train_with_lora.py

import os
import sys
import json

import argparse
import itertools
import datetime
import logging
import inspect
import math
from typing import Dict, Optional, Tuple
from omegaconf import OmegaConf
from collections import OrderedDict
from PIL import Image

import torch
import torch.nn.functional as F
import torch.utils.checkpoint

import diffusers
import transformers
from accelerate import Accelerator
from accelerate.logging import get_logger
from accelerate.utils import set_seed
from diffusers import AutoencoderKL, DDPMScheduler, DDIMScheduler
from diffusers.optimization import get_scheduler
from diffusers.utils import check_min_version
from diffusers.utils.import_utils import is_xformers_available
from diffusers.loaders import AttnProcsLayers
from tqdm.auto import tqdm
from transformers import CLIPTextModel, CLIPTokenizer
from diffusers.models.attention_processor import LoRAAttnProcessor
from safetensors import safe_open

from diffusers.loaders import LoraLoaderMixin
from diffusers.models.attention_processor import (
    AttnAddedKVProcessor,
    AttnAddedKVProcessor2_0,
    SlicedAttnAddedKVProcessor,
)
from diffusers.models.lora import LoRALinearLayer
from diffusers.training_utils import unet_lora_state_dict

import wandb

from animatediff.models.unet import UNet3DConditionModel
from animatediff.pipelines.pipeline_animation import AnimationPipeline
from animatediff.utils.convert_from_ckpt import convert_ldm_unet_checkpoint, convert_ldm_clip_checkpoint, convert_ldm_vae_checkpoint
from animatediff.utils.util import save_videos_grid, ddim_inversion
from einops import rearrange, repeat

from animatediff.utils.dataset import AnimateDiffDataset

from inspect import currentframe

def text_encoder_lora_state_dict(text_encoder):
    state_dict = {}

    for name, module in text_encoder_attn_modules(text_encoder):
        for k, v in module.q_proj.lora_linear_layer.state_dict().items():
            state_dict[f"{name}.q_proj.lora_linear_layer.{k}"] = v

        for k, v in module.k_proj.lora_linear_layer.state_dict().items():
            state_dict[f"{name}.k_proj.lora_linear_layer.{k}"] = v

        for k, v in module.v_proj.lora_linear_layer.state_dict().items():
            state_dict[f"{name}.v_proj.lora_linear_layer.{k}"] = v

        for k, v in module.out_proj.lora_linear_layer.state_dict().items():
            state_dict[f"{name}.out_proj.lora_linear_layer.{k}"] = v

    return state_dict

def get_linenumber():
    cf = currentframe()
    return cf.f_back.f_lineno

def log_memory(line):
    """
    memory = torch.cuda.mem_get_info()
    allocated = memory[1] - memory[0]
    scale = 1000*1000*1000
    logger.info("memory line:%d %.2fG" % (line, allocated/scale))
    """

# Will error if the minimal version of diffusers is not installed. Remove at your own risks.
check_min_version("0.10.0.dev0")

logger = get_logger(__name__, log_level="INFO")


def load_checkpoint(path, unet, vae, text_encoder):
    if path.endswith(".ckpt"):
        state_dict = torch.load(path)
        unet.load_state_dict(state_dict)

    elif path.endswith(".safetensors"):
        state_dict = {}
        with safe_open(path, framework="pt", device="cpu") as f:
            for key in f.keys():
                state_dict[key] = f.get_tensor(key)

        base_state_dict = state_dict

        # vae
        converted_vae_checkpoint = convert_ldm_vae_checkpoint(base_state_dict, vae.config)
        vae.load_state_dict(converted_vae_checkpoint)
        # unet
        converted_unet_checkpoint = convert_ldm_unet_checkpoint(base_state_dict, unet.config)
        unet.load_state_dict(converted_unet_checkpoint, strict=False)
        # text_model
        text_encoder = convert_ldm_clip_checkpoint(base_state_dict)

        return unet, vae, text_encoder


def main(
    pretrained_model_path: str,
    output_dir: str,
    train_data: Dict,
    validation_data: Dict,
    validation_steps: int = 100,
    dreambooth_path: str = None,
    train_whole_module: bool = False,
    trainable_modules: Tuple[str] = (
        "to_q",
    ),
    train_batch_size: int = 1,
    max_train_steps: int = 500,
    learning_rate: float = 3e-5,
    scale_lr: bool = False,
    lr_scheduler: str = "constant",
    lr_warmup_steps: int = 0,
    adam_beta1: float = 0.9,
    adam_beta2: float = 0.999,
    adam_weight_decay: float = 1e-2,
    adam_epsilon: float = 1e-08,
    max_grad_norm: float = 1.0,
    gradient_accumulation_steps: int = 1,
    gradient_checkpointing: bool = True,
    checkpointing_steps: int = 500,
    resume_from_checkpoint: Optional[str] = None,
    mixed_precision: Optional[str] = "fp16",
    use_optimizer: str = 'AdamW',
    enable_xformers_memory_efficient_attention: bool = True,
    seed: Optional[int] = None,

    motion_module: str = "models/Motion_Module/mm_sd_v15.ckpt",
    inference_config_path: str = "configs/inference/inference.yaml",
    motion_module_pe_multiplier: int = 1,

    train_dreambooth: bool = False,
    train_lora: bool = False,
    train_text_encoder: bool = False,

    lora_rank: int = 4,
    lora_path: str = None,

    report_to: str = None,
):
    *_, config = inspect.getargvalues(inspect.currentframe())

    inference_config = OmegaConf.load(inference_config_path)

    accelerator = Accelerator(
        gradient_accumulation_steps=gradient_accumulation_steps,
        mixed_precision=mixed_precision,
        log_with=report_to,
    )

    # Make one log on every process with the configuration for debugging.
    logging.basicConfig(
        format="%(asctime)s - %(levelname)s - %(name)s - %(message)s",
        datefmt="%m/%d/%Y %H:%M:%S",
        level=logging.INFO,
    )
    logger.info(accelerator.state, main_process_only=False)
    log_memory(get_linenumber())

    if accelerator.is_local_main_process:
        transformers.utils.logging.set_verbosity_warning()
        diffusers.utils.logging.set_verbosity_info()
    else:
        transformers.utils.logging.set_verbosity_error()
        diffusers.utils.logging.set_verbosity_error()

    log_memory(get_linenumber())
    # If passed along, set the training seed now.
    if seed is not None:
        set_seed(seed)

    # Handle the output folder creation
    if accelerator.is_main_process:
        # now = datetime.datetime.now().strftime("%Y-%m-%dT%H-%M-%S")
        # output_dir = os.path.join(output_dir, now)
        os.makedirs(output_dir, exist_ok=True)
        os.makedirs(f"{output_dir}/samples", exist_ok=True)
        os.makedirs(f"{output_dir}/inv_latents", exist_ok=True)
        OmegaConf.save(config, os.path.join(output_dir, 'config.yaml'))

    log_memory(get_linenumber())

    # Load scheduler, tokenizer and models.
    noise_scheduler = DDPMScheduler.from_pretrained(pretrained_model_path, subfolder="scheduler")
    tokenizer = CLIPTokenizer.from_pretrained(pretrained_model_path, subfolder="tokenizer")
    text_encoder = CLIPTextModel.from_pretrained(pretrained_model_path, subfolder="text_encoder")
    vae = AutoencoderKL.from_pretrained(pretrained_model_path, subfolder="vae")
    unet = UNet3DConditionModel.from_pretrained_2d(
            pretrained_model_path,
            subfolder="unet",
            unet_additional_kwargs=OmegaConf.to_container(inference_config.unet_additional_kwargs))

    log_memory(get_linenumber())

    if dreambooth_path != None and dreambooth_path != "":
        unet, vae, text_encoder = load_checkpoint(dreambooth_path, unet, vae, text_encoder)

    log_memory(get_linenumber())

    if motion_module != None and motion_module != "":
        motion_module_state_dict = torch.load(motion_module, map_location="cpu")

        # Multiply pe weights by multiplier for training more than 24 frames
        if motion_module_pe_multiplier > 1:
            for key in motion_module_state_dict:
              if 'pe' in key:
                t = motion_module_state_dict[key]
                t = repeat(t, "b f d -> b (f m) d", m=motion_module_pe_multiplier)
                motion_module_state_dict[key] = t

        if "global_step" in motion_module_state_dict: func_args.update({"global_step": motion_module_state_dict["global_step"]})
        missing, unexpected = unet.load_state_dict(motion_module_state_dict, strict=False)
        assert len(unexpected) == 0

    log_memory(get_linenumber())

    text_lora_parameters = []
    unet_lora_parameters = []

    if train_lora:
        if lora_path is not None:
            logging.info(f"Loading LoRA {lora_path}")
            lora_state_dict, network_alphas = LoraLoaderMixin.lora_state_dict(lora_path, use_safetensors=True)
            LoraLoaderMixin.load_lora_into_unet(lora_state_dict, network_alphas=network_alphas, unet=unet)
            LoraLoaderMixin.load_lora_into_text_encoder(
                lora_state_dict, network_alphas=network_alphas, text_encoder=text_encoder
            )

        def find_lora_compatible_layers(module, path, found):
            for n, m in module.named_children():
                p = path + [n]
                if find_lora_compatible_layers(m, p, found):
                    found.append(p)

            if 'LoRACompatibleLinear' in str(type(module)) and hasattr(module, 'set_lora_layer'):
                for m in ['to_q', 'to_k', 'to_v', 'to_out']:
                    if m in ".".join(path):
                        return True

            return False

        def update_lora_parameters(unet):
            lora_layers = []
            find_lora_compatible_layers(unet, [], lora_layers)

            for path in lora_layers:
                path = ".".join(path)

                lora_module = unet
                for n in path.split("."):
                    lora_module = getattr(lora_module, n)

                lora_module.set_lora_layer(
                    LoRALinearLayer(
                        in_features=lora_module.in_features,
                        out_features=lora_module.out_features,
                        rank=lora_rank,
                        device=unet.device,
                        dtype=unet.dtype,
                    )
                )

                unet_lora_parameters.extend(lora_module.lora_layer.parameters())

            return unet_lora_parameters

        unet_lora_parameters = update_lora_parameters(unet)

        # The text encoder comes from 🤗 transformers, so we cannot directly modify it.
        # So, instead, we monkey-patch the forward calls of its attention-blocks.
        if train_text_encoder:
            # ensure that dtype is float32, even if rest of the model that isn't trained is loaded in fp16
            text_lora_parameters = LoraLoaderMixin._modify_text_encoder(text_encoder, dtype=torch.float32, rank=lora_rank)

        # create custom saving & loading hooks so that `accelerator.save_state(...)` serializes in a nice format
        def save_model_hook(models, weights, output_dir):
            if accelerator.is_main_process and len(weights) > 0:
                # there are only two options here. Either are just the unet attn processor layers
                # or there are the unet and text encoder atten layers
                unet_lora_layers_to_save = None
                text_encoder_lora_layers_to_save = None

                for model in models:
                    if isinstance(model, type(accelerator.unwrap_model(unet))):
                        unet_lora_layers_to_save = unet_lora_state_dict(model)
                    elif isinstance(model, type(accelerator.unwrap_model(text_encoder))):
                        text_encoder_lora_layers_to_save = text_encoder_lora_state_dict(model)
                    else:
                        raise ValueError(f"unexpected save model: {model.__class__}")

