mlop

What mlop is

mlop is an open-source experiment tracking platform for machine learning teams. It tracks and compares experiments, model versions, hyperparameters, metrics, artifacts, and gradients. The documentation describes real-time training monitoring, interactive performance dashboards, and experiment comparison views. mlop can alert users when training performance is poor and provide suggestions, including ways to address exploding gradients. It also tracks experiment versions and uncommitted files in real time, which the maker presents as support for reproducibility. The site says mlop is compatible with the Weights & Biases API, and its documentation offers migration support, including instructions for PyTorch Lightning and Hugging Face Transformers. A Python SDK is installed with `pip install "mlop[full]"`. The Free plan costs nothing and includes one seat, unlimited logging hours, and 10 GB of storage. Pro includes up to 10 seats and 100 GB of storage; Enterprise has unlimited seats and storage, a self-hosted option, a security audit, and 24/7 founder support. The platform is available on web and self-hosted deployments; Compute is labeled Private Beta and Inference Coming Soon.

Who it suits

mlop is aimed at people training machine learning models, with its Free tier described for individual developers and small teams. It may suit teams that need experiment comparisons, tracking for training artifacts, and compatibility with the Weights & Biases API.

Limits

The Free plan includes one seat and 10 GB of storage, with unlimited logging hours. Pro is capped at 10 seats and 100 GB of storage; Enterprise lists unlimited seats and storage. Compute is Private Beta and Inference is Coming Soon.

Getting started

1. Open the mlop website.
2. Install the Python SDK with `pip install "mlop[full]"`.
3. Start with the Free plan, which includes one seat, unlimited logging hours, and 10 GB of storage.
4. Use the documentation's migration instructions if moving from PyTorch Lightning or Hugging Face Transformers.
5. Contact mlop about Pro or Enterprise pricing and options.

Verdict

Choose mlop if you want experiment tracking with real-time monitoring and a Free plan for one seat. Teams needing more seats, storage, a self-hosted option, or Enterprise support can review its paid tiers, but their prices are available only by discussion.