Launchables

Launchables are an easy way to bundle a hardware and software environment into an easily shareable link, allowing for faster onboarding and consistency across your GPU-accelerated projects.

Creating a Launchable

1. Create an account on brev.nvidia.com.

2. On the top navigation bar, click the Launchables tab.

3. Click Create Launchable.

Step 1: Files and Runtime

Specify how to add the Code Files to the Launchable. There are three options:

• Git Repository: Select this option if you have code stored in a repository. Only public repositories are currently supported.

  • Enter the repository URL or link to the notebook.

  • Select a Runtime: VM Mode (recommended) or Custom Container

  • Click Next.

• Embedded in Container: Select this option if you have code embedded in a container.

  • No action is required. Click Next.

• No Code: Select this option if you just want a GPU sandbox environment without any guiding code.

  • Select a Runtime: VM Mode (recommended) or Custom Container

  • Click Next.

Best Practices

VM Mode provides a GPU VM with Ubuntu 22.04 OS, Docker, Python, and CUDA pre-installed. You can install additional dependencies in a terminal/notebook session.

Select VM Mode if you have a container that requires a private registry or API keys to pull the image.

Step 2: Configure the Runtime

Configure the selected runtime environment.

• If you had selected VM Mode in Step 1:

  • (Optional) Upload or paste a setup script to run on the VM to customize the environment.

  • Click Next.

• If you had selected Custom Container in Step 1:

  • Select a Featured Container or build/upload your own Docker Compose.

  • If using Docker Compose, validate your compose configuration.

  • Click Next.

Best Practices

Featured Containers are pre-developed, commonly used Docker images for simple setup and demos using a jupyter notebook experience.

Use Docker Compose for complex, multi-container workflows.

Step 3: Jupyter and Networking

1. You may be asked to configure your Jupyter Notebook Experience. Select Yes or No.

   • Yes: This installs Jupyter on the host and provides the user with a one-click button to access the code files via Jupyter.

   • No: This provides the user with a button to access the GPU instance via SSH using the Brev CLI.

2. You may be asked to pre-expose tunnels or services to internet traffic. You can add a secure link, or specify and assign permissions to TCP/UDP Ports.

Step 4: Compute Configuration

1. Select an NVIDIA GPU to run your use case.

2. Review the configuration, adjust Disk Storage if desired.

3. Click Next.

Best Practices

Select the most reasonable minimum-cost configuration for your needs.

You can filter NVIDIA GPUs by VRAM, Cloud Provider, Compute Attributes, and Price.

Step 5: Final Review

1. Enter a Name for your Launchable

2. Review the Launchable configuration you have built. You can Preview the deploy page.

3. Click Create Launchable.

   • This generates a shareable link of the Launchable.

   • You can also view the final Deploy Page for the Launchable.

Congratulations! You have successfully created an NVIDIA Brev Launchable. To deploy this Launchable, open the Deploy Page, select Deploy Launchable > Go to Instance Page.

Next Steps

All Launchables will live under the Launchables tab on the top navigation bar. From here, you can:

• View Launchable Metrics: See the number of views and deployments for each Launchable.

• Toggle Launchable Access: Control for who can view and deploy each Launchable

• Delete Launchable: Permanently remove any Launchable

• Access Launchable Page: Preview code and deploy each Launchable

• Copy Markdown Badge: Embed a deep link to the Launchable in any markdown file

• Show Configuration: Quickly glance at the hardware and software details for each Launchable