How to use the mmcv.runner.Runner function in mmcv

try:
        import torch
        import torch.nn as nn
        from torch.utils.data import DataLoader
    except ImportError:
        warnings.warn('Skipping test_save_checkpoint in the absense of torch')
        return

    import mmcv.runner
    wandb_mock = MagicMock()
    hook = mmcv.runner.hooks.WandbLoggerHook()
    hook.wandb = wandb_mock
    loader = DataLoader(torch.ones((5, 5)))

    model = nn.Linear(1, 1)
    runner = mmcv.runner.Runner(
        model=model,
        batch_processor=lambda model, x, **kwargs: {
            'log_vars': {
                "accuracy": 0.98
            },
            'num_samples': 5
        })
    runner.register_hook(hook)
    runner.run([loader, loader], [('train', 1), ('val', 1)], 1)
    wandb_mock.init.assert_called()
    wandb_mock.log.assert_called_with({'accuracy/val': 0.98}, step=5)
    wandb_mock.join.assert_called()

from mmcv.runner import Runner
from mmcv.runner.utils import obj_from_dict
import torch

class MultiLRRunner(Runner):
    def init_optimizer(self, optimizer):
        """Init the optimizer.

        Args:
            optimizer (dict or :obj:`~torch.optim.Optimizer`): Either an
                optimizer object or a dict used for constructing the optimizer.

        Returns:
            :obj:`~torch.optim.Optimizer`: An optimizer object.

        Examples:
             optimizer = dict(type='SGD', lr=0.01, momentum=0.9)
             type(runner.init_optimizer(optimizer))
            
        """
        if isinstance(optimizer, dict):

def _non_dist_train(model, dataset, cfg, validate=False):
    # prepare data loaders
    data_loaders = [
        build_dataloader(
            dataset,
            cfg.data.videos_per_gpu,
            cfg.data.workers_per_gpu,
            cfg.gpus,
            dist=False)
    ]
    # put model on gpus
    model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()
    # build runner
    runner = Runner(model, batch_processor, cfg.optimizer, cfg.work_dir,
                    cfg.log_level)
    runner.register_training_hooks(cfg.lr_config, cfg.optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)

    if cfg.resume_from:
        runner.resume(cfg.resume_from)
    elif cfg.load_from:
        runner.load_checkpoint(cfg.load_from)
    runner.run(data_loaders, cfg.workflow, cfg.total_epochs)

amp_opt_level = 'O1' if cfg.apex.type == "float16" else 'O0'
        model, optimizer = amp.initialize(
            model, optimizer, opt_level=amp_opt_level, loss_scale=cfg.apex.loss_scale
        )

    # put model on gpus
    find_unused_parameters = cfg.get('find_unused_parameters', False)
    # Sets the `find_unused_parameters` parameter in
    # torch.nn.parallel.DistributedDataParallel
    model = MMDistributedDataParallel(
        model.cuda(),
        device_ids=[torch.cuda.current_device()],
        broadcast_buffers=False,
        find_unused_parameters=find_unused_parameters)
    # build runner
    runner = Runner(
        model, batch_processor, optimizer, cfg.work_dir, cfg.log_level, logger
    )

    # register optimizer hooks
    if cfg.apex.use_mixed_precision:
        optimizer_config = DistApexOptimizerHook(**cfg.optimizer_config)
    else:
        optimizer_config = DistOptimizerHook(**cfg.optimizer_config)
    logger.info("Register Optimizer Hook...")
    runner.register_training_hooks(
        cfg.lr_config, optimizer_config, cfg.checkpoint_config,
        log_config={"interval": cfg.log_config['interval'], 'hooks': []}
    )

    # register self-defined logging hooks
    for info in cfg.log_config['hooks']:

eval_dataset=None, vis_dataset=None, validate=False, logger=None
):
    # prepare data loaders
    data_loaders = [
        build_data_loader(
            train_dataset,
            cfg.data.imgs_per_gpu,
            cfg.data.workers_per_gpu,
            cfg.gpus,
            dist=False)
    ]
    # put model on gpus
    model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()
    # build runner
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(
        model, batch_processor, optimizer, cfg.work_dir, cfg.log_level, logger
    )
    logger.info("Register Optimizer Hook...")
    runner.register_training_hooks(
        cfg.lr_config, cfg.optimizer_config, cfg.checkpoint_config, cfg.log_config
    )
    logger.info("Register EmptyCache Hook...")
    runner.register_hook(
        EmptyCacheHook(before_epoch=True, after_iter=False, after_epoch=True),
        priority='VERY_LOW'
    )

    if cfg.resume_from:
        runner.resume(cfg.resume_from)
    elif cfg.load_from:
        runner.load_checkpoint(cfg.load_from)

def _dist_train_runner(model, dataset, cfg, validate=False)->Runner:

    # put model on gpus
    model = MMDistributedDataParallel(model.cuda())

    # build runner
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(model, batch_processor, optimizer, cfg.work_dir,
                    cfg.log_level)

    # fp16 setting
    fp16_cfg = cfg.get('fp16', None)
    if fp16_cfg is not None:
        optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
                                             **fp16_cfg)
    else:
        optimizer_config = DistOptimizerHook(**cfg.optimizer_config)

    # register hooks
    runner.register_training_hooks(cfg.lr_config, optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)
    runner.register_hook(DistSamplerSeedHook())
    # register eval hooks
    if validate:

validate=False,
                logger=None,
                timestamp=None):
    # prepare data loaders
    dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
    data_loaders = [
        build_dataloader(
            ds, cfg.data.imgs_per_gpu, cfg.data.workers_per_gpu, dist=True)
        for ds in dataset
    ]
    # put model on gpus
    model = MMDistributedDataParallel(model.cuda())

    # build runner
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(
        model, batch_processor, optimizer, cfg.work_dir, logger=logger)
    # an ugly walkaround to make the .log and .log.json filenames the same
    runner.timestamp = timestamp

    # fp16 setting
    fp16_cfg = cfg.get('fp16', None)
    if fp16_cfg is not None:
        optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
                                             **fp16_cfg)
    else:
        optimizer_config = DistOptimizerHook(**cfg.optimizer_config)

    # register hooks
    runner.register_training_hooks(cfg.lr_config, optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)
    runner.register_hook(DistSamplerSeedHook())

def _dist_train(model, dataset, cfg, validate=False):
    # prepare data loaders
    data_loaders = [
        build_dataloader(
            dataset,
            cfg.data.imgs_per_gpu,
            cfg.data.workers_per_gpu,
            dist=True)
    ]
    # put model on gpus
    model = MMDistributedDataParallel(model.cuda())

    # build runner
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(model, batch_processor, optimizer, cfg.work_dir,
                    cfg.log_level)

    # fp16 setting
    fp16_cfg = cfg.get('fp16', None)
    if fp16_cfg is not None:
        optimizer_config = Fp16OptimizerHook(**cfg.optimizer_config,
                                             **fp16_cfg)
    else:
        optimizer_config = DistOptimizerHook(**cfg.optimizer_config)

    # register hooks
    runner.register_training_hooks(cfg.lr_config, optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)
    runner.register_hook(DistSamplerSeedHook())
    # register eval hooks
    if validate:

# prepare data loaders
    dataset = dataset if isinstance(dataset, (list, tuple)) else [dataset]
    data_loaders = [
        build_dataloader(
            ds,
            cfg.data.imgs_per_gpu,
            cfg.data.workers_per_gpu,
            cfg.gpus,
            dist=False) for ds in dataset
    ]
    # put model on gpus
    model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()

    # build runner
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(
        model, batch_processor, optimizer, cfg.work_dir, logger=logger)
    # an ugly walkaround to make the .log and .log.json filenames the same
    runner.timestamp = timestamp
    # fp16 setting
    fp16_cfg = cfg.get('fp16', None)
    if fp16_cfg is not None:
        optimizer_config = Fp16OptimizerHook(
            **cfg.optimizer_config, **fp16_cfg, distributed=False)
    else:
        optimizer_config = cfg.optimizer_config
    runner.register_training_hooks(cfg.lr_config, optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)

    if cfg.resume_from:
        runner.resume(cfg.resume_from)
    elif cfg.load_from:

# prepare data loaders
    # 返回dataloader的迭代器，采用pytorch的DataLoader方法封装数据集
    data_loaders = [
        build_dataloader(
            dataset,
            cfg.data.imgs_per_gpu,
            cfg.data.workers_per_gpu,
            cfg.gpus,
            dist=False)
    ]
    # put model on gpus 这里多GPU输入没用list而是迭代器，注意单GPU是range(0,1),遍历的时候只有0
    model = MMDataParallel(model, device_ids=range(cfg.gpus)).cuda()

    # build runner 
    optimizer = build_optimizer(model, cfg.optimizer)
    runner = Runner(model, batch_processor, optimizer, cfg.work_dir,
                    cfg.log_level)
    # fp16 setting
    fp16_cfg = cfg.get('fp16', None)
    if fp16_cfg is not None:
        optimizer_config = Fp16OptimizerHook(
            **cfg.optimizer_config, **fp16_cfg, distributed=False)
    else:
        optimizer_config = cfg.optimizer_config

    # 注册钩子    
    runner.register_training_hooks(cfg.lr_config, optimizer_config,
                                   cfg.checkpoint_config, cfg.log_config)
    # 断点加载或文件加载数据
    if cfg.resume_from:
        runner.resume(cfg.resume_from)
    elif cfg.load_from: