Quick Start Guide

This guide walks you through installing the MLPipeline CLI, creating a project, defining a pipeline, running it in the cloud, and monitoring the results.

1. Install the CLI

Install mlpipeline-cli using pip. Python 3.9 or later is required.

$ pip install mlpipeline-cli

# Verify installation
$ mlpipeline --version
mlpipeline-cli 1.8.3

2. Authenticate

$ mlpipeline auth login
Opening browser for authentication...
Authenticated as user@company.com (org: acme-ml)

3. Initialize a Project

Create a new project directory with the standard structure. This generates a pipeline.yaml template and a .mlpipeline/ configuration directory.

$ mlpipeline init my-first-pipeline
Created project: my-first-pipeline/
  pipeline.yaml
  .mlpipeline/config.yaml
  stages/
  data/

$ cd my-first-pipeline

4. Define Your Pipeline

Open pipeline.yaml and define your stages. Each stage specifies an image, compute requirements, and dependencies. Here is a minimal example with two stages:

pipeline.yaml

name: my-first-pipeline
version: "1.0"

stages:
  - name: preprocess
    image: mlpipeline/python:3.11
    script: stages/preprocess.py
    compute: cpu-2x8
    outputs:
      - data/processed/

  - name: train
    depends_on: preprocess
    image: mlpipeline/pytorch:2.3
    script: stages/train.py
    compute: gpu-a100-1x
    params:
      epochs: 20
      batch_size: 64
    metrics:
      - loss
      - accuracy

5. Run the Pipeline

Submit the pipeline to MLPipeline Cloud. The CLI validates your configuration, uploads stage scripts, and starts execution in your selected region.

$ mlpipeline run --region eu-west-1
Validating pipeline.yaml... OK
Uploading stages (2 files, 4.2 KB)... done
Pipeline submitted: run-2026-04-11-001
Dashboard: https://app.mlpipeline-cloud.com/runs/run-2026-04-11-001

6. Monitor Execution

Watch pipeline progress in real time from the terminal or check the web dashboard for detailed metrics and logs.

$ mlpipeline logs --follow run-2026-04-11-001
[preprocess] Starting on cpu-2x8 (eu-west-1)...
[preprocess] Loaded 142,583 records from data/raw/
[preprocess] Output: data/processed/ (23.4 MB)
[preprocess] Completed in 1m 42s
[train] Starting on gpu-a100-1x (eu-west-1)...
[train] Epoch 1/20 — loss: 0.684, accuracy: 0.712
[train] Epoch 2/20 — loss: 0.521, accuracy: 0.803
...

Next Steps

Now that your first pipeline is running, explore these topics:

Add more stages and configure parallel execution with depends_on
Register trained models in the Model Registry for versioning and promotion
Set up scheduled runs with cron expressions in your pipeline definition
Deploy models as REST endpoints using mlpipeline deploy
Configure alerting for metric thresholds and pipeline failures

Getting Started