The trainML job execution environment presets a variety of environment variables to make the development of worker training scripts easier and more flexible.
TRAINML_JOB_NAME: The name of the job with spaces converted to underscores.
TRAINML_JOB_ID: A unique ID specific to this job.
TRAINML_WORKER_NUM: The number of the worker in a multi-worker training job.
TRAINML_WORKER_ID: A unique ID specific to this job worker.
TRAINML_DATA_PATH: The local directory of the data loaded through the
Datasetoption. This directory is read-only.
TRAINML_OUTPUT_PATH: The local directory that will be uploaded to the
Output storage pathafter job completion.
TRAINML_MODEL_PATH: The local directory containing the model code. This is the default working directory when a worker starts.
TRAINML_CHECKPOINT_PATH: The local directory of the checkpoints loaded through the
Checkpointoption. This directory is read-only.
TRAINML_CLIENT_IP: The IP address a worker can use to communicate back to the customer running the connection utility. This will only be accessible when the customer is connected.
TRAINML_DEVICE_NAME: (CloudBender Devices Only) The friendly name of the device the endpoint is running on.
TRAINML_DEVICE_ID: (CloudBender Devices Only) The unique ID of the device the endpoint is running on.