[SDK] Use dictionary unpacking to pass trainer function arguments

[astefanutti]

What you would like to be added?

A more straightforward way to pass arguments to the training function, e.g.:

def train_func(batch_size=64,
               test_batch_size=1000,
               epochs=10,
               lr=1.0,
               gamma=0.7,
               dry_run=False,
               seed=1,
               log_interval=10,
               save_model=False):
...

client.train(
    trainer=Trainer(
        func=train_func,
        func_args={
            "batch_size": 64,
            "test_batch_size": 1000,
            "epochs": 10,
            "lr": 1.0,
            "gamma": 0.7,
            "dry_run": False,
            "seed": 1,
            "log_interval": 10,
            "save_model": False,
        },
    ),
)

Why is this needed?

It feels closer to what would be typically written if the training script would run standalone.

Love this feature?

Give it a 👍 We prioritize the features with most 👍

[andreyvelich]

/remove-label lifecycle/needs-triage

[andreyvelich]

/area sdk