                    # make sure to pop weight so that corresponding model is not saved again
                    weights.pop()

                LoraLoaderMixin.save_lora_weights(
                    output_dir,
                    unet_lora_layers=unet_lora_layers_to_save,
                    text_encoder_lora_layers=text_encoder_lora_layers_to_save,
                    )

        def load_model_hook(models, input_dir):
            unet_ = None
            text_encoder_ = None

            while len(models) > 0:
                model = models.pop()

                if isinstance(model, type(accelerator.unwrap_model(unet))):
                    unet_ = model
                elif isinstance(model, type(accelerator.unwrap_model(text_encoder))):
                    text_encoder_ = model
                else:
                    raise ValueError(f"unexpected save model: {model.__class__}")

            if unet_ is not None:
                lora_state_dict, network_alphas = LoraLoaderMixin.lora_state_dict(input_dir)
                LoraLoaderMixin.load_lora_into_unet(lora_state_dict, network_alphas=network_alphas, unet=unet_)
                LoraLoaderMixin.load_lora_into_text_encoder(
                    lora_state_dict, network_alphas=network_alphas, text_encoder=text_encoder_
                )
                unet_lora_parameters = update_lora_parameters(unet_)

        accelerator.register_save_state_pre_hook(save_model_hook)
        accelerator.register_load_state_pre_hook(load_model_hook)

    log_memory(get_linenumber())

    # Freeze vae and text_encoder
    vae.requires_grad_(False)
    text_encoder.requires_grad_(False)
    unet.requires_grad_(False)

    log_memory(get_linenumber())

    mm_parameters = []
    unet_parameters = []
    if train_dreambooth:
        unet_parameters.extend(unet.parameters())
    elif train_whole_module or len(trainable_modules) > 0:
        for name, module in unet.named_modules():
            if "motion_modules" in name and (train_whole_module or name.endswith(tuple(trainable_modules))):
                mm_parameters.extend(module.parameters())

    log_memory(get_linenumber())

    if enable_xformers_memory_efficient_attention:
        if is_xformers_available():
            unet.enable_xformers_memory_efficient_attention()
        else:
            raise ValueError("xformers is not available. Make sure it is installed correctly")

    log_memory(get_linenumber())

    if gradient_checkpointing:
        unet.enable_gradient_checkpointing()
        if train_text_encoder:
            text_encoder.gradient_checkpointing_enable()

    log_memory(get_linenumber())

    if scale_lr:
        learning_rate = (
            learning_rate * gradient_accumulation_steps * train_batch_size * accelerator.num_processes
        )

    logger.info("Optimize:")
    logger.info("* unet_params: %d" % len(unet_parameters))
    logger.info("* unet_lora_params: %d" % len(unet_lora_parameters))
    logger.info("* text_lora_params: %d" % len(text_lora_parameters))
    logger.info("* mm_params: %d" % len(mm_parameters))

    all_parameters = unet_parameters + unet_lora_parameters + text_lora_parameters + mm_parameters
    for param in all_parameters:
        param.requires_grad_(True)

    log_memory(get_linenumber())

    # Optimizer creation
    params_to_optimize = itertools.chain(unet_parameters, unet_lora_parameters, text_lora_parameters, mm_parameters)

    # Initialize the optimizer
    optimizer = None
    if use_optimizer == 'AdamW8bit':
        try:
            import bitsandbytes as bnb
        except ImportError:
            raise ImportError(
                "Please install bitsandbytes to use 8-bit Adam. You can do so by running `pip install bitsandbytes`"
            )

        optimizer_cls = bnb.optim.AdamW8bit
    elif use_optimizer == 'DAdaptation':
        try:
            import dadaptation
        except ImportError:
            raise ImportError(
                "Please install dadaptation. You can do so by running `pip install dadaptation`"
            )

        optimizer = dadaptation.DAdaptAdam(
            params_to_optimize,
            lr=1.,
            weight_decay=adam_weight_decay,
        )
    elif use_optimizer == 'Prodigy':
        try:
            import prodigyopt
        except ImportError:
            raise ImportError(
                "Please install prodigyopt. You can do so by running `pip install prodigyopt`"
            )

