I - Python SDK v2 syntax

# 1. Authenticate / connect
credential = DefaultAzureCredential()
ml_client = MLClient(...)

# 2. Define → submit
job = command(...)
returned_job = ml_client.jobs.create_or_update(job)

# 3. Track inside training code
mlflow.set_experiment(...)
mlflow.autolog()

# 4. Sweep + early termination
BanditPolicy(
    slack_factor=...,
    evaluation_interval=...,
    delay_evaluation=...
)

I1 - authentication, connection and jobs
I2 - MLflow
[I3 - sweep]