train.py

#!/usr/bin/env python3

"""Trains Karras et al. (2022) diffusion models."""

import argparse
from copy import deepcopy
from functools import partial
import importlib.util
import math
import json
from pathlib import Path
import time

import accelerate
import safetensors.torch as safetorch
import torch
import torch._dynamo
from torch import distributed as dist
from torch import multiprocessing as mp
from torch import optim
from torch.utils import data, flop_counter
from torchvision import datasets, transforms, utils
from tqdm.auto import tqdm

import k_diffusion as K


def ensure_distributed():
    if not dist.is_initialized():
        dist.init_process_group(world_size=1, rank=0, store=dist.HashStore())


def main():
    p = argparse.ArgumentParser(description=__doc__,
                                formatter_class=argparse.ArgumentDefaultsHelpFormatter)
    p.add_argument('--batch-size', type=int, default=64,
                   help='the batch size')
    p.add_argument('--checkpointing', action='store_true',
                   help='enable gradient checkpointing')
    p.add_argument('--clip-model', type=str, default='ViT-B/16',
                   choices=K.evaluation.CLIPFeatureExtractor.available_models(),
                   help='the CLIP model to use to evaluate')
    p.add_argument('--compile', action='store_true',
                   help='compile the model')
    p.add_argument('--config', type=str, required=True,
                   help='the configuration file')
    p.add_argument('--demo-every', type=int, default=500,
                   help='save a demo grid every this many steps')
    p.add_argument('--dinov2-model', type=str, default='vitl14',
                   choices=K.evaluation.DINOv2FeatureExtractor.available_models(),
                   help='the DINOv2 model to use to evaluate')
    p.add_argument('--end-step', type=int, default=None,
                   help='the step to end training at')
    p.add_argument('--evaluate-every', type=int, default=10000,
                   help='evaluate every this many steps')
    p.add_argument('--evaluate-n', type=int, default=2000,
                   help='the number of samples to draw to evaluate')
    p.add_argument('--evaluate-only', action='store_true',
                   help='evaluate instead of training')
    p.add_argument('--evaluate-with', type=str, default='inception',
                   choices=['inception', 'clip', 'dinov2'],
                   help='the feature extractor to use for evaluation')
    p.add_argument('--gns', action='store_true',
                   help='measure the gradient noise scale (DDP only, disables stratified sampling)')
    p.add_argument('--grad-accum-steps', type=int, default=1,
                   help='the number of gradient accumulation steps')
    p.add_argument('--lr', type=float,
                   help='the learning rate')
    p.add_argument('--mixed-precision', type=str,
                   help='the mixed precision type')
    p.add_argument('--name', type=str, default='model',
                   help='the name of the run')
    p.add_argument('--num-workers', type=int, default=8,
                   help='the number of data loader workers')
    p.add_argument('--reset-ema', action='store_true',
                   help='reset the EMA')
    p.add_argument('--resume', type=str,
                   help='the checkpoint to resume from')
    p.add_argument('--resume-inference', type=str,
                   help='the inference checkpoint to resume from')
    p.add_argument('--sample-n', type=int, default=64,
                   help='the number of images to sample for demo grids')
    p.add_argument('--save-every', type=int, default=10000,
                   help='save every this many steps')
    p.add_argument('--seed', type=int,
                   help='the random seed')
    p.add_argument('--start-method', type=str, default='spawn',
                   choices=['fork', 'forkserver', 'spawn'],
                   help='the multiprocessing start method')
    p.add_argument('--wandb-entity', type=str,
                   help='the wandb entity name')
    p.add_argument('--wandb-group', type=str,
                   help='the wandb group name')
    p.add_argument('--wandb-project', type=str,
                   help='the wandb project name (specify this to enable wandb)')
    p.add_argument('--wandb-save-model', action='store_true',
                   help='save model to wandb')
    args = p.parse_args()

    mp.set_start_method(args.start_method)
    torch.backends.cuda.matmul.allow_tf32 = True
    try:
        torch._dynamo.config.automatic_dynamic_shapes = False
    except AttributeError:
        pass

    config = K.config.load_config(args.config)
    model_config = config['model']
    dataset_config = config['dataset']
    opt_config = config['optimizer']
    sched_config = config['lr_sched']
    ema_sched_config = config['ema_sched']

