ToriLinux - Linux LiveCD for offline AI training and inference

LiveCD distribution based on ArchLinux and currently includes the following projects preinstalled, along with their dependencies:

• automatic
• ComfyUI
• llama.cpp
• SillyTavern
• SillyTavern-Extras
• text-generation-webui
• vllm

If you would like to see another AI-related project included in ToriLinux, please open an issue.

Features

• Easy setup: just boot the ISO, and you will have a working setup for training and/or inferencing Large Language Models/Stable Diffusion/etc.
• Fully offline training and/or inference.
• Includes performance state switcher, which reduces GPU temperatures when inference is not running (only on NVIDIA, automatic, llama.cpp, vllm supported).

Usage

To use ToriLinux:

1. Install Ventoy on a USB drive.
2. Download the latest ISO from workflows and copy it to the USB drive.
3. Boot from the USB drive (select it as the boot device in BIOS/UEFI).
4. Log in with the username tori and password tori. You can also use SSH.

The following options are currently available:

1. Headless: Headless variant without GUI, for servers.
2. With GUI: A variant with a GUI, for desktops.
3. Empty: A variant WITHOUT projects, with CUDA/ROCm only.

	Headless	With GUI	Empty
NVIDIA	NVIDIA headless	NVIDIA with GUI	NVIDIA empty
AMD	AMD headless	AMD with GUI	AMD empty

Note: You need to be logged in to GitHub to download artifacts. You can use nightly.link to download artifacts without authorization.

Misc

Note that you need pre-downloaded models on a local hard drive or NFS server, or enough RAM and internet connection to download models directly into RAM.

Note that following projects is not available on ROCm version:

• vllm

The server for building the ROCm version is provided by @Sepera-okeq.