COMET-ML

Comet ML experiment tracking + Opik LLM observability.

Définition

Comet ML products : (1) Comet ML Experiments : ML training runs tracking similar W&B (hyperparameters + metrics + system metrics + artifacts), comet_ml.Experiment Python SDK, integration TensorFlow + PyTorch + scikit-learn + XGBoost + LightGBM + Keras + HuggingFace Transformers, etc. (2) Comet ML Model Registry : model versions + stages + lifecycle management. (3) Comet ML Workspaces : team + project collaboration. (4) Comet Artifacts : versioned datasets + models + arbitrary files lineage. (5) Comet Reports : embed Markdown + experiment plots collaborative reports. (6) Comet Production : Predictions monitoring + drift detection + alerting deployed models. (7) Opik (Open-source MIT, launched 2024) : LLM observability + evaluation platform, traces LLM calls + evaluations + datasets, similar Phoenix Arize + LangSmith + Weave competitor, deepens LLM workflow support Comet ecosystem. Customers : Etsy + Booking.com + Citi + Affirm + Uber + Boeing. Funding : Series A $13M 2020 + Series B $50M 2021 (Yardi Capital lead).

Origine

Comet ML fondee 2017 a New York par Gideon Mendels (ex-Google) + Nimrod Lahav + Dor Moshe ; Series A $13M 2020 + Series B $50M 2021 ; Opik launched 2024 OSS MIT.

Exemple en contexte

Booking.com data science teams use Comet ML to track ~10000+ ML experiments per year : recommendation models + pricing optimization + search ranking + ETA prediction, Comet integration PyTorch + scikit-learn + LightGBM, model registry production deployment, Opik LLM observability emerging adoption customer service AI agents.

Termes liés

• W&B — main competitor.