workspace/train_diffusion_unet_lowdim_workspace.py

if __name__ == "__main__":
    import sys
    import os
    import pathlib

    ROOT_DIR = str(pathlib.Path(__file__).parent.parent.parent)
    sys.path.append(ROOT_DIR)
    os.chdir(ROOT_DIR)

import os
import hydra
import torch
from omegaconf import OmegaConf
import pathlib
from torch.utils.data import DataLoader
import copy
import numpy as np
import random
import wandb
import tqdm
import shutil

from diffusion_policy.common.pytorch_util import dict_apply, optimizer_to
from diffusion_policy.workspace.base_workspace import BaseWorkspace
from diffusion_policy.policy.diffusion_unet_lowdim_policy import DiffusionUnetLowdimPolicy
from diffusion_policy.dataset.base_dataset import BaseLowdimDataset
from diffusion_policy.env_runner.base_lowdim_runner import BaseLowdimRunner
from diffusion_policy.common.checkpoint_util import TopKCheckpointManager
from diffusion_policy.common.json_logger import JsonLogger
from diffusion_policy.model.common.lr_scheduler import get_scheduler
from diffusers.training_utils import EMAModel

OmegaConf.register_new_resolver("eval", eval, replace=True)

# %%
class TrainDiffusionUnetLowdimWorkspace(BaseWorkspace):
    include_keys = ['global_step', 'epoch']

    def __init__(self, cfg: OmegaConf, output_dir=None):
        super().__init__(cfg, output_dir=output_dir)

        # set seed
        seed = cfg.training.seed
        torch.manual_seed(seed)
        np.random.seed(seed)
        random.seed(seed)

        # configure model
        self.model: DiffusionUnetLowdimPolicy
        self.model = hydra.utils.instantiate(cfg.policy)

        self.ema_model: DiffusionUnetLowdimPolicy = None
        if cfg.training.use_ema:
            self.ema_model = copy.deepcopy(self.model)

        # configure training state
        self.optimizer = hydra.utils.instantiate(
            cfg.optimizer, params=self.model.parameters())

        self.global_step = 0
        self.epoch = 0

    def run(self):
        cfg = copy.deepcopy(self.cfg)

        # resume training
        if cfg.training.resume:
            lastest_ckpt_path = self.get_checkpoint_path()
            if lastest_ckpt_path.is_file():
                print(f"Resuming from checkpoint {lastest_ckpt_path}")
                self.load_checkpoint(path=lastest_ckpt_path)

        # configure dataset
        dataset: BaseLowdimDataset
        dataset = hydra.utils.instantiate(cfg.task.dataset)
        assert isinstance(dataset, BaseLowdimDataset)
        train_dataloader = DataLoader(dataset, **cfg.dataloader)
        normalizer = dataset.get_normalizer()

        # configure validation dataset
        val_dataset = dataset.get_validation_dataset()
        val_dataloader = DataLoader(val_dataset, **cfg.val_dataloader)

        self.model.set_normalizer(normalizer)
        if cfg.training.use_ema:
            self.ema_model.set_normalizer(normalizer)

        # configure lr scheduler
        lr_scheduler = get_scheduler(
            cfg.training.lr_scheduler,
            optimizer=self.optimizer,
            num_warmup_steps=cfg.training.lr_warmup_steps,
            num_training_steps=(
                len(train_dataloader) * cfg.training.num_epochs) \
                    // cfg.training.gradient_accumulate_every,
            # pytorch assumes stepping LRScheduler every epoch
            # however huggingface diffusers steps it every batch
            last_epoch=self.global_step-1
        )

        # configure ema
        ema: EMAModel = None
        if cfg.training.use_ema:
            ema = hydra.utils.instantiate(
                cfg.ema,
                model=self.ema_model)

        # configure env runner
        env_runner: BaseLowdimRunner
        env_runner = hydra.utils.instantiate(
            cfg.task.env_runner,
            output_dir=self.output_dir)
        assert isinstance(env_runner, BaseLowdimRunner)

        # configure logging
        wandb_run = wandb.init(
            dir=str(self.output_dir),
            config=OmegaConf.to_container(cfg, resolve=True),
            **cfg.logging
        )
        wandb.config.update(
            {
                "output_dir": self.output_dir,
            }
        )

        # configure checkpoint
        topk_manager = TopKCheckpointManager(
            save_dir=os.path.join(self.output_dir, 'checkpoints'),
            **cfg.checkpoint.topk
        )

        # device transfer
        device = torch.device(cfg.training.device)
        self.model.to(device)
        if self.ema_model is not None:
            self.ema_model.to(device)
        optimizer_to(self.optimizer, device)

        # save batch for sampling
        train_sampling_batch = None

        if cfg.training.debug:
            cfg.training.num_epochs = 2
            cfg.training.max_train_steps = 3
            cfg.training.max_val_steps = 3
            cfg.training.rollout_every = 1
            cfg.training.checkpoint_every = 1
            cfg.training.val_every = 1
            cfg.training.sample_every = 1

