Integrations

Neptune integrates with various machine learning libraries.

Most integrations are implemented as loggers or callbacks that you pass along in your model training code.

Example

To log metadata when using Keras, add a Neptune callback to model.fit():

import neptune
from neptune.integrations.tensorflow_keras import NeptuneCallback

neptune_run = neptune.init_run()

model.fit(
    x_train,
    y_train,
    ...,
    callbacks=[NeptuneCallback(run=neptune_run)],
)

In the above example, Neptune automatically logs metadata that is typically generated while training models with Keras.

You can use both the Neptune client library and Neptune integrations in combination. This way, you can benefit both from the integrated logging and customized metadata tracking.

You can also use multiple integrations together in the same script. For a tutorial, see Using multiple integrations at once.

Integrations by category

TrainingAutomationExperiment trackingHPOMonitoringDebugging

• Catalyst
• CatBoost
• Detectron2
• fastai
• Transformers
• Keras
• LightGBM
• Lightning
• MosaicML Composer
• Prophet
• PyTorch
• PyTorch Ignite
• Sacred
• scikit-learn
• skorch
• TensorFlow
• XGBoost

• Airflow
• Azure ML
• Docker
• GitHub Actions
• Kedro
• SageMaker
• ZenML

• MLflow
• Sacred
• TensorBoard

• Optuna

• Arize
• Evidently

• DALEX
• Great Expectations

CI/CD

Neptune integrates with the following continuous integration and delivery tools:

• Docker
• GitHub Actions

See also:

• Service accounts
• Models in Neptune

Migration tools

• Copy runs from one project to another 
• Copy model metadata from the model registry to experiments 

You can also import your metadata from other frameworks.

• MLflow
• TensorBoard
• Weights and Biases

Other supported tools

While some libraries and tools may require more manual setup than others, Neptune generally supports anything you can do in Python.

We provide examples and guides for the following:

IDEsLibraries Community-developed

• Amazon SageMaker
• Deepnote
• Google Colab

• Altair
• argparse
• Bokeh
• Leaflet (Folium)
• Matplotlib
• pandas
• Plotly
• Seaborn

• Flyte
• H2O Hydrogen Torch
• H2O LLM Studio
• MMEngine
• Modelbit
• Open Metric Learning
• Slideflow
• RecList
• Ultralytics YOLO

My library is not here. What now?

1. Use the Neptune client library (neptune ). If you get stuck with setup or usage, we're here to help.
2. Contact us via mail (contact@neptune.ai) to discuss what you need and how we can deliver it.

Tip

To see what's planned or already in development, have a look at the Neptune product roadmap .