XPU Support #1072

alexfreyre · 2025-10-22T19:57:12Z

alexfreyre
Oct 22, 2025

This discussion is focused on tracking the progress for the implementation of XPU backend.

Testing platform:

Ubuntu 24.04.2 LTS

Hardware

CPU AMD Ryzen 9 5900X x 24
RAM 32GB
GPU DG2 Intel ARC A770

Steps from Ubuntu 24.04 clean installation

Graphics Drivers & Compute Libraries Installation

https://dgpu-docs.intel.com/driver/client/overview.html

# Required to switch to the HWE Kernel
sudo apt install --install-recommends linux-generic-hwe-24.04

#  intel tools and software repositories
sudo apt install -y software-properties-common intel-gpu-tools 

# Intel Official PPA
sudo add-apt-repository -y ppa:kobuk-team/intel-graphics
sudo apt update
sudo apt upgrade -y

# GPGPU Intel Compute Runtime
sudo apt install -y libze-intel-gpu1 libze1 intel-metrics-discovery intel-opencl-icd clinfo intel-gsc

# media
sudo apt install -y intel-media-va-driver-non-free libmfx-gen1 libvpl2 libvpl-tools libva-glx2 va-driver-all vainfo

# Pytorch
sudo apt install -y libze-dev intel-ocloc

# OneAPI Base Toolkit
https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html

# Required build tools
sudo apt -y install cmake pkg-config build-essential

# download the key to system keyring
wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB \
| gpg --dearmor | sudo tee /usr/share/keyrings/oneapi-archive-keyring.gpg > /dev/null

# add signed entry to apt sources and configure the APT client to use Intel repository:
echo "deb [signed-by=/usr/share/keyrings/oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main" | sudo tee /etc/apt/sources.list.d/oneAPI.list

sudo apt update
sudo apt install intel-oneapi-base-toolkit-2025.0

echo 'source /opt/intel/oneapi/setvars.sh' >> ~/.bashrc

sudo gpasswd -a ${USER} render
newgrp render

# Verifying Installation
clinfo | grep "Device Name"

  Device Name                                     Intel(R) Arc(TM) A770 Graphics


source /opt/intel/oneapi/setvars.sh

:: initializing oneAPI environment ...
   -bash: BASH_VERSION = 5.2.21(1)-release
   args: Using "$@" for setvars.sh arguments: 
:: advisor -- latest
:: ccl -- latest
:: compiler -- latest
:: dal -- latest
:: debugger -- latest
:: dev-utilities -- latest
:: dnnl -- latest
:: dpcpp-ct -- latest
:: dpl -- latest
:: ipp -- latest
:: ippcp -- latest
:: mkl -- latest
:: mpi -- latest
:: pti -- latest
:: tbb -- latest
:: umf -- latest
:: vtune -- latest
:: oneAPI environment initialized ::

OneTrainer installation

Dependency needed

sudo apt install python3.12-venv

git clone https://github.com/Nerogar/OneTrainer.git
cd OneTrainer
python3 -m venv venv
source venv/bin/activate

python -m pip install torch==2.7.0 torchvision==0.22.0 torchaudio==2.7.0 --index-url https://download.pytorch.org/whl/xpu
python -m pip install intel-extension-for-pytorch==2.7.10+xpu oneccl_bind_pt==2.7.0+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/us/

pip install -r requirements-global.txt

pip install onnxruntime==1.22.1

First Run

./start-ui.sh 
[OneTrainer] Using Python Venv environment in "venv"...
[OneTrainer] + python scripts/util/version_check.py 3.10 3.13
[OneTrainer] + python scripts/train_ui.py
Traceback (most recent call last):
  File "/home/alex/OneTrainer/scripts/train_ui.py", line 5, in <module>
    from modules.ui.TrainUI import TrainUI
  File "/home/alex/OneTrainer/modules/ui/TrainUI.py", line 15, in <module>
    from tkinter import filedialog
ModuleNotFoundError: No module named 'tkinter'

Dependency needed

sudo apt install python3-tk

Second Run

OneTrainer opens, model path and concepts configured for sd15.