        # training loop
        log_path = os.path.join(self.output_dir, 'logs.json.txt')
        json_logger = JsonLogger(log_path)
        json_logger.start()
        for local_epoch_idx in range(cfg.training.num_epochs):
            step_log = dict()
            # ========= train for this epoch ==========
            train_losses = list()
            with tqdm.tqdm(train_dataloader, desc=f"Training epoch {self.epoch}",
                    leave=False, mininterval=cfg.training.tqdm_interval_sec) as tepoch:
                for batch_idx, batch in enumerate(tepoch):
                    # device transfer
                    batch = dict_apply(batch, lambda x: x.to(device, non_blocking=True))
                    if train_sampling_batch is None:
                        train_sampling_batch = batch

                    # compute loss
                    raw_loss = self.model.compute_loss(batch)
                    loss = raw_loss / cfg.training.gradient_accumulate_every
                    loss.backward()

                    # step optimizer
                    if self.global_step % cfg.training.gradient_accumulate_every == 0:
                        self.optimizer.step()
                        self.optimizer.zero_grad()
                        lr_scheduler.step()

                    # update ema
                    if cfg.training.use_ema:
                        ema.step(self.model)

                    # logging
                    raw_loss_cpu = raw_loss.item()
                    tepoch.set_postfix(loss=raw_loss_cpu, refresh=False)
                    train_losses.append(raw_loss_cpu)
                    step_log = {
                        'train_loss': raw_loss_cpu,
                        'global_step': self.global_step,
                        'epoch': self.epoch,
                        'lr': lr_scheduler.get_last_lr()[0]
                    }

                    is_last_batch = (batch_idx == (len(train_dataloader)-1))
                    if not is_last_batch:
                        # log of last step is combined with validation and rollout
                        wandb_run.log(step_log, step=self.global_step)
                        json_logger.log(step_log)
                        self.global_step += 1

                    if (cfg.training.max_train_steps is not None) \
                        and batch_idx >= (cfg.training.max_train_steps-1):
                        break

            # at the end of each epoch
            # replace train_loss with epoch average
            train_loss = np.mean(train_losses)
            step_log['train_loss'] = train_loss

            # ========= eval for this epoch ==========
            policy = self.model
            if cfg.training.use_ema:
                policy = self.ema_model
            policy.eval()

            # run rollout
            if (self.epoch % cfg.training.rollout_every) == 0 or self.epoch == cfg.training.num_epochs - 1:
                runner_log = env_runner.run(policy)
                # log all
                step_log.update(runner_log)

            # run validation
            if (self.epoch % cfg.training.val_every) == 0:
                with torch.no_grad():
                    val_losses = list()
                    with tqdm.tqdm(val_dataloader, desc=f"Validation epoch {self.epoch}",
                            leave=False, mininterval=cfg.training.tqdm_interval_sec) as tepoch:
                        for batch_idx, batch in enumerate(tepoch):
                            batch = dict_apply(batch, lambda x: x.to(device, non_blocking=True))
                            loss = self.model.compute_loss(batch)
                            val_losses.append(loss)
                            if (cfg.training.max_val_steps is not None) \
                                and batch_idx >= (cfg.training.max_val_steps-1):
                                break
                    if len(val_losses) > 0:
                        val_loss = torch.mean(torch.tensor(val_losses)).item()
                        # log epoch average validation loss
                        step_log['val_loss'] = val_loss

            # run diffusion sampling on a training batch
            if (self.epoch % cfg.training.sample_every) == 0:
                with torch.no_grad():
                    # sample trajectory from training set, and evaluate difference
                    batch = train_sampling_batch
                    obs_dict = {'obs': batch['obs']}
                    gt_action = batch['action']

                    result = policy.predict_action(obs_dict)
                    if cfg.pred_action_steps_only:
                        pred_action = result['action']
                        start = cfg.n_obs_steps - 1
                        end = start + cfg.n_action_steps
                        gt_action = gt_action[:,start:end]
                    else:
                        pred_action = result['action_pred']
                    mse = torch.nn.functional.mse_loss(pred_action, gt_action)
                    # log
                    step_log['train_action_mse_error'] = mse.item()
                    # release RAM
                    del batch
                    del obs_dict
                    del gt_action
                    del result
                    del pred_action
                    del mse

            # checkpoint
            if self.epoch != 0 and ((self.epoch % cfg.training.checkpoint_every) == 0 or self.epoch == cfg.training.num_epochs - 1):
                # checkpointing
                if cfg.checkpoint.save_last_ckpt:
                    self.save_checkpoint()
                if cfg.checkpoint.save_last_snapshot:
                    self.save_snapshot()

                # sanitize metric names
                metric_dict = dict()
                for key, value in step_log.items():
                    new_key = key.replace('/', '_')
                    metric_dict[new_key] = value

                # We can't copy the last checkpoint here
                # since save_checkpoint uses threads.
                # therefore at this point the file might have been empty!
                topk_ckpt_path = topk_manager.get_ckpt_path(metric_dict)

                if topk_ckpt_path is not None:
                    self.save_checkpoint(path=topk_ckpt_path)
            # ========= eval end for this epoch ==========
            policy.train()

            # end of epoch
            # log of last step is combined with validation and rollout
            wandb_run.log(step_log, step=self.global_step)
            json_logger.log(step_log)
            self.global_step += 1
            self.epoch += 1
        json_logger.stop()

@hydra.main(
    version_base=None,
    config_path=str(pathlib.Path(__file__).parent.parent.joinpath("config")), 
    config_name=pathlib.Path(__file__).stem)
def main(cfg):
    workspace = TrainDiffusionUnetLowdimWorkspace(cfg)
    workspace.run()

if __name__ == "__main__":
    main()