        # https://rentry.org/59xed3
        optimizer = prodigyopt.Prodigy(
            params_to_optimize,
            lr=1.,
            decouple=True,
            weight_decay=0.01,
            d_coef=2,
            use_bias_correction=True,
            safeguard_warmup=True,
        )
    else:
        optimizer_cls = torch.optim.AdamW

    if optimizer == None:
        optimizer = optimizer_cls(
            params_to_optimize,
            lr=float(learning_rate),
            betas=(adam_beta1, adam_beta2),
            weight_decay=adam_weight_decay,
            eps=adam_epsilon,
        )

    log_memory(get_linenumber())

    # Get the training dataset
    train_dataset = AnimateDiffDataset(tokenizer=tokenizer, **train_data)

    log_memory(get_linenumber())

    # DataLoaders creation:
    train_dataloader = torch.utils.data.DataLoader(
        train_dataset, batch_size=train_batch_size
    )

    log_memory(get_linenumber())

    # Get the validation pipeline
    validation_pipeline = AnimationPipeline(
        vae=vae, text_encoder=text_encoder, tokenizer=tokenizer, unet=unet,
        scheduler=DDIMScheduler(**OmegaConf.to_container(inference_config.noise_scheduler_kwargs['DDIMScheduler'])),
    )
    validation_pipeline.enable_vae_slicing()
    ddim_inv_scheduler = DDIMScheduler.from_pretrained(pretrained_model_path, subfolder='scheduler')
    ddim_inv_scheduler.set_timesteps(validation_data.num_inv_steps)

    log_memory(get_linenumber())

    # Scheduler
    lr_scheduler = get_scheduler(
        lr_scheduler,
        optimizer=optimizer,
        num_warmup_steps=lr_warmup_steps * gradient_accumulation_steps,
        num_training_steps=max_train_steps * gradient_accumulation_steps,
    )

    log_memory(get_linenumber())

    # Prepare everything with our `accelerator`.
    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
        )

    log_memory(get_linenumber())

    # For mixed precision training we cast the text_encoder and vae weights to half-precision
    # as these models are only used for inference, keeping weights in full precision is not required.
    weight_dtype = torch.float32
    if accelerator.mixed_precision == "fp16":
        weight_dtype = torch.float16
    elif accelerator.mixed_precision == "bf16":
        weight_dtype = torch.bfloat16

    log_memory(get_linenumber())

    # Move text_encode and vae to gpu and cast to weight_dtype
    text_encoder.to(accelerator.device, dtype=weight_dtype)
    vae.to(accelerator.device, dtype=weight_dtype)

    log_memory(get_linenumber())

    # 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) / gradient_accumulation_steps)
    # Afterwards we recalculate our number of training epochs
    num_train_epochs = math.ceil(max_train_steps / num_update_steps_per_epoch)

    # 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:
        accelerator.init_trackers("animatediff")

    log_memory(get_linenumber())

    # Train!
    total_batch_size = train_batch_size * accelerator.num_processes * gradient_accumulation_steps

    logger.info("***** Running training *****")
    logger.info(f"  Num examples = {len(train_dataset)}")
    logger.info(f"  Num Epochs = {num_train_epochs}")
    logger.info(f"  Instantaneous batch size per device = {train_batch_size}")
    logger.info(f"  Total train batch size (w. parallel, distributed & accumulation) = {total_batch_size}")
    logger.info(f"  Gradient Accumulation steps = {gradient_accumulation_steps}")
    logger.info(f"  Total optimization steps = {max_train_steps}")
    global_step = 0
    first_epoch = 0

    # Potentially load in the weights and states from a previous save
    if resume_from_checkpoint:
        if resume_from_checkpoint != "latest":
            path = os.path.basename(resume_from_checkpoint)
        else:
            # Get the most recent checkpoint
            dirs = os.listdir(output_dir)
            dirs = [d for d in dirs if d.startswith("checkpoint")]
            dirs = sorted(dirs, key=lambda x: int(x.split("-")[1]))
            path = dirs[-1]
        accelerator.print(f"Resuming from checkpoint {path}")
        accelerator.load_state(os.path.join(output_dir, path))
        global_step = int(path.split("-")[1])

        first_epoch = global_step // num_update_steps_per_epoch
        resume_step = global_step % num_update_steps_per_epoch

    log_memory(get_linenumber())

