ROCm-AI-Installer

A script that automatically installs all the required stuff to run selected AI interfaces on AMD Radeon 7900XTX. It should also work on 7900XT cards. For other cards, change HSA_OVERRIDE_GFX_VERSION and GFX at the beginning of the script (Not tested).

Info

Note

Ubuntu 24.04.1 is recommended. Version 6.x is not tested on older systems.

Test platform:

Name	Info
CPU	AMD Ryzen 7900X3D (iGPU disabled in BIOS)
GPU	AMD Radeon 7900XTX
RAM	64GB DDR5 6600MHz
Motherboard	ASRock B650E PG Riptide WiFi (3.08)
OS	Ubuntu 24.04.1
Kernel	6.8.0-51-generic
ROCm	6.3.1

Text generation

Name	Enviroment	Links	Additional information
KoboldCPP	Python 3.12 venv	https://github.com/YellowRoseCx/koboldcpp-rocm	GGML and GGUF models.
Text generation web UI	Python 3.12 venv	https://github.com/oobabooga/text-generation-webui https://github.com/ROCm/bitsandbytes.git https://github.com/ROCmSoftwarePlatform/flash-attention https://github.com/turboderp/exllamav2	ExLlamaV2 and Transformers models.
SillyTavern (1.12.9)	Node	https://github.com/SillyTavern/SillyTavern
llama.cpp	C++	https://github.com/ggerganov/llama.cpp	1. Put model.gguf into llama.cpp folder. 2. Change context size in run.sh file (Default: 32768). 3. Set GPU offload layers in run.sh file (Default: 1)

Image generation

Name	Enviroment	Links	Additional information
ANIMAGINE XL 3.1	Python 3.12 venv	https://huggingface.co/spaces/cagliostrolab/animagine-xl-3.1 https://huggingface.co/cagliostrolab/animagine-xl-3.1
ComfyUI	Python 3.12 venv	https://github.com/comfyanonymous/ComfyUI	1. Prepared workflows examples.
Artist	Python 3.12 venv	https://github.com/songrise/Artist/

ComfyUI Addons

Name	Link	Additional information
ComfyUI-Manager	https://github.com/ltdrdata/ComfyUI-Manager	Manage nodes of ComfyUI.
ComfyUI-GGUF	https://github.com/city96/ComfyUI-GGUF https://huggingface.co/city96/t5-v1_1-xxl-encoder-bf16 https://huggingface.co/openai/clip-vit-large-patch14 https://huggingface.co/black-forest-labs/FLUX.1-schnell	GGUF models loader.
ComfyUI-AuraSR	https://github.com/alexisrolland/ComfyUI-AuraSR https://huggingface.co/fal/AuraSR https://huggingface.co/fal/AuraSR-v2	ComfyUI node to upscale images.
AuraFlow-v0.3	https://huggingface.co/fal/AuraFlow-v0.3	Text to image model.
FLUX.1-schnell GGUF	https://huggingface.co/black-forest-labs/FLUX.1-schnell https://huggingface.co/city96/FLUX.1-schnell-gguf	Text to image model.
AnimePro FLUX GGUF	https://civitai.com/models/934628?modelVersionId=1053818	Text to image model.

Video generation

Name	Enviroment	Links	Additional information
Cinemo	Python 3.12 venv	https://huggingface.co/spaces/maxin-cn/Cinemo https://github.com/maxin-cn/Cinemo

Music generation

Name	Enviroment	Links	Additional information
AudioCraft	Python 3.12 venv	https://github.com/facebookresearch/audiocraft

Voice generation

Name	Enviroment	Links	Additional information
WhisperSpeech web UI	Python 3.12 venv	https://github.com/Mateusz-Dera/whisperspeech-webui https://github.com/collabora/WhisperSpeech https://github.com/ROCmSoftwarePlatform/flash-attention
MeloTTS	Python 3.12 venv	https://github.com/myshell-ai/MeloTTS
MetaVoice	Python 3.12 venv	https://github.com/metavoiceio/metavoice-src https://github.com/metavoiceio/metavoice-src/tree/sidroopdaska/faster_decoding	1. Script uses the faster_decoding branch. 2. Telemetry is disabled by default
F5-TTS	Python 3.12 venv	https://github.com/SWivid/F5-TTS	1. Remember to select the voice file when using the interface.
Matcha-TTS	Python 3.12 venv	https://github.com/shivammehta25/Matcha-TTS

3D generation

Name	Enviroment	Links	Additional information
TripoSR	Python3.12 venv	https://github.com/VAST-AI-Research/TripoSR https://github.com/ROCmSoftwarePlatform/flash-attention	1. It uses PyTorch ROCm, but torchmcubes is built for the CPU. This method is still faster than using just PyTorch CPU-only version.

Tools

Name	Enviroment	Links	Additional information
Fastfetch	Shell	https://github.com/fastfetch-cli/fastfetch	Custom Fastfetch configuration with GPU memory info.

Instalation:

Note

First startup after installation of the selected interface may take longer.

Important

This script does not download any models. If the interface does not have defaults, download your own.

Caution

If you update, back up your settings and models. Reinstallation deletes the previous directories.

1. Add the user to the required groups.

sudo adduser `whoami` video
sudo adduser `whoami` render

2. Reboot

sudo reboot

3. Clone repository

git clone https://github.com/Mateusz-Dera/ROCm-AI-Installer.git

4. Run installer

bash ./install.sh

5. Select installation path.

6. Select ROCm installation if you are upgrading or running the script for the first time.

7. Install selected interfaces

8. Go to the installation path with the selected interface and run:

./run.sh