Experiments tracking with

Marie-Hélène Burle

November 25, 2025

What is MLflow?

FOSS

Multi-tool platform

draw map of the LLM/data usage

this webinar focuses on experiment tracking

Experiment tracking

allows to:

Combine with DVC

for dataset versioning

here is an example workflow

Limitations

MLflow projects do not (yet) support uv

Experiment tracking usage

Installation

uv not officially supported and might come issues. In case of problem, use official method with pip
Don’t call your script mlflow.py if you install with uv as it causes unexpected behaviours

uv init --bare
uv add mlflow

Tracking models

Overview

Tracking models at checkpoints

Compare with different datasets

Visualization with tracking UI

Collect tracking data

import mlflow

with mlflow.start_run():
    mlflow.log_param("lr", 0.001)
    # Your ml code
    ...
    mlflow.log_metric("val_loss", val_loss)

Organize runs

experiments child runs tags

Visualize tracking data

Can run on multiple platforms or locally

Local tracking data:

uv run mlflow ui --port 5000

Remote tracking data (team development), launch a tracking server:

uv run mlflow server --host 127.0.0.1 --port 5000

In both cases, open http://127.0.0.1:5000 in your browser

You can choose any unused port

MLflow tracking setups