Introduction¶

Mantik is a web-based platform for hosting and maintaining your Machine Learning (ML) projects. It allows experiment tracking with MLflow, and training and inference on high-performance computing (HPC) systems.

Getting Started¶

To get started with the Mantik Platform and its main features, have a look at our quickstart guide.

Mantik projects form the basis of the platform, which are introduced on the Mantik projects page.

Since the Mantik Platform is the main access point to Mantik, we recommend new users to familiarize themselves with it first. Some features (i.e. data handling), though, currently require usage of the Mantik Python Client and CLI for which we also have a quickstart guide. For more detailed information on the CLI, see Command-Line Interface.

Further API documentations:

Python SDK

Experiment Tracking¶

The Mantik platform offers the full capabilities of MLflow. Tracking results are stored and accessible remotely so that you can work from any machine. The tracking server is secured by username and password (see Credentials and Environment Variables).

For a detailed introduction on MLflow with Mantik see here.

Remote Execution on HPC¶

Mantik can submit runs to HPC and automatically track ML tasks on the platform. It supports any HPC resource accessible via UNICORE or firecREST (e.g. JSC or CSCS, see Supported HPC Sites). The client can be called either via the web-platform, Command-Line Interface or Python bindings.

For a tutorial on how to run applications on HPC, see Mantik on HPC.

Using Stored Models for Inference¶

MLflow enables training and storing models for later use in inference. To get started see the “Saving and loading models with Mlflow” introduction.

Bug Reports and Feature Requests¶

For bug reports and feature requests, you’re welcome to write an email to our helpdesk.

FAQ, Common Issues, and Helpdesk¶

For a collection of frequently-asked questions (FAQs) and common issues, see FAQ and Common Issues, for how to contact the helpdesk see here.

Acknowledgements

The MyST markdown language and MyST parser are both supported by the open community, The Executable Book Project.