Start Training

./start-ui.sh 
[OneTrainer] Using Python Venv environment in "venv"...
[OneTrainer] + python scripts/util/version_check.py 3.10 3.13
[OneTrainer] + python scripts/train_ui.py
[W1022 14:49:02.081481304 OperatorEntry.cpp:154] Warning: Warning only once for all operators,  other operators may also be overridden.
  Overriding a previously registered kernel for the same operator and the same dispatch key
  operator: aten::geometric_(Tensor(a!) self, float p, *, Generator? generator=None) -> Tensor(a!)
    registered at /pytorch/build/aten/src/ATen/RegisterSchema.cpp:6
  dispatch key: XPU
  previous kernel: registered at /pytorch/aten/src/ATen/VmapModeRegistrations.cpp:37
       new kernel: registered at /build/intel-pytorch-extension/build/Release/csrc/gpu/csrc/gpu/xpu/ATen/RegisterXPU_0.cpp:186 (function operator())
Failed to load config from training_concepts/concepts.json: [Errno 2] No such file or directory: 'training_concepts/concepts.json'
Failed to load config from training_samples/samples.json: [Errno 2] No such file or directory: 'training_samples/samples.json'



Exception in thread Thread-3 (__training_thread_function):
Traceback (most recent call last):
  File "/usr/lib/python3.12/threading.py", line 1073, in _bootstrap_inner
    self.run()
  File "/usr/lib/python3.12/threading.py", line 1010, in run
    self._target(*self._args, **self._kwargs)
  File "/home/alex/OneTrainer/modules/ui/TrainUI.py", line 752, in __training_thread_function
    trainer = create.create_trainer(self.train_config, self.training_callbacks, self.training_commands, reattach=self.cloud_tab.reattach)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/alex/OneTrainer/modules/util/create.py", line 1610, in create_trainer
    ZLUDA.initialize_devices(config)
  File "/home/alex/OneTrainer/modules/zluda/ZLUDA.py", line 37, in initialize_devices
    if not is_zluda(config.train_device) and not is_zluda(config.temp_device):
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/alex/OneTrainer/modules/zluda/ZLUDA.py", line 12, in is_zluda
    return torch.cuda.get_device_name(device).endswith("[ZLUDA]")
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/alex/OneTrainer/venv/lib/python3.12/site-packages/torch/cuda/__init__.py", line 544, in get_device_name
    return get_device_properties(device).name
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/alex/OneTrainer/venv/lib/python3.12/site-packages/torch/cuda/__init__.py", line 576, in get_device_properties
    _lazy_init()  # will define _get_device_properties
    ^^^^^^^^^^^^
  File "/home/alex/OneTrainer/venv/lib/python3.12/site-packages/torch/cuda/__init__.py", line 363, in _lazy_init
    raise AssertionError("Torch not compiled with CUDA enabled")
AssertionError: Torch not compiled with CUDA enabled

dxqb · 2025-10-23T07:51:11Z

dxqb
Oct 23, 2025
Collaborator

Are you planning to work on this? Or is this meant as a report that it doesn't work?

4 replies

alexfreyre Oct 23, 2025
Author

I am able to do recurrent tests and submit the results as discussed on discord. This is the initial report that shows the xpu backend is not able to start. The purpose of this discussion is to track the errors until the backend gets implemented. I am not implementing it on my own, just providing the gpu as playground and the reports of installations with further training testings.

O-J1 Oct 23, 2025
Collaborator

@alexfreyre I think I wasnt clear enough in the discord 😅 we (Dxqb and I) cannot implement something we dont have. Let me restate my response to you with tweaks to make it clearer:

"Your comment frames it as if this is our fault and responsibility to support something we don't have. This is open source software. I would invite you submit a pull request after doing extensive testing to demonstrate training works. The UI launching means nothing and has no relation to your GPU, its using the CPU only to render the UI.

If your testing of training results satisfies us we can merge.

The tests being the most crucial part. You would need to demonstrate to us it works on Ubuntu and ideally Windows, though windows is not a blocker."

alexfreyre Oct 23, 2025
Author

Got it 😅! It is clear now, thanks! I will close this discussion then.

O-J1 Oct 23, 2025
Collaborator

Sorry for not being clearer initially, because what you were asking us is to drive a car blind. Thats a no go. We need someone with knowledge of Arc and owns a card to submit changes.

kainatquaderee · 2026-04-08T18:57:28Z

kainatquaderee
Apr 8, 2026

I have managed to make it work without breaking anything. should i create a pr?

here is what i did ->

XPU (Intel GPU) Integration Notes — ontrainer

1. Clone IPEX bridge into module folder

git clone https://github.com/Disty0/ipex_to_cuda  modules/ipex_to_cuda/

Make sure the path is: modules/ipex_to_cuda/src/ipex_to_cuda/init.py

2. Modify scripts/train_ui.py

Add this block after importing torch, before any CUDA ops run:

import torch

if [torch.xpu.is](http://torch.xpu.is/)_available():

    from modules.ipex_to_cuda.src.ipex_to_cuda.__init__ import ipex_init

    ipex_init()

    print("✓ CUDA → XPU hijacking active")

3. Fix modules/util/torch_util.py

Function