    # TODO: allow non-square input sizes
    assert len(model_config['input_size']) == 2 and model_config['input_size'][0] == model_config['input_size'][1]
    size = model_config['input_size']

    accelerator = accelerate.Accelerator(gradient_accumulation_steps=args.grad_accum_steps, mixed_precision=args.mixed_precision)
    ensure_distributed()
    device = accelerator.device
    unwrap = accelerator.unwrap_model
    print(f'Process {accelerator.process_index} using device: {device}', flush=True)
    accelerator.wait_for_everyone()
    if accelerator.is_main_process:
        print(f'World size: {accelerator.num_processes}', flush=True)
        print(f'Batch size: {args.batch_size * accelerator.num_processes}', flush=True)

    if args.seed is not None:
        seeds = torch.randint(-2 ** 63, 2 ** 63 - 1, [accelerator.num_processes], generator=torch.Generator().manual_seed(args.seed))
        torch.manual_seed(seeds[accelerator.process_index])
    demo_gen = torch.Generator().manual_seed(torch.randint(-2 ** 63, 2 ** 63 - 1, ()).item())
    elapsed = 0.0

    inner_model = K.config.make_model(config)
    inner_model_ema = deepcopy(inner_model)

    if args.compile:
        inner_model.compile()
        # inner_model_ema.compile()

    if accelerator.is_main_process:
        print(f'Parameters: {K.utils.n_params(inner_model):,}')

    # If logging to wandb, initialize the run
    use_wandb = accelerator.is_main_process and args.wandb_project
    if use_wandb:
        import wandb
        log_config = vars(args)
        log_config['config'] = config
        log_config['parameters'] = K.utils.n_params(inner_model)
        wandb.init(project=args.wandb_project, entity=args.wandb_entity, group=args.wandb_group, config=log_config, save_code=True)

    lr = opt_config['lr'] if args.lr is None else args.lr
    groups = inner_model.param_groups(lr)
    if opt_config['type'] == 'adamw':
        opt = optim.AdamW(groups,
                          lr=lr,
                          betas=tuple(opt_config['betas']),
                          eps=opt_config['eps'],
                          weight_decay=opt_config['weight_decay'])
    elif opt_config['type'] == 'adam8bit':
        import bitsandbytes as bnb
        opt = bnb.optim.Adam8bit(groups,
                                 lr=lr,
                                 betas=tuple(opt_config['betas']),
                                 eps=opt_config['eps'],
                                 weight_decay=opt_config['weight_decay'])
    elif opt_config['type'] == 'sgd':
        opt = optim.SGD(groups,
                        lr=lr,
                        momentum=opt_config.get('momentum', 0.),
                        nesterov=opt_config.get('nesterov', False),
                        weight_decay=opt_config.get('weight_decay', 0.))
    else:
        raise ValueError('Invalid optimizer type')

    if sched_config['type'] == 'inverse':
        sched = K.utils.InverseLR(opt,
                                  inv_gamma=sched_config['inv_gamma'],
                                  power=sched_config['power'],
                                  warmup=sched_config['warmup'])
    elif sched_config['type'] == 'exponential':
        sched = K.utils.ExponentialLR(opt,
                                      num_steps=sched_config['num_steps'],
                                      decay=sched_config['decay'],
                                      warmup=sched_config['warmup'])
    elif sched_config['type'] == 'constant':
        sched = K.utils.ConstantLRWithWarmup(opt, warmup=sched_config['warmup'])
    else:
        raise ValueError('Invalid schedule type')