    # Only show the progress bar once on each machine.
    progress_bar = tqdm(range(global_step, max_train_steps), disable=not accelerator.is_local_main_process)
    progress_bar.set_description("Steps")

    for epoch in range(first_epoch, num_train_epochs):
        unet.train()
        train_loss = 0.0
        for step, batch in enumerate(train_dataloader):
            log_memory(get_linenumber())

            # Skip steps until we reach the resumed step
            if resume_from_checkpoint and epoch == first_epoch and step < resume_step:
                if step % gradient_accumulation_steps == 0:
                    progress_bar.update(1)
                continue

            with accelerator.accumulate(unet):
                # Convert videos to latent space
                pixel_values = batch["pixel_values"].to(weight_dtype)
                video_length = pixel_values.shape[1]
                pixel_values = rearrange(pixel_values, "b f c h w -> (b f) c h w")
                latents = vae.encode(pixel_values).latent_dist.sample()
                latents = rearrange(latents, "(b f) c h w -> b c f h w", f=video_length)
                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 video
                timesteps = torch.randint(0, noise_scheduler.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 = text_encoder(batch["prompt_ids"])[0]

                # Get the target for loss depending on the prediction type
                if noise_scheduler.prediction_type == "epsilon":
                    target = noise
                elif noise_scheduler.prediction_type == "v_prediction":
                    target = noise_scheduler.get_velocity(latents, noise, timesteps)
                else:
                    raise ValueError(f"Unknown prediction type {noise_scheduler.prediction_type}")

                # Predict the noise residual and compute loss
                model_pred = unet(noisy_latents, timesteps, encoder_hidden_states).sample
                loss = F.mse_loss(model_pred.float(), target.float(), reduction="mean")

                # Gather the losses across all processes for logging (if we use distributed training).
                avg_loss = accelerator.gather(loss.repeat(train_batch_size)).mean()
                train_loss += avg_loss.item() / gradient_accumulation_steps

                # Backpropagate
                accelerator.backward(loss)
                if accelerator.sync_gradients:
                    accelerator.clip_grad_norm_(params_to_optimize, max_grad_norm)
                optimizer.step()
                lr_scheduler.step()
                optimizer.zero_grad()

            # Checks if the accelerator has performed an optimization step behind the scenes
            if accelerator.sync_gradients:
                progress_bar.update(1)
                global_step += 1
                accelerator.log({"train_loss": train_loss}, step=global_step)
                train_loss = 0.0

                if global_step % checkpointing_steps == 0:
                    if accelerator.is_main_process:
                        log_memory(get_linenumber())

                        if train_whole_module or len(trainable_modules) > 0:
                            save_path = os.path.join(output_dir, f"mm-{global_step}.pth")
                            save_mm_checkpoint(unet, save_path)
                            logger.info(f"Saved mm state to {save_path}")

                        if train_lora:
                            save_path = os.path.join(output_dir, f"lora-{global_step}")
                            save_lora_checkpoint(unet, save_path, text_encoder if train_text_encoder else None, accelerator)
                            logger.info(f"Saved lora state to {save_path}")

                        save_path = os.path.join(output_dir, f"checkpoint-{global_step}")
                        accelerator.save_state(save_path)
                        logger.info(f"Saved state to {save_path}")

                        log_memory(get_linenumber())


                if global_step % validation_steps == 0:
                    if accelerator.is_main_process:
                        log_memory(get_linenumber())

                        samples = []
                        generator = torch.Generator(device=latents.device)
                        generator.manual_seed(seed)

