How to choose a template for your Hostinger GPU instance

Hostinger GPU instance template options define the pre-installed operating system and AI application environment available during server deployment. Your choice depends on the workflow you’re running, how many CPU cores your instance has, and how much default security each template provides.

Before you start

• Ensure your selected server configuration includes at least 2 CPU cores to access pre-configured application templates.

Understand Hostinger GPU instance templates

A template is the operating system and application image that deploys automatically on your Hostinger GPU instance. Every Hostinger GPU instance runs Ubuntu 24.04 as its base operating system. Certain templates include pre-installed software stacks so you can start a specific AI workflow immediately instead of installing and configuring it yourself.

Check template availability by CPU cores

Available templates depend strictly on your selected CPU core count, not your GPU model.

Choose a template based on your workflow

Ubuntu 24.04

Choose Ubuntu 24.04 if you want to install and configure your own software from scratch rather than use a pre-built stack. It provides a bare operating system with no pre-installed applications or exposed web services, giving you total manual control.

Ollama

Choose Ollama if you want to run a local language model and interact with it through a browser without setting anything up yourself. The template includes Ollama for executing local language models and Open WebUI for browser-based interaction. The internal Ollama Application Programming Interface (API) stays private within the instance, while Open WebUI exposes automatically so you can access it in a browser immediately after deployment.

JupyterLab

Choose JupyterLab if you want a ready-to-use development environment for experimenting with models or running training code. The template provides a web-based notebook environment pre-configured with PyTorch and CUDA 12.4 drivers, and it exposes automatically upon deployment.

Unsloth

Choose Unsloth if your workflow is fine-tuning Large Language Models (LLMs) rather than training from scratch. The template includes Unsloth Studio, pre-configured for fine-tuning using Low-Rank Adaptation (LoRA) and Quantized Low-Rank Adaptation (QLoRA), and it exposes automatically with pre-set credentials.

InvokeAI

Choose InvokeAI if your workflow is image generation rather than language model work. The template delivers a web-based studio supporting Stable Diffusion and FLUX models. It deploys without any pre-installed models, so you download the models you need through the internal Model Manager interface after deployment.

Compare security settings across templates

Each template also comes with a different default authentication level, which affects what you need to secure before exposing it publicly.

Frequently asked questions

Why are application templates missing during deployment?

Select a larger instance size with at least 2 CPU cores if application templates do not appear during deployment.

Can you switch templates after deploying?

Deploy a new Hostinger GPU instance to switch templates, as templates cannot be modified after initial deployment. Destroying an existing Hostinger GPU instance permanently removes all stored data.

Can you install additional software on application templates?

Install additional software packages on any application template because every Hostinger GPU instance retains full Ubuntu 24.04 root access.

Choosing the right Hostinger GPU instance template ensures your server deploys with the exact operating system, application stack, and security configuration required for your project.

NOTES

• Select an instance size with at least 2 CPU cores before picking a GPU model if you require application templates.
• Create your account immediately after server deployment when using Open WebUI because the service features no authentication until the first user registers.
• Keep the JupyterLab access token requirement enabled because an unauthenticated environment allows remote code execution.
• Select a GPU model with 48 GB of Video Random Access Memory (VRAM) or more for full fine-tuning runs, as smaller instances only support QLoRA fine-tuning.
• Export fine-tuned models to external storage like Hugging Face before destroying an Unsloth instance because Unsloth Studio does not persist data automatically.
• Configure custom access control before sharing the server address when using InvokeAI because the service includes no default authentication.
• Download generated images and custom models to external storage before destroying an InvokeAI instance because local data does not persist automatically.