Automatically track PyTorch Ignite model training progress to Neptune
- Log parameters, metrics, losses, hardware utilization and monitor it live
- Save model checkpoints, performance charts like ROC curve or confusion matrix
- Analyze and compare the results across multiple experiments and share them with others
PyTorch Ignite metrics, losses, and hardware utilization logged to Neptune automatically.
Experiment tracking tool with open source integration with Pytorch Ignite, to automatically log metrics model checkpoints and more.
NeptuneLogger and attach your trainer and evaluator to it to log metrics and losses.
They will be automatically plotted in the Neptune app.
NeptuneSaver to Ignites’
Checkpoint to keep the best checkpoints saved and backed-up in Neptune as artifacts.
npt_logger = NeptuneLogger( api_token="ANONYMOUS", project_name='shared/pytorch-ignite-integration',...) npt_logger.attach(validation_evaluator, log_handler=OutputHandler(tag="validation", metric_names=["loss", "accuracy"], another_engine=trainer), event_name=Events.EPOCH_COMPLETED) handler = Checkpoint(to_save, NeptuneSaver(npt_logger), …) validation_evaluator.add_event_handler(Events.COMPLETED, handler)
Support for deep learning libraries
- Start instantly with out-of-the-box integration.
- Track rich data (metrics, text, images, files and more).
- Save model weights.
- All charts visible for team members.
- Compare across multiple experiments and gain insight.
- Download charts from UI.
Backed up experiments history
- All visualizations are stored and secured.
- Keep all history – review when needed.
- Secured Intellectual Property (IP).
Register and try it out!