PyTorch Lightning provides a lightweight wrapper for organizing your PyTorch code and easily adding advanced features such as distributed training and 16-bit precision. W&B provides a lightweight wrapper for logging your ML experiments. We're incorporated directly into the PyTorch Lightning library, so you can always check out their documentation.

⚡ Get going lightning-fast with just two lines:

from pytorch_lightning.loggers import WandbLogger
from pytorch_lightning import Trainer
​
wandb_logger = WandbLogger()
trainer = Trainer(logger=wandb_logger)

✅ Check out real examples!

We've created a few examples for you to see how the integration works:

• ​Demo in Google Colab with hyperparameter optimization

• ​Tutorial: Supercharge your Training with Pytorch Lightning + Weights & Biases

• ​Semantic Segmentation with Lightning: optimize neural networks for self-driving cars

• ​A step by step guide to tracking your Lightning model performance

💻 API Reference

WandbLogger

Optional parameters:

• name (str) – display name for the run.

• save_dir (str) – path where data is saved (wandb dir by default).

• offline (bool) – run offline (data can be streamed later to wandb servers).

• id (str) – sets the version, mainly used to resume a previous run.

• version (str) – same as version (legacy).

• anonymous (bool) – enables or explicitly disables anonymous logging.

• project (str) – the name of the project to which this run will belong.

• log_model (bool) – save checkpoints in wandb dir to upload on W&B servers.

• prefix (str) – string to put at the beginning of metric keys.

• sync_step (bool) - Sync Trainer step with wandb step (True by default).

• **kwargs – Additional arguments like entity, group, tags, etc. used by wandb.init can be passed as keyword arguments in this logger.

WandbLogger.watch

Log model topology as well as optionally gradients and weights.

wandb_logger.watch(model, log='gradients', log_freq=100)

Parameters:

• model (nn.Module) – model to be logged.

• log (str) – can be "gradients" (default), "parameters", "all" or None.

• log_freq (int) – step count between logging of gradients and parameters (100 by default).

WandbLogger.log_hyperparams

Record hyperparameter configuration.

Note: this function is called automatically when using LightningModule.save_hyperparameters()

wandb_logger.log_hyperparams(params)

Parameters:

• params (dict) – dictionary with hyperparameter names as keys and configuration values as values

WandbLogger.log_metrics

Record training metrics.

Note: this function is called automatically by LightningModule.log('metric', value)

wandb_logger.log_metrics(metrics, step=None)

Parameters:

• metric (numeric) – dictionary with metric names as keys and measured quantities as values

• step (int|None) – step number at which the metrics should be recorded

****