ClearML

ClearML

Part of a broader platform

Open-source with closed-source paid features

Ability to be deployed on-prem and all clouds either self hosted with OSS or managed with Enterprise

Standard SLA that can be customized for Enterprise customers

Open-source: community
Enterprise: customizable 24X7 SLA based on needs

ClearML Server supports self user management.
Enterprise offers full user management with SSO and user roles and permissions

SSO and LDAP in Enterprise as well as user management and roles and resource-based permissions

Server:
OSS: supports docker based installation, takes 10-30 minutes to set up as well as run on a Kubernetes Cluster Free, Pro & Enterprise: Fully Managed
SDK:
2 lines of code integrate to user's codebase. Same integration for all tiers.

Any type of artifact can be saved (Files, folders and python objects), most artifacts have previews.
Multiple frameworks models are automatically captured.
Metadata can be accessed from python code or REST API.
Supports storage on various mediums (NFS, Azure blob, S3, etc...)

Model artifacts are stored alongside it's inputs (code, params, data, metrics).
Supports full reproducibility of models.
Supports tagging models and publishing them.
ClearML Agent also supports reproducing experiment including environment (almost 0 setup needed).

Experiment information is automatically logged:
Git info, docker container info, parameters (using argparse, click & Hydra).
Model checkpoints from multiple frameworks such as pytorch, TF, sklearn are automatically logged.
Everything reported to Tensorboard or matplotlib is automatically captured.
Users can manually report scalars histograms, plots and custom plotly objects as well as images, audio files and console output.
Machine monitoring (GPU, CPU, Mem, Network)
Live tracking is supported.

Full experiment comparison including code, docker, parameters, output metrics.
Parallel coordinates for Hyperparameter optimization.

Organized into projects and sub-projects.
Further management with status and tags is available.
Has API for retrieving tasks according to filters.

Models and experiments can be reviewed and tracked with built-in markdown editor.
Supports sharing views between colleagues.
Supports locking models for production.

Supports continuous integration with triggers and scheduling mechanisms.
Job run via ClearML Agent as the orchestrator.
Pipeline mechanism also available for complex processes.

Integrates with a host of tools.
3rd party integrations (Slurm, Apache Airflow, and more) are also available.