    assert ema_sched_config['type'] == 'inverse'
    ema_sched = K.utils.EMAWarmup(power=ema_sched_config['power'],
                                  max_value=ema_sched_config['max_value'])
    ema_stats = {}

    tf = transforms.Compose([
        transforms.Resize(size[0], interpolation=transforms.InterpolationMode.BICUBIC),
        transforms.CenterCrop(size[0]),
        K.augmentation.KarrasAugmentationPipeline(model_config['augment_prob'], disable_all=model_config['augment_prob'] == 0),
    ])

    if dataset_config['type'] == 'imagefolder':
        train_set = K.utils.FolderOfImages(dataset_config['location'], transform=tf)
    elif dataset_config['type'] == 'imagefolder-class':
        train_set = datasets.ImageFolder(dataset_config['location'], transform=tf)
    elif dataset_config['type'] == 'cifar10':
        train_set = datasets.CIFAR10(dataset_config['location'], train=True, download=True, transform=tf)
    elif dataset_config['type'] == 'mnist':
        train_set = datasets.MNIST(dataset_config['location'], train=True, download=True, transform=tf)
    elif dataset_config['type'] == 'huggingface':
        from datasets import load_dataset
        train_set = load_dataset(dataset_config['location'])
        train_set.set_transform(partial(K.utils.hf_datasets_augs_helper, transform=tf, image_key=dataset_config['image_key']))
        train_set = train_set['train']
    elif dataset_config['type'] == 'custom':
        location = (Path(args.config).parent / dataset_config['location']).resolve()
        spec = importlib.util.spec_from_file_location('custom_dataset', location)
        module = importlib.util.module_from_spec(spec)
        spec.loader.exec_module(module)
        get_dataset = getattr(module, dataset_config.get('get_dataset', 'get_dataset'))
        custom_dataset_config = dataset_config.get('config', {})
        train_set = get_dataset(custom_dataset_config, transform=tf)
    else:
        raise ValueError('Invalid dataset type')

    if accelerator.is_main_process:
        try:
            print(f'Number of items in dataset: {len(train_set):,}')
        except TypeError:
            pass

    image_key = dataset_config.get('image_key', 0)
    num_classes = dataset_config.get('num_classes', 0)
    cond_dropout_rate = dataset_config.get('cond_dropout_rate', 0.1)
    class_key = dataset_config.get('class_key', 1)

    train_dl = data.DataLoader(train_set, args.batch_size, shuffle=True, drop_last=True,
                               num_workers=args.num_workers, persistent_workers=True, pin_memory=True)

    inner_model, inner_model_ema, opt, train_dl = accelerator.prepare(inner_model, inner_model_ema, opt, train_dl)

    with torch.no_grad(), K.models.flops.flop_counter() as fc:
        x = torch.zeros([1, model_config['input_channels'], size[0], size[1]], device=device)
        sigma = torch.ones([1], device=device)
        extra_args = {}
        if getattr(unwrap(inner_model), "num_classes", 0):
            extra_args['class_cond'] = torch.zeros([1], dtype=torch.long, device=device)
        inner_model(x, sigma, **extra_args)
        if accelerator.is_main_process:
            print(f"Forward pass GFLOPs: {fc.flops / 1_000_000_000:,.3f}", flush=True)

    if use_wandb:
        wandb.watch(inner_model)
    if accelerator.num_processes == 1:
        args.gns = False
    if args.gns:
        gns_stats_hook = K.gns.DDPGradientStatsHook(inner_model)
        gns_stats = K.gns.GradientNoiseScale()
    else:
        gns_stats = None
    sigma_min = model_config['sigma_min']
    sigma_max = model_config['sigma_max']
    sample_density = K.config.make_sample_density(model_config)

    model = K.config.make_denoiser_wrapper(config)(inner_model)
    model_ema = K.config.make_denoiser_wrapper(config)(inner_model_ema)