                        ddim_inv_latent = None
                        if validation_data.use_inv_latent:
                            inv_latents_path = os.path.join(output_dir, f"inv_latents/ddim_latent-{global_step}.pt")
                            ddim_inv_latent = ddim_inversion(
                                validation_pipeline, ddim_inv_scheduler, video_latent=latents,
                                num_inv_steps=validation_data.num_inv_steps, prompt="")[-1].to(weight_dtype)
                            torch.save(ddim_inv_latent, inv_latents_path)

                        wandb_images = []
                        for idx, prompt in enumerate(set(validation_data.prompts)):
                            sample = validation_pipeline(prompt, generator=generator,
                                                         latents=ddim_inv_latent,
                                                         fp16=True,
                                                         **validation_data).videos
                            outputs = save_videos_grid(sample, f"{output_dir}/samples/sample-{global_step}/{idx}.gif")
                            samples.append(sample)
                            for frame in outputs:
                                wandb_images.append(wandb.Image(Image.fromarray(frame), caption=f"{prompt}:{idx}"))

                        if len(samples) > 0:
                            samples = torch.concat(samples)
                            save_path = f"{output_dir}/samples/sample-{global_step}.gif"
                            save_videos_grid(samples, save_path)
                            logger.info(f"Saved samples to {save_path}")

                        for tracker in accelerator.trackers:
                            if tracker.name == "wandb":
                                tracker.log(
                                    {
                                        "validation": wandb_images,
                                    }
                                )

                        log_memory(get_linenumber())


            logs = {"step_loss": loss.detach().item(), "lr": lr_scheduler.get_last_lr()[0]}
            progress_bar.set_postfix(**logs)

            if global_step >= max_train_steps:
                break

    # Create the pipeline using the trained modules and save it.
    accelerator.wait_for_everyone()
    if accelerator.is_main_process:
        log_memory(get_linenumber())

        unet = accelerator.unwrap_model(unet)
        unet = unet.to(torch.float32)
        pipeline = AnimationPipeline.from_pretrained(
            pretrained_model_path,
            text_encoder=text_encoder,
            vae=vae,
            unet=unet,
        )

        if train_whole_module or len(trainable_modules) > 0:
            mm_path = "%s/mm.pth" % output_dir
            save_mm_checkpoint(unet, mm_path)

        if train_lora:
            unet_lora_layers = unet_lora_state_dict(unet)

            if text_encoder is not None and train_text_encoder:
                text_encoder = accelerator.unwrap_model(text_encoder)
                text_encoder = text_encoder.to(torch.float32)
                text_encoder_lora_layers = text_encoder_lora_state_dict(text_encoder)
            else:
                text_encoder_lora_layers = None

            LoraLoaderMixin.save_lora_weights(
                save_directory=output_dir,
                unet_lora_layers=unet_lora_layers,
                text_encoder_lora_layers=text_encoder_lora_layers,
            )
        log_memory(get_linenumber())

    accelerator.end_training()

def save_mm_checkpoint(unet, mm_path):
    mm_state_dict = OrderedDict()
    state_dict = unet.state_dict()
    for key in state_dict:
        if "motion_module" in key:
            mm_state_dict[key] = state_dict[key]

    torch.save(mm_state_dict, mm_path)

def save_lora_checkpoint(unet, output_dir, text_encoder, accelerator):
    unet = accelerator.unwrap_model(unet)
    unet = unet.to(torch.float32)
    unet_lora_layers = unet_lora_state_dict(unet)

    if text_encoder is not None:
        text_encoder = accelerator.unwrap_model(text_encoder)
        text_encoder = text_encoder.to(torch.float32)
        text_encoder_lora_layers = text_encoder_lora_state_dict(text_encoder)
    else:
        text_encoder_lora_layers = None

    LoraLoaderMixin.save_lora_weights(
        save_directory=output_dir,
        unet_lora_layers=unet_lora_layers,
        text_encoder_lora_layers=text_encoder_lora_layers,
    )


if __name__ == "__main__":
    parser = argparse.ArgumentParser()
    parser.add_argument("--config", type=str, default="./configs/tuneavideo.yaml")
    args = parser.parse_args()

    config = OmegaConf.load(args.config)
    main(**config)