Overview

Relevant source files

Purpose and Scope

This document provides a high-level introduction to MLflow's architecture, core components, and codebase organization. It describes the platform's purpose as a complete ML lifecycle management system and maps high-level concepts to specific code entities (classes, modules, files) to help developers navigate the repository.

For detailed information about specific subsystems:

System Architecture: Overall design and subsystem interactions. See System Architecture.
Package Distributions: Differences between mlflow, mlflow-skinny, and mlflow-tracing. See Package Distributions.
Tracking System: Logging experiments, runs, parameters, metrics, and artifacts. See Tracking System.
ML Framework Integration: Version management and autologging for 40+ libraries. See ML Framework Integration.
PyFunc Model System: The universal model interface and deployment format. See PyFunc Model System.
Model Evaluation: Assessing model performance and GenAI quality. See Model Evaluation.
Tracing and Observability: LLM execution flows and OpenTelemetry integration. See Tracing and Observability.

Sources: README.md8-15 pyproject.toml11-15

What is MLflow?

MLflow is an open-source platform for managing the complete machine learning lifecycle, covering both classical ML and GenAI/LLM applications README.md6-11 It provides:

Tracking: Log and query experiments, parameters, metrics, and artifacts via the fluent API mlflow/tracking/fluent.py1-4 or MlflowClient mlflow/tracking/client.py216
Models: Deploy models from various ML libraries in diverse serving environments using the pyfunc abstraction mlflow/__init__.py94
Model Registry: Centralized model store with versioning and alias management mlflow/tracking/client.py119
Prompt Registry: Version and manage prompt templates for GenAI applications via Prompt and PromptVersion entities mlflow/tracking/client.py50-52 mlflow/tracking/client.py79-88
Tracing: Observability for LLM/agentic applications with OpenTelemetry-based span tracking mlflow/version.py24 mlflow/__init__.py184-199
Evaluation: Assess performance using mlflow.evaluate() and GenAI-specific judges mlflow/ml-package-versions.yml100-108
AI Gateway: Unified API for managing access to LLM providers like OpenAI and Anthropic pyproject.toml96-103 pyproject.toml140-141

The platform is implemented primarily in Python with support for TypeScript/JavaScript, Java, and R README.md80 mlflow/java/pom.xml1-11 mlflow/R/mlflow/DESCRIPTION1-4

Sources: README.md6-15 pyproject.toml11-15 mlflow/java/pom.xml1-11 mlflow/R/mlflow/DESCRIPTION1-4 mlflow/version.py5-30 mlflow/tracking/client.py50-52

Package Architecture

MLflow is distributed as three distinct Python packages with shared infrastructure. The build is managed via dev/pyproject.py which auto-generates the distribution metadata pyproject.toml1

Package Component Mapping

Title: Package Component Mapping

Package Definitions:

Package	Purpose	Key Characteristics
`mlflow`	Full platform with all features	Includes ML frameworks, UI, and serving dependencies pyproject.toml11-68
`mlflow-skinny`	Lightweight core tracking	Excludes SQL storage, server, UI, or heavy data science libs libs/skinny/pyproject.toml9-48
`mlflow-tracing`	Tracing SDK for LLM observability	Focused on OpenTelemetry conventions and span capture libs/tracing/pyproject.toml1-10

For details on how these packages are constructed and the flags IS_TRACING_SDK_ONLY, IS_MLFLOW_SKINNY, or IS_FULL_MLFLOW, see Package Distributions.

Sources: mlflow/version.py20-30 pyproject.toml1-68 libs/skinny/pyproject.toml1-48 pyproject.release.toml29-31

System Architecture Overview

MLflow follows a layered client-server architecture where multiple client types communicate with backend services via REST API.

Title: Client-Server Code Entity Interaction

Key Code Entities:

Client Entry Point: MlflowClient class provides a lower-level CRUD interface for runs, models, and prompts mlflow/tracking/client.py216
Fluent API: High-level interface in mlflow.tracking.fluent (e.g., start_run, log_param, set_experiment) mlflow/tracking/fluent.py1-10
Environment Flags: Global state flags like IS_TRACING_SDK_ONLY determine module availability at runtime mlflow/version.py24
Dependency Tracking: Use of uv for exact dependency resolution and lockfile management uv.lock1-35 pyproject.toml240-255

For details on subsystem interactions and design patterns, see System Architecture.

Sources: mlflow/tracking/client.py216-220 mlflow/version.py20-30 uv.lock1-35 pyproject.toml29-68 pyproject.toml128

ML Framework Integrations

MLflow integrates with dozens of ML frameworks (Scikit-Learn, PyTorch, Transformers, etc.) through a unified "flavor" system. These integrations are managed with specific version constraints defined in ml-package-versions.yml mlflow/ml-package-versions.yml1-26

Framework Version Management

Title: Version Configuration to Code Flavor Mapping

Key Integration Components:

Version Management: mlflow/ml-package-versions.yml defines the minimum and maximum supported versions for each flavor, as well as test commands mlflow/ml-package-versions.yml1-10
Autologging: Automatically captures metrics, parameters, and models during training via autologging hooks defined in the configuration mlflow/ml-package-versions.yml14-20
Cross-Version Testing: CI matrices and supported version ranges are generated based on the version configuration mlflow/ml-package-versions.yml11-20 mlflow/ml_package_versions.py4-25

Sources: mlflow/ml-package-versions.yml1-210 mlflow/ml_package_versions.py1-210 mlflow/version.py24-30

Child Pages

System Architecture — Detailed view of subsystem interactions and internal design patterns.
Package Distributions — Technical breakdown of mlflow, mlflow-skinny, and mlflow-tracing builds.

