# Single Train run
trainer = TorchTrainer(
    train_fn,
    train_loop_config={"lr": 0.01},
    scaling_config=air.ScalingConfig(num_workers=2),
    run_config=air.RunConfig(storage_path="/storage/path/1", callbacks=[WandbLoggerCallback()]),
)
result = trainer.fit() # <-- This creates a Tuner() under the hood.

# Great, that works! Let's tune the hyperparameters now.

tuner = Tuner(
    trainer,
    param_space={
        # Hmm, what happens to my original train_loop_config?
        # My IDE isn't able to autofill these magic dict keys...
        "train_loop_config": {"lr": tune.uniform(0, 1)},
        "scaling_config": air.ScalingConfig(
            num_workers=tune.choice([2, 4])
        ),
        # ScalingConfig can be here, but RunConfig can not.
    },
    # Why am I able to specify a new RunConfig? Which one gets used?
    run_config=air.RunConfig(storage_path="/storage/path/2"),
    tune_config=tune.TuneConfig(num_samples=2),
)
results = tuner.fit()

# Why don't I see any wandb logs?
# -> It's because the wandb callback that was defined in the Trainer's RunConfig gets thrown away in favor of the Tuner's RunConfig.

            logger.info(
                "A `RunConfig` was passed to both the `Tuner` and the "
                f"`{trainer.__class__.__name__}`. The run config passed to "
                "the `Tuner` is the one that will be used."
            )