    state_path = Path(f'{args.name}_state.json')

    if state_path.exists() or args.resume:
        if args.resume:
            ckpt_path = args.resume
        if not args.resume:
            state = json.load(open(state_path))
            ckpt_path = state['latest_checkpoint']
        if accelerator.is_main_process:
            print(f'Resuming from {ckpt_path}...')
        ckpt = torch.load(ckpt_path, map_location='cpu')
        unwrap(model.inner_model).load_state_dict(ckpt['model'])
        unwrap(model_ema.inner_model).load_state_dict(ckpt['model_ema'])
        opt.load_state_dict(ckpt['opt'])
        sched.load_state_dict(ckpt['sched'])
        ema_sched.load_state_dict(ckpt['ema_sched'])
        ema_stats = ckpt.get('ema_stats', ema_stats)
        epoch = ckpt['epoch'] + 1
        step = ckpt['step'] + 1
        if args.gns and ckpt.get('gns_stats', None) is not None:
            gns_stats.load_state_dict(ckpt['gns_stats'])
        demo_gen.set_state(ckpt['demo_gen'])
        elapsed = ckpt.get('elapsed', 0.0)

        del ckpt
    else:
        epoch = 0
        step = 0

    if args.reset_ema:
        unwrap(model.inner_model).load_state_dict(unwrap(model_ema.inner_model).state_dict())
        ema_sched = K.utils.EMAWarmup(power=ema_sched_config['power'],
                                      max_value=ema_sched_config['max_value'])
        ema_stats = {}

    if args.resume_inference:
        if accelerator.is_main_process:
            print(f'Loading {args.resume_inference}...')
        ckpt = safetorch.load_file(args.resume_inference)
        unwrap(model.inner_model).load_state_dict(ckpt)
        unwrap(model_ema.inner_model).load_state_dict(ckpt)
        del ckpt

    evaluate_enabled = args.evaluate_every > 0 and args.evaluate_n > 0
    metrics_log = None
    if evaluate_enabled:
        if args.evaluate_with == 'inception':
            extractor = K.evaluation.InceptionV3FeatureExtractor(device=device)
        elif args.evaluate_with == 'clip':
            extractor = K.evaluation.CLIPFeatureExtractor(args.clip_model, device=device)
        elif args.evaluate_with == 'dinov2':
            extractor = K.evaluation.DINOv2FeatureExtractor(args.dinov2_model, device=device)
        else:
            raise ValueError('Invalid evaluation feature extractor')
        train_iter = iter(train_dl)
        if accelerator.is_main_process:
            print('Computing features for reals...')
        reals_features = K.evaluation.compute_features(accelerator, lambda x: next(train_iter)[image_key][1], extractor, args.evaluate_n, args.batch_size)
        if accelerator.is_main_process and not args.evaluate_only:
            metrics_log = K.utils.CSVLogger(f'{args.name}_metrics.csv', ['step', 'time', 'loss', 'fid', 'kid'])
        del train_iter

    cfg_scale = 1.

    def make_cfg_model_fn(model):
        def cfg_model_fn(x, sigma, class_cond):
            x_in = torch.cat([x, x])
            sigma_in = torch.cat([sigma, sigma])
            class_uncond = torch.full_like(class_cond, num_classes)
            class_cond_in = torch.cat([class_uncond, class_cond])
            out = model(x_in, sigma_in, class_cond=class_cond_in)
            out_uncond, out_cond = out.chunk(2)
            return out_uncond + (out_cond - out_uncond) * cfg_scale
        if cfg_scale != 1:
            return cfg_model_fn
        return model

    @torch.no_grad()
    @K.utils.eval_mode(model_ema)
    def demo():
        if accelerator.is_main_process:
            tqdm.write('Sampling...')
        filename = f'{args.name}_demo_{step:08}.png'
        n_per_proc = math.ceil(args.sample_n / accelerator.num_processes)
        x = torch.randn([accelerator.num_processes, n_per_proc, model_config['input_channels'], size[0], size[1]], generator=demo_gen).to(device)
        dist.broadcast(x, 0)
        x = x[accelerator.process_index] * sigma_max
        model_fn, extra_args = model_ema, {}
        if num_classes:
            class_cond = torch.randint(0, num_classes, [accelerator.num_processes, n_per_proc], generator=demo_gen).to(device)
            dist.broadcast(class_cond, 0)
            extra_args['class_cond'] = class_cond[accelerator.process_index]
            model_fn = make_cfg_model_fn(model_ema)
        sigmas = K.sampling.get_sigmas_karras(50, sigma_min, sigma_max, rho=7., device=device)
        x_0 = K.sampling.sample_dpmpp_2m_sde(model_fn, x, sigmas, extra_args=extra_args, eta=0.0, solver_type='heun', disable=not accelerator.is_main_process)
        x_0 = accelerator.gather(x_0)[:args.sample_n]
        if accelerator.is_main_process:
            grid = utils.make_grid(x_0, nrow=math.ceil(args.sample_n ** 0.5), padding=0)
            K.utils.to_pil_image(grid).save(filename)
            if use_wandb:
                wandb.log({'demo_grid': wandb.Image(filename)}, step=step)

    @torch.no_grad()
    @K.utils.eval_mode(model_ema)
    def evaluate():
        if not evaluate_enabled:
            return
        if accelerator.is_main_process:
            tqdm.write('Evaluating...')
        sigmas = K.sampling.get_sigmas_karras(50, sigma_min, sigma_max, rho=7., device=device)
        def sample_fn(n):
            x = torch.randn([n, model_config['input_channels'], size[0], size[1]], device=device) * sigma_max
            model_fn, extra_args = model_ema, {}
            if num_classes:
                extra_args['class_cond'] = torch.randint(0, num_classes, [n], device=device)
                model_fn = make_cfg_model_fn(model_ema)
            x_0 = K.sampling.sample_dpmpp_2m_sde(model_fn, x, sigmas, extra_args=extra_args, eta=0.0, solver_type='heun', disable=True)
            return x_0
        fakes_features = K.evaluation.compute_features(accelerator, sample_fn, extractor, args.evaluate_n, args.batch_size)
        if accelerator.is_main_process:
            fid = K.evaluation.fid(fakes_features, reals_features)
            kid = K.evaluation.kid(fakes_features, reals_features)
            print(f'FID: {fid.item():g}, KID: {kid.item():g}')
            if accelerator.is_main_process and metrics_log is not None:
                metrics_log.write(step, elapsed, ema_stats['loss'], fid.item(), kid.item())
            if use_wandb:
                wandb.log({'FID': fid.item(), 'KID': kid.item()}, step=step)

    def save():
        accelerator.wait_for_everyone()
        filename = f'{args.name}_{step:08}.pth'
        if accelerator.is_main_process:
            tqdm.write(f'Saving to {filename}...')
        inner_model = unwrap(model.inner_model)
        inner_model_ema = unwrap(model_ema.inner_model)
        obj = {
            'config': config,
            'model': inner_model.state_dict(),
            'model_ema': inner_model_ema.state_dict(),
            'opt': opt.state_dict(),
            'sched': sched.state_dict(),
            'ema_sched': ema_sched.state_dict(),
            'epoch': epoch,
            'step': step,
            'gns_stats': gns_stats.state_dict() if gns_stats is not None else None,
            'ema_stats': ema_stats,
            'demo_gen': demo_gen.get_state(),
            'elapsed': elapsed,
        }
        accelerator.save(obj, filename)
        if accelerator.is_main_process:
            state_obj = {'latest_checkpoint': filename}
            json.dump(state_obj, open(state_path, 'w'))
        if args.wandb_save_model and use_wandb:
            wandb.save(filename)

