Skip to content

Mateusz-Dera/ROCm-AI-Installer

Repository files navigation

ROCm-AI-Installer

Installation scripts for an AI applications using ROCm on Linux.

Info:

Version

ROCm

Note

From version 10.0, the script is distribution-independent thanks to the use of Podman.
All you need is a correctly configured Podman and amdgpu.

Important

All models and applications are tested on a GPU with 24GB of VRAM.
Some applications may not work on GPUs with less VRAM.

Test platform:

Name Info
CPU AMD Ryzen 9 9950X3D
GPU AMD Radeon 7900XTX
RAM 64GB DDR5 6600MHz
Motherboard Gigabyte X870 AORUS ELITE WIFI7 (BIOS F8)
OS Debian 13.5
Kernel 6.12.90+deb13.1-amd64

Text generation:

Name Links Additional information
KoboldCPP https://github.com/YellowRoseCx/koboldcpp-rocm
SillyTavern https://github.com/SillyTavern/SillyTavern
llama.cpp https://github.com/ggml-org/llama.cpp
https://huggingface.co/unsloth/gemma-4-12b-it-GGUF
1. Multi-Token Prediction (MTP)
2. By default, it downloads gemma-4-12b-it Q8_0 model and MTP head
3. A Vulkan version is also available.
turboquant-rocm-llamacpp https://github.com/jagsan-cyber/turboquant-rocm-llamacpp
https://huggingface.co/unsloth/gemma-4-12b-it-GGUF
1. TurboQuant 4-bit KV-cache compression (q4_0)
2. By default, it downloads gemma-4-12b-it Q8_0 model
3. ROCm only.

SillyTavern Extensions:

Name Link Additional information
WhisperSpeech web UI https://github.com/Mateusz-Dera/whisperspeech-webui Install and run WhisperSpeech web UI first.

Image & video generation:

Name Links Additional information
ComfyUI https://github.com/comfyanonymous/ComfyUI
https://github.com/city96/ComfyUI-GGUF
Workflows templates are in the workflows folder.
Extension manager is installed by default.
ComfyUI-GGUF is installed by default.

ComfyUI Addons:

Name Link Additional information
Qwen-Image-2512-GGUF https://huggingface.co/Qwen/Qwen-Image-2512
https://huggingface.co/unsloth/Qwen-Image-2512-GGUF
https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI
https://huggingface.co/Wuli-art/Qwen-Image-2512-Turbo-LoRA-2-Steps
Uses Q5_0 quant.
Uses 2-step turbo LoRA.
Qwen-Image-2511-Edit-GGUF https://huggingface.co/Qwen/Qwen-Image-Edit-2511
https://huggingface.co/unsloth/Qwen-Image-Edit-2511-GGUF
https://huggingface.co/Comfy-Org/Qwen-Image_ComfyUI
https://huggingface.co/lightx2v/Qwen-Image-Edit-2511-Lightning
Uses Q5_0 quant.
Uses 4-step Lightning LoRA
Z-Image-Turbo https://huggingface.co/Tongyi-MAI/Z-Image-Turbo
https://huggingface.co/Comfy-Org/z_image_turbo
Z-Anime https://huggingface.co/SeeSee21/Z-Anime
https://huggingface.co/Comfy-Org/z_image_turbo
Wan2.2-TI2V-5B https://huggingface.co/Wan-AI/Wan2.2-TI2V-5B
https://huggingface.co/Comfy-Org/Wan_2.2_ComfyUI_Repackaged

Music generation:

Name Links Additional information
ACE-Step-1.5 https://github.com/ace-step/ACE-Step-1.5

Voice generation:

Name Links Additional information
WhisperSpeech web UI https://github.com/Mateusz-Dera/whisperspeech-webui
https://github.com/collabora/WhisperSpeech
Soprano https://github.com/ekwek1/soprano
https://github.com/Mateusz-Dera/soprano-rocm
1. Uses my experimental fork for ROCm with vLLM
OmniVoice https://github.com/k2-fsa/OmniVoice

3D generation:

Name Links Additional information
PartCrafter https://github.com/wgsxm/PartCrafter 1. Added custom simple UI.
2. Uses a modified version of PyTorch Cluster for ROCm https://github.com/Mateusz-Dera/pytorch_cluster_rocm.
TRELLIS.2_rocm https://github.com/hqnicolas/TRELLIS.2_rocm 1. Go to https://huggingface.co/facebook/dinov3-vitl16-pretrain-lvd1689m
2. If you haven't set the HuggingFace Token in Variables, add it and run Create a container again.
3. Recommended resolution: 512.
Kimodo https://github.com/nv-tlabs/kimodo 1. Set HuggingFace Token in Variables.
2. Request access to meta-llama/Meta-Llama-3-8B-Instruct (required for text encoder).
TripoSplat https://github.com/VAST-AI-Research/TripoSplat

Instalation:

1. Install Podman.

Note

If you are using Debian 13.5, you can use sudo apt-get update && sudo apt-get -y install podman podman-compose qemu-system (should also work on Ubuntu 26.04)

2. Make sure that /dev/dri and /dev/kfd are accessible.

ls /dev/dri
ls /dev/kfd

Important

Your distribution must have amdgpu configured.

3. Make sure that your user has permissions for the video and render groups.

sudo usermod -aG video,render $USER

Important

If not, you need reboot after this step.

4. Clone repository.

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

5. Run installer.

./install.sh

6. Set variables

Note

By default, the script is configured for AMD Radeon 7900XTX.
For other cards and architectures, edit GFX and HSA_OVERRIDE_GFX_VERSION.

7. Create a container if you are upgrading or running the script for the first time.

8. Install the applications of your choice.

9. Go to the application folder and run:

./run.sh

Note

Everything is configured to start from the host side (You don't need to enter the container).

Container:

Checking the container

To check if the container is running:

podman ps

Starting the container

If the container is not running, start it with:

podman start rocm

Accessing container bash

To enter the container's bash shell:

podman exec -it rocm bash

Removing the container

To stop and remove the container:

podman stop rocm
podman rm rocm

Or force remove (stop and remove in one command):

podman rm -f rocm