Merlin

ray_train.py

import ray
from ray.train import Trainer

def train_func(config):
    """
    Training function passed to Ray Train to be distributed.
    Loads the dataset, builds, fits and saves the model.
    """
	...

def main():
	ray.init("<RAY_CLUSTER_ADDRESS>")

	config = {...}

	trainer = Trainer(backend="tensorflow", num_workers=num_workers, use_gpu=True)
	trainer.start()
	log.info("Starting to train.")
	start = time.time()
	results = trainer.run(train_func, config=config)
	log.info(f"Training finished in {time.time() - start} seconds.")
	trainer.shutdown()
	log.info(f"Training results: {results[0]}")