+        disable=not accelerator.is_local_main_process,
+        dynamic_ncols=True
+    )
+    local_progress_bar.set_description("Epoch X / Y")
+
+    global_progress_bar = tqdm(
+        range(args.max_train_steps + val_steps),
+        disable=not accelerator.is_local_main_process,
+        dynamic_ncols=True
+    )
+    global_progress_bar.set_description("Total progress")
+
+    try:
+        for epoch in range(num_epochs):
+            local_progress_bar.set_description(f"Epoch {epoch + 1} / {num_epochs}")
+            local_progress_bar.reset()
+
+            unet_lora.train()
+
+            for step, batch in enumerate(train_dataloader):
+                with accelerator.accumulate(unet_lora):
+                    # Convert images to latent space
+                    latents = vae.encode(batch["pixel_values"]).latent_dist.sample()
+                    latents = latents * 0.18215
+
+                    # Sample noise that we'll add to the latents
+                    noise = torch.randn_like(latents)
+                    bsz = latents.shape[0]
+                    # Sample a random timestep for each image
+                    timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps,
+                                              (bsz,), 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)
+
+                    # Get the text embedding for conditioning
+                    encoder_hidden_states = prompt_processor.get_embeddings(batch["input_ids"], batch["attention_mask"])
+
+                    # Predict the noise residual
+                    model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                    # Get the target for loss depending on the prediction type
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        target = noise
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                    else:
+                        raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                    if args.num_class_images != 0:
+                        # Chunk the noise and model_pred into two parts and compute the loss on each part separately.
+                        model_pred, model_pred_prior = torch.chunk(model_pred, 2, dim=0)
+                        target, target_prior = torch.chunk(target, 2, dim=0)
+
+                        # Compute instance loss
+                        loss = F.mse_loss(model_pred.float(), target.float(), reduction="none").mean([1, 2, 3]).mean()
+
+                        # Compute prior loss
+                        prior_loss = F.mse_loss(model_pred_prior.float(), target_prior.float(), reduction="mean")
+
+                        # Add the prior loss to the instance loss.
+                        loss = loss + args.prior_loss_weight * prior_loss
+                    else:
+                        loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                    acc = (model_pred == latents).float().mean()
+
+                    accelerator.backward(loss)
+
+                    if accelerator.sync_gradients:
+                        accelerator.clip_grad_norm_(unet_lora.parameters(), args.max_grad_norm)
+
+                    optimizer.step()
+                    if not accelerator.optimizer_step_was_skipped:
+                        lr_scheduler.step()
+                    optimizer.zero_grad(set_to_none=True)
+
+                    with torch.no_grad():
+                        text_encoder.get_input_embeddings(
+                        ).weight[index_fixed_tokens] = original_token_embeds[index_fixed_tokens]
+
+                    avg_loss.update(loss.detach_(), bsz)
+                    avg_acc.update(acc.detach_(), bsz)
+
+                # Checks if the accelerator has performed an optimization step behind the scenes
+                if accelerator.sync_gradients:
+                    local_progress_bar.update(1)
+                    global_progress_bar.update(1)
+
+                    global_step += 1
+
+                logs = {
+                    "train/loss": avg_loss.avg.item(),
+                    "train/acc": avg_acc.avg.item(),
+                    "train/cur_loss": loss.item(),
+                    "train/cur_acc": acc.item(),
+                    "lr/unet": lr_scheduler.get_last_lr()[0],
+                    "lr/text": lr_scheduler.get_last_lr()[1]
+                }
+
+                accelerator.log(logs, step=global_step)
+
+                local_progress_bar.set_postfix(**logs)
+
+                if global_step >= args.max_train_steps:
+                    break
+
+            accelerator.wait_for_everyone()
+
+            unet_lora.eval()
+
+            with torch.inference_mode():
+                for step, batch in enumerate(val_dataloader):
+                    latents = vae.encode(batch["pixel_values"]).latent_dist.sample()
+                    latents = latents * 0.18215
+
+                    noise = torch.randn_like(latents)
+                    bsz = latents.shape[0]
+                    timesteps = torch.randint(0, noise_scheduler.config.num_train_timesteps,
+                                              (bsz,), device=latents.device)
+                    timesteps = timesteps.long()
+
+                    noisy_latents = noise_scheduler.add_noise(latents, noise, timesteps)
+
+                    encoder_hidden_states = prompt_processor.get_embeddings(batch["input_ids"], batch["attention_mask"])
+
+                    model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
+
+                    # Get the target for loss depending on the prediction type
+                    if noise_scheduler.config.prediction_type == "epsilon":
+                        target = noise
+                    elif noise_scheduler.config.prediction_type == "v_prediction":
+                        target = noise_scheduler.get_velocity(latents, noise, timesteps)
+                    else:
+                        raise ValueError(f"Unknown prediction type {noise_scheduler.config.prediction_type}")
+
+                    loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")
+
+                    acc = (model_pred == latents).float().mean()
+
+                    avg_loss_val.update(loss.detach_(), bsz)
+                    avg_acc_val.update(acc.detach_(), bsz)
+
+                    if accelerator.sync_gradients:
+                        local_progress_bar.update(1)
+                        global_progress_bar.update(1)
+
+                    logs = {
+                        "val/loss": avg_loss_val.avg.item(),
+                        "val/acc": avg_acc_val.avg.item(),
+                        "val/cur_loss": loss.item(),
+                        "val/cur_acc": acc.item(),
+                    }
+                    local_progress_bar.set_postfix(**logs)
+
+            accelerator.log({
+                "val/loss": avg_loss_val.avg.item(),
+                "val/acc": avg_acc_val.avg.item(),
+            }, step=global_step)
+
+            local_progress_bar.clear()
+            global_progress_bar.clear()
+
+            if avg_acc_val.avg.item() > max_acc_val:
+                accelerator.print(
+                    f"Global step {global_step}: Validation accuracy reached new maximum: {max_acc_val:.2e} -> {avg_acc_val.avg.item():.2e}")
+                max_acc_val = avg_acc_val.avg.item()
+
+            if accelerator.is_main_process:
+                if (epoch + 1) % args.sample_frequency == 0:
+                    checkpointer.save_samples(global_step, args.sample_steps)
+
+        # Create the pipeline using using the trained modules and save it.
+        if accelerator.is_main_process:
+            print("Finished! Saving final checkpoint and resume state.")
+            checkpointer.save_model()
+
+            accelerator.end_training()
+
+    except KeyboardInterrupt:
+        if accelerator.is_main_process:
+            print("Interrupted, saving checkpoint and resume state...")
+            checkpointer.save_model()
+            accelerator.end_training()
+        quit()
+
+
+if __name__ == "__main__":
+    main()
diff --git a/train_ti.py b/train_ti.py
index 5c0299e..9616db6 100644
--- a/train_ti.py
+++ b/train_ti.py
@@ -24,7 +24,6 @@ from slugify import slugify
 
 from common import load_text_embeddings, load_text_embedding
 from pipelines.stable_diffusion.vlpn_stable_diffusion import VlpnStableDiffusion
-from pipelines.util import set_use_memory_efficient_attention_xformers
 from data.csv import CSVDataModule, CSVDataItem
 from training.optimization import get_one_cycle_schedule
 from models.clip.prompt import PromptProcessor
@@ -557,8 +556,8 @@ def main():
         args.pretrained_model_name_or_path, subfolder='scheduler')
 
     vae.enable_slicing()
-    set_use_memory_efficient_attention_xformers(unet, True)
-    set_use_memory_efficient_attention_xformers(vae, True)
+    vae.set_use_memory_efficient_attention_xformers(True)
+    unet.set_use_memory_efficient_attention_xformers(True)
 
     if args.gradient_checkpointing:
         unet.enable_gradient_checkpointing()
diff --git a/training/lora.py b/training/lora.py
new file mode 100644
index 0000000..d8dc147
--- /dev/null
+++ b/training/lora.py
@@ -0,0 +1,27 @@
+import torch.nn as nn
+from diffusers import ModelMixin, ConfigMixin, XFormersCrossAttnProcessor, register_to_config
+
+
+class LoraAttention(ModelMixin, ConfigMixin):
+    @register_to_config
+    def __init__(self, in_features, out_features, r=4):
+        super().__init__()
+
+        if r > min(in_features, out_features):
+            raise ValueError(
+                f"LoRA rank {r} must be less or equal than {min(in_features, out_features)}"
+            )
+
+        self.lora_down = nn.Linear(in_features, r, bias=False)
+        self.lora_up = nn.Linear(r, out_features, bias=False)
+        self.scale = 1.0
+
+        self.processor = XFormersCrossAttnProcessor()
+
+        nn.init.normal_(self.lora_down.weight, std=1 / r**2)
+        nn.init.zeros_(self.lora_up.weight)
+
+    def __call__(self, attn, hidden_states, encoder_hidden_states=None, attention_mask=None, number=None):
+        hidden_states = self.processor(attn, hidden_states, encoder_hidden_states, attention_mask, number)
+        hidden_states = hidden_states + self.lora_up(self.lora_down(hidden_states)) * self.scale
+        return hidden_states