Skip to content
Last Dictionary
Plane
AI Engineering & MLOpsMachine Learning
trademark

MLflow

proper noun/EM-el-flow/Also: MLflow Tracking, MLflow Model Registry, MLflow Models

In plain wordsA free software platform that records machine-learning experiments, packages the resulting models and keeps track of which versions are ready to be used.

  1. An open-source platform that organizes the work surrounding model code — recording experiments, packaging models and controlling their release — whatever training library is used. Its main components are Tracking, which records each run's parameters, metrics and output artifacts through an API and a web interface (experiment trackingexperiment tracking, n.: Keeping an automatic logbook of every model-training attempt — settings, data, code and results — so a team can compare attempts, repeat the good ones and see what changed.); Models, a packaging convention with 'flavors' that lets one saved model be loaded by different tools or served as a REST endpoint; the Model Registry, for versioning models and managing their promotion to production (model registrymodel registry, n.: A central catalog where a team stores every version of its trained models, with records of how each was made and which version is approved for use.); and Projects, a format for packaging reproducible code. Later releases added evaluation utilities, a gateway for routing calls to language-model providers, prompt management, and OpenTelemetry-compatible tracing of generative-AI applications, extending it toward LLMOpsLLMOps, n.: The everyday work of running chatbots and other products built on large language models: tracking prompts, testing answers, watching cost and quality, and updating safely when models change.. MLflow runs locally or on a self-hosted server and is offered as a managed service by Databricks and several cloud providers.

    Log the run to MLflow and register the model if validation AUC beats the current champion.

Origin

Announced by Databricks in June 2018 at its Spark + AI Summit, where Matei Zaharia, the creator of Apache Spark, presented it as an open alternative to the proprietary in-house ML platforms of large technology companies. Databricks contributed it to the Linux Foundation in 2020.

Related

See also
  • experiment trackingexperiment tracking, n.: Keeping an automatic logbook of every model-training attempt — settings, data, code and results — so a team can compare attempts, repeat the good ones and see what changed.
  • model registrymodel registry, n.: A central catalog where a team stores every version of its trained models, with records of how each was made and which version is approved for use.
  • MLOpsMLOps, n.: The working habits and tools that keep machine-learning models dependable inside real products once the experiment is over: automated testing, release, monitoring and retraining; the factory and maintenance crew around a model, not the model itself.
  • Weights & BiasesWeights & Biases, prop. n.: A company and online service that machine-learning teams use to log, chart and compare their training experiments, models and AI-app test results in shared dashboards.
  • LLM observabilityLLM observability, n.: Recording and inspecting what an AI application actually did on each request — prompts sent, documents found, tools used, answers given, time and cost — so problems can be found and fixed.
  • machine learning pipelinemachine learning pipeline, n.: An automated chain of steps that takes raw data, prepares it, trains and tests a model and hands it over for use, so the whole process can be rerun the same way.

Sources

  1. Zaharia, M. et al. (2018). Accelerating the Machine Learning Lifecycle with MLflow. IEEE Data Engineering Bulletin 41(4).

Mentioned in

experiment trackingexperiment tracking, n.: Keeping an automatic logbook of every model-training attempt — settings, data, code and results — so a team can compare attempts, repeat the good ones and see what changed., model registrymodel registry, n.: A central catalog where a team stores every version of its trained models, with records of how each was made and which version is approved for use., Weights & BiasesWeights & Biases, prop. n.: A company and online service that machine-learning teams use to log, chart and compare their training experiments, models and AI-app test results in shared dashboards.