PyTorch Lightning Introduces Improved Lightning Logger Support for Weight & Biases and Neputne.ai

The Lightning Team recently announced that it has collaborated closely with Neptune.ai and Weight & Biases Teams to provide the best experience in terms of Logging and Experiment Tracking. The launch of PyTorch Lightning v1.5 marked a major leap in reliability to support the increasingly complex demands of the leading AI organizations and prestigious research labs that rely on Lightning to develop and deploy AI at scale. To better support the fast-growing community, PyTorch Lightning aims at becoming the simplest, most flexible framework for expediting any kind of deep learning research to production.

Lightning 1.5 adds new methods to WandbLogger that help you elevate your logging experience inside PL by giving you the ability to monitor your model weights and give you the functionality to log other artifacts such as text, tables, images, and model checkpoints. By using watch and unwatch methods, you can track gradients, parameter histogram, model topology and it can be used directly within the LightningModule. 

But the most exciting feature is the support of log_table directly though the WandbLogger.

Below is an example of how W&B tables can be used to provide insightful reports.

With the v1.5 release, NeptuneLogger is revamped and relies on the new improved neptune-client.

To utilize the NeptuneLogger with Lightning, simply do the following:

Using the NeptuneLogger, you can log images, texts, and artifacts.

And using the log_model_summary method, you can track the summary of the parameters of your models.

The Lightning Team seems more than ever committed to provide users with the best experience possible to anyone doing optimization with PyTorch and the PyTorch Lightning API being already stable, breaking changes will be minimal. You can learn more though the WandbLogger documentation and the NeptuneLogger documentation.