NeptuneLogger

class composer.loggers.NeptuneLogger(*, project=None, api_token=None, rank_zero_only=True, upload_checkpoints=False, base_namespace='training', mode=None, **neptune_kwargs)

Log to neptune.ai.

For instructions, see the integration guide.

Parameters

• project (str, optional) – The name of your Neptune project, in the form “workspace-name/project-name”. If you leave it empty, the NEPTUNE_PROJECT environment variable will be used.

• api_token (str, optional) – Your Neptune API token. You can leave out this argument if you save your token to the NEPTUNE_API_TOKEN environment variable (recommended). You can find your API token in the user menu of the Neptune web app.

• rank_zero_only (bool) – Whether to log only on the rank-zero process (default: True).

• upload_checkpoints (bool) – Whether the logger should upload checkpoints to Neptune (default: False).

• base_namespace (str, optional) – The name of the base namespace where the metadata is logged (default: “training”).

• neptune_kwargs (dict[str, Any], optional) – Any additional keyword arguments to the neptune.init_run() function. For options, see the Run API reference.

property base_handler

Gets a handler for the base logging namespace.

Use the handler to log extra metadata to the run and organize it under the base namespace (default: “training”). You can operate on it like a run object: Access a path inside the handler and assign metadata to it with = or other Neptune logging methods.

can_upload_files()

Whether the logger supports uploading files.

property neptune_run

Gets the Neptune run object from a NeptuneLogger instance.

To log additional metadata to the run, access a path inside the run and assign metadata with = or other Neptune logging methods.