wandb

allennlp.training.callbacks.wandb

[SOURCE]

---

WandBCallback

@TrainerCallback.register("wandb")
class WandBCallback(LogWriterCallback):
 | def __init__(
 |     self,
 |     serialization_dir: str,
 |     summary_interval: int = 100,
 |     distribution_interval: Optional[int] = None,
 |     batch_size_interval: Optional[int] = None,
 |     should_log_parameter_statistics: bool = True,
 |     should_log_learning_rate: bool = False,
 |     project: Optional[str] = None,
 |     entity: Optional[str] = None,
 |     group: Optional[str] = None,
 |     name: Optional[str] = None,
 |     notes: Optional[str] = None,
 |     tags: Optional[List[str]] = None,
 |     watch_model: bool = True,
 |     files_to_save: Tuple[str, ...] = ("config.json", "out.log"),
 |     wandb_kwargs: Optional[Dict[str, Any]] = None
 | ) -> None

Logs training runs to Weights & Biases.

Note

This requires the environment variable 'WANDB_API_KEY' to be set in order to authenticate with Weights & Biases. If not set, you may be prompted to log in or upload the experiment to an anonymous account.

In addition to the parameters that LogWriterCallback takes, there are several other parameters specific to WandBWriter listed below.

Parameters

• project : Optional[str], optional (default = None)
  The name of the W&B project to save the training run to.
• entity : Optional[str], optional (default = None)
  The username or team name to send the run to. If not specified, the default will be used.
• group : Optional[str], optional (default = None)
  Specify a group to organize individual runs into a larger experiment.
• name : Optional[str], optional (default = None)
  A short display name for this run, which is how you'll identify this run in the W&B UI. By default a random name is generated.
• notes : Optional[str], optional (default = None)
  A description of the run.
• tags : Optional[List[str]], optional (default = None)
  Tags to assign to the training run in W&B.
• watch_model : bool, optional (default = True)
  Whether or not W&B should watch the Model.
• files_to_save : Tuple[str, ...], optional (default = ("config.json", "out.log"))
  Extra files in the serialization directory to save to the W&B training run.
• wandb_kwargs : Optional[Dict[str, Any]], optional (default = None)
  Additional key word arguments to pass to wandb.init().

log_scalars

class WandBCallback(LogWriterCallback):
 | ...
 | def log_scalars(
 |     self,
 |     scalars: Dict[str, Union[int, float]],
 |     log_prefix: str = "",
 |     epoch: Optional[int] = None
 | ) -> None

log_tensors

class WandBCallback(LogWriterCallback):
 | ...
 | def log_tensors(
 |     self,
 |     tensors: Dict[str, torch.Tensor],
 |     log_prefix: str = "",
 |     epoch: Optional[int] = None
 | ) -> None

on_start

class WandBCallback(LogWriterCallback):
 | ...
 | def on_start(
 |     self,
 |     trainer: "GradientDescentTrainer",
 |     is_primary: bool = True,
 |     **kwargs
 | ) -> None

close

class WandBCallback(LogWriterCallback):
 | ...
 | def close(self) -> None

state_dict

class WandBCallback(LogWriterCallback):
 | ...
 | def state_dict(self) -> Dict[str, Any]

load_state_dict

class WandBCallback(LogWriterCallback):
 | ...
 | def load_state_dict(self, state_dict: Dict[str, Any]) -> None