    if args.evaluate_only:
        if not evaluate_enabled:
            raise ValueError('--evaluate-only requested but evaluation is disabled')
        evaluate()
        return

    losses_since_last_print = []

    try:
        while True:
            for batch in tqdm(train_dl, smoothing=0.1, disable=not accelerator.is_main_process):
                if device.type == 'cuda':
                    start_timer = torch.cuda.Event(enable_timing=True)
                    end_timer = torch.cuda.Event(enable_timing=True)
                    torch.cuda.synchronize()
                    start_timer.record()
                else:
                    start_timer = time.time()

                with accelerator.accumulate(model):
                    reals, _, aug_cond = batch[image_key]
                    class_cond, extra_args = None, {}
                    if num_classes:
                        class_cond = batch[class_key]
                        drop = torch.rand(class_cond.shape, device=class_cond.device)
                        class_cond.masked_fill_(drop < cond_dropout_rate, num_classes)
                        extra_args['class_cond'] = class_cond
                    noise = torch.randn_like(reals)
                    with K.utils.enable_stratified_accelerate(accelerator, disable=args.gns):
                        sigma = sample_density([reals.shape[0]], device=device)
                    with K.models.checkpointing(args.checkpointing):
                        losses = model.loss(reals, noise, sigma, aug_cond=aug_cond, **extra_args)
                    loss = accelerator.gather(losses).mean().item()
                    losses_since_last_print.append(loss)
                    accelerator.backward(losses.mean())
                    if args.gns:
                        sq_norm_small_batch, sq_norm_large_batch = gns_stats_hook.get_stats()
                        gns_stats.update(sq_norm_small_batch, sq_norm_large_batch, reals.shape[0], reals.shape[0] * accelerator.num_processes)
                    if accelerator.sync_gradients:
                        accelerator.clip_grad_norm_(model.parameters(), 1.)
                    opt.step()
                    sched.step()
                    opt.zero_grad()

                    ema_decay = ema_sched.get_value()
                    K.utils.ema_update_dict(ema_stats, {'loss': loss}, ema_decay ** (1 / args.grad_accum_steps))
                    if accelerator.sync_gradients:
                        K.utils.ema_update(model, model_ema, ema_decay)
                        ema_sched.step()

                if device.type == 'cuda':
                    end_timer.record()
                    torch.cuda.synchronize()
                    elapsed += start_timer.elapsed_time(end_timer) / 1000
                else:
                    elapsed += time.time() - start_timer

                if step % 25 == 0:
                    loss_disp = sum(losses_since_last_print) / len(losses_since_last_print)
                    losses_since_last_print.clear()
                    avg_loss = ema_stats['loss']
                    if accelerator.is_main_process:
                        if args.gns:
                            tqdm.write(f'Epoch: {epoch}, step: {step}, loss: {loss_disp:g}, avg loss: {avg_loss:g}, gns: {gns_stats.get_gns():g}')
                        else:
                            tqdm.write(f'Epoch: {epoch}, step: {step}, loss: {loss_disp:g}, avg loss: {avg_loss:g}')

                if use_wandb:
                    log_dict = {
                        'epoch': epoch,
                        'loss': loss,
                        'lr': sched.get_last_lr()[0],
                        'ema_decay': ema_decay,
                    }
                    if args.gns:
                        log_dict['gradient_noise_scale'] = gns_stats.get_gns()
                    wandb.log(log_dict, step=step)

                step += 1

                if step % args.demo_every == 0:
                    demo()

                if evaluate_enabled and step > 0 and step % args.evaluate_every == 0:
                    evaluate()

                if step == args.end_step or (step > 0 and step % args.save_every == 0):
                    save()

                if step == args.end_step:
                    if accelerator.is_main_process:
                        tqdm.write('Done!')
                    return

            epoch += 1
    except KeyboardInterrupt:
        pass


if __name__ == '__main__':
    main()