Sources: mlflow/version.py20-30 pyproject.release.toml30-31

Overview

Relevant source files

Purpose and Scope

For detailed information about specific subsystems:

System Architecture: Overall design and subsystem interactions. See System Architecture.
Package Distributions: Differences between mlflow, mlflow-skinny, and mlflow-tracing. See Package Distributions.
Tracking System: Logging experiments, runs, parameters, metrics, and artifacts. See Tracking System.
ML Framework Integration: Version management and autologging for 40+ libraries. See ML Framework Integration.
PyFunc Model System: The universal model interface and deployment format. See PyFunc Model System.
Model Evaluation: Assessing model performance and GenAI quality. See Model Evaluation.
Tracing and Observability: LLM execution flows and OpenTelemetry integration. See Tracing and Observability.

Sources: README.md8-15 pyproject.toml11-15

What is MLflow?

MLflow is an open-source platform for managing the complete machine learning lifecycle, covering both classical ML and GenAI/LLM applications README.md6-11 It provides:

Tracking: Log and query experiments, parameters, metrics, and artifacts via the fluent API mlflow/tracking/fluent.py1-4 or MlflowClient mlflow/tracking/client.py216
Models: Deploy models from various ML libraries in diverse serving environments using the pyfunc abstraction mlflow/__init__.py94
Model Registry: Centralized model store with versioning and alias management mlflow/tracking/client.py119
Prompt Registry: Version and manage prompt templates for GenAI applications via Prompt and PromptVersion entities mlflow/tracking/client.py50-52 mlflow/tracking/client.py79-88
Tracing: Observability for LLM/agentic applications with OpenTelemetry-based span tracking mlflow/version.py24 mlflow/__init__.py184-199
Evaluation: Assess performance using mlflow.evaluate() and GenAI-specific judges mlflow/ml-package-versions.yml100-108
AI Gateway: Unified API for managing access to LLM providers like OpenAI and Anthropic pyproject.toml96-103 pyproject.toml140-141

The platform is implemented primarily in Python with support for TypeScript/JavaScript, Java, and R README.md80 mlflow/java/pom.xml1-11 mlflow/R/mlflow/DESCRIPTION1-4

Sources: README.md6-15 pyproject.toml11-15 mlflow/java/pom.xml1-11 mlflow/R/mlflow/DESCRIPTION1-4 mlflow/version.py5-30 mlflow/tracking/client.py50-52

Package Architecture

MLflow is distributed as three distinct Python packages with shared infrastructure. The build is managed via dev/pyproject.py which auto-generates the distribution metadata pyproject.toml1

Package Component Mapping

Title: Package Component Mapping

Package Definitions:

Package	Purpose	Key Characteristics
`mlflow`	Full platform with all features	Includes ML frameworks, UI, and serving dependencies pyproject.toml11-68
`mlflow-skinny`	Lightweight core tracking	Excludes SQL storage, server, UI, or heavy data science libs libs/skinny/pyproject.toml9-48
`mlflow-tracing`	Tracing SDK for LLM observability	Focused on OpenTelemetry conventions and span capture libs/tracing/pyproject.toml1-10

For details on how these packages are constructed and the flags IS_TRACING_SDK_ONLY, IS_MLFLOW_SKINNY, or IS_FULL_MLFLOW, see Package Distributions.

Sources: mlflow/version.py20-30 pyproject.toml1-68 libs/skinny/pyproject.toml1-48 pyproject.release.toml29-31

System Architecture Overview

MLflow follows a layered client-server architecture where multiple client types communicate with backend services via REST API.

Title: Client-Server Code Entity Interaction

Key Code Entities:

Client Entry Point: MlflowClient class provides a lower-level CRUD interface for runs, models, and prompts mlflow/tracking/client.py216
Fluent API: High-level interface in mlflow.tracking.fluent (e.g., start_run, log_param, set_experiment) mlflow/tracking/fluent.py1-10
Environment Flags: Global state flags like IS_TRACING_SDK_ONLY determine module availability at runtime mlflow/version.py24
Dependency Tracking: Use of uv for exact dependency resolution and lockfile management uv.lock1-35 pyproject.toml240-255

For details on subsystem interactions and design patterns, see System Architecture.

Sources: mlflow/tracking/client.py216-220 mlflow/version.py20-30 uv.lock1-35 pyproject.toml29-68 pyproject.toml128

ML Framework Integrations

Framework Version Management

Title: Version Configuration to Code Flavor Mapping

Key Integration Components:

Version Management: mlflow/ml-package-versions.yml defines the minimum and maximum supported versions for each flavor, as well as test commands mlflow/ml-package-versions.yml1-10
Autologging: Automatically captures metrics, parameters, and models during training via autologging hooks defined in the configuration mlflow/ml-package-versions.yml14-20
Cross-Version Testing: CI matrices and supported version ranges are generated based on the version configuration mlflow/ml-package-versions.yml11-20 mlflow/ml_package_versions.py4-25

Sources: mlflow/ml-package-versions.yml1-210 mlflow/ml_package_versions.py1-210 mlflow/version.py24-30

Child Pages

System Architecture — Detailed view of subsystem interactions and internal design patterns.
Package Distributions — Technical breakdown of mlflow, mlflow-skinny, and mlflow-tracing builds.

Sources: mlflow/version.py20-30 pyproject.release.toml30-31

Overview

Purpose and Scope

What is MLflow?

Package Architecture

Package Component Mapping

System Architecture Overview

ML Framework Integrations

Framework Version Management

Child Pages

On this page

Overview

Purpose and Scope

What is MLflow?

Package Architecture

Package Component Mapping

System Architecture Overview

ML Framework Integrations

Framework Version Management

Child Pages

On this page