MLFlowClient.jl

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)

What's covered

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

| Area | Operations | |---|---| | 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
mlf = MLFlow("http://localhost:5000"; headers=Dict("Authorization" => "Bearer <token>"))

Environment variables MLFLOWTRACKINGURI, MLFLOWTRACKINGUSERNAME, and MLFLOWTRACKINGPASSWORD are respected when set.

Documentation

See the full documentation for the complete API reference and tutorial.