---
title: Mlflow Server
emoji: 🏃
colorFrom: gray
colorTo: pink
sdk: docker
pinned: false
license: apache-2.0
short_description: A sample MLFlow server for demo
---

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

# 🚀 MLflow Tracking Server Configuration

This Space hosts a remote MLflow Tracking Server. It allows you to log parameters, metrics, and models from your training pipelines (like GitHub Actions) to a centralized location.

## 🛠 Environment Variables (Secrets)

To make this server functional, you need to go to **Settings > Variables and secrets** in your Hugging Face Space and add the following secrets.

| Variable Name | Description | Example Value |
| :--- | :--- | :--- |
| `BACKEND_STORE_URI` | The database URI where metrics and parameters are stored. | `postgresql://user:password@host:port/db` or `sqlite:///mlflow.db` |
| `ARTIFACT_STORE_URI` | The remote storage location for model artifacts (S3, GCS, etc.). | `s3://my-mlflow-bucket/artifacts` |
| `AWS_ACCESS_KEY_ID` | Required if your artifact store is on S3. | `AKIA...` |
| `AWS_SECRET_ACCESS_KEY` | Required if your artifact store is on S3. | `wJalr...` |
| `PORT` | The port the application listens on (HF Spaces defaults to 7860). | `7860` |

> **Note:** The `AWS_` credentials are required because the Dockerfile installs the AWS CLI to handle interactions with S3 buckets.

## 🐳 Dockerfile Overview

The `Dockerfile` is built on top of `continuumio/miniconda3` to ensure a robust Python environment. Here is what happens during the build:

1.  **Tool Installation:** Installs utility tools (`nano`, `unzip`, `curl`) and the **AWS CLI** (via the official installer) to allow MLflow to communicate with S3 buckets.
2.  **Dependencies:** Installs Python packages listed in `requirements.txt` (typically `mlflow`, `boto3`, `psycopg2`, etc.).
3.  **Startup Command:** The container launches the MLflow server with specific flags to ensure it works in a cloud environment.

### Understanding the Launch Parameters

The `CMD` in the Dockerfile uses the following flags:

* `--host 0.0.0.0`: Binds the server to all network interfaces so it is accessible from outside the container.
* `--serve-artifacts`: Enables the MLflow server to act as a proxy for artifact downloads/uploads (useful if the client doesn't have direct S3 access).
* `--allowed-hosts '*'`: Disables the Host Header validation check. This is necessary in cloud environments (like HF Spaces) where the internal IP and public DNS might mismatch.
* `--cors-allowed-origins '*'`: Sets the Cross-Origin Resource Sharing policy to allow any domain to access this API. This is useful for teaching environments but can be restricted for production.