LMRun

Multi-cloud AI environment built on SkyPilot

• Turns multiple cloud accounts into an integrated AI platform
• Provides on-demand GPUs with a seamless DX through SkyPilot

E.g. LMRun connects isolated clusters and MLOps servers in one bootstrap command

LMRun is a distribution: it has already made infrastructure choices to include a cloud setup, networking and services, that run below, across and on top of SkyPilot.

features

• Global VPN network of GPUs for secure availability and cheaper compute
• Open stack: work with any tool or framework on cloud-agnostic infrastructure
• No data lock-in: free or low-cost egress bandwidth for high throughput
• Hyperscaler (AWS) extension to outsource AI compute

Quickstart

1. Set up your local environment
2. Follow instructions in init to initialize the cloud environment
3. Follow instructions in mesh to deploy the cluster mesh
4. Go to workspace for example templates and use cases
5. Make them yours: change models, tune configuration, add resources, etc

Local Environment

Prerequisite: if necessary, install system dependencies. At the very least, you need to configure an AWS profile.

1. Install uv
2. Run uv sync --extra dev in this directory to create the virtual environment and install all dependencies
3. Set the export of variables when activating the environment

echo 'set -a; source "$VIRTUAL_ENV/../.env"; set +a' >> .venv/bin/activate

4. Activate the environment: source .venv/bin/activate

System Dependencies

1. Install Pulumi to manage cloud resources
2. Use AWS CLI V2 to create an lmrun profile with admin permissions: aws configure --profile lmrun. We recommend creating a new AWS account for this profile. If you already have a main account, create the new account from AWS Organizations. To use another profile name, edit AWS_PROFILE in .env.