mlop

Mlop is an open source platform for tracking, optimizing and collaborating on machine learning experiments, backed by Y Combinator. It is aimed at ML engineers and research teams who need a record of how each training run performed and why. Users add a short Python client to their training code, initialize a run and log metrics, parameters, gradients and media such as images. The dashboard shows model accuracy and performance over time, tracks parameter and gradient behavior as a model trains, and records git status and uncommitted files so that experiments stay reproducible. Email alerts and notifications flag problems with model performance. The API is described as fully compatible with Weights & Biases, which is intended to make migration easy. A compute product is in private beta and inference is listed as coming soon. There is a free tier at $0 per month for personal use or small teams, with one seat, unlimited logging hours and 10 GB of storage, and no credit card is required to start. The Pro plan is for businesses scaling with AI, with up to 10 seats, unlimited logging hours, 100 GB of storage and email support, and its price is given on request. The Enterprise plan has custom pricing with unlimited seats and storage, a self hosted option, a security audit, a private Slack channel and 24/7 support from the founders.