MLFlowClient.jl

MLFlowClient.MLFlowClient — Module

MLFlowClient

MLFlowClient is a Julia package for working with MLFlow using the REST API v2.0.

MLFlowClient allows you to create and manage MLFlow experiments, runs, and log parameters, metrics, and artifacts. If you are not familiar with MLFlow and its concepts, please refer to MLFlow documentation.

source

A Julia client for the MLflow REST API. Track experiments, log metrics and parameters, manage models, and more — directly from Julia.

Tested against MLflow 3.11.1.

Installation

using Pkg
Pkg.add("MLFlowClient")

Quick start

using MLFlowClient

# Connect to an MLflow server
mlf = MLFlow("http://localhost:5000")

# Create an experiment and a run
experiment_id = createexperiment(mlf, "my-experiment")
run = createrun(mlf, experiment_id)

# Log parameters, metrics, and tags
logparam(mlf, run, "learning_rate", "0.01")
logmetric(mlf, run, "accuracy", 0.95)
setruntag(mlf, run, "model_type", "linear")

# Complete the run
updaterun(mlf, run; status=RunStatus.FINISHED)

API coverage

MLFlowClient implements the full MLflow REST API (v2.0 and v3.0) and the Authentication REST API:

Experiments — create, get, search, update, delete, restore, tags
Runs — create, get, search, update, delete, restore, tags
Logging — metrics, parameters, batch, model, inputs
Artifacts — list, upload, download, delete, multipart upload, presigned URLs
Registered models — create, get, search, rename, update, delete, tags, aliases
Model versions — create, get, search, update, delete, transition stage, tags
Scorers — register, list, get, delete (v3.0)
Gateway — secrets, model definitions, endpoints, bindings, tags, budgets (v3.0)
Prompt optimization — create, get, search, cancel, delete jobs (v3.0)
Webhooks — create, get, list, update, delete, test
Users & permissions — create, get, update, delete users; experiment and model permissions

Authentication

# Basic auth
mlf = MLFlow("http://localhost:5000"; username="admin", password="password")

# Token-based auth (e.g., Databricks)
mlf = MLFlow("https://my-server.cloud.databricks.com";
    headers=Dict("Authorization" => "Bearer <token>"))

The environment variables MLFLOW_TRACKING_URI, MLFLOW_TRACKING_USERNAME, and MLFLOW_TRACKING_PASSWORD are respected when set, and will override the corresponding constructor arguments.

Next steps

Head to the Tutorial for a walkthrough of the common tracking workflow, or browse the API Reference for the full list of types and operations.