Getting Started

Welcome to Databricks Mosaic AI Training! With just a few simple steps, get started training and deploying your models.

First, install MCLI, the command line interface to Mosaic AI Training, via pip into your python3 environment:

pip install --upgrade mosaicml-cli

Register your API key by following the instructions in this command:

mcli init

To manually manage your account and create API keys, visit the Mosaic AI console. This API key can be reset in mcli by using:

mcli set api-key <value>

You can also skip this step and use the environment variable MOSAICML_API_KEY to automatically configure access.

Run “Hello World” 🌎

To submit your first run, copy the below yaml into a file called ‘hello_world.yaml’:

name: hello-world
gpu_num: 0
image: bash
command: |
  sleep 2
  echo Hello World!

Then, run:

mcli run -f hello_world.yaml --follow

If you see “Hello World!”, congratulations on setting up MCLI! Follow the documentation for next steps:

• Learn how to configure and customize training runs

• Add integrations with github, docker, cloud storage, and experiment trackers

• Train your first 1 billion parameter GPT model

• Explore what happens at each phase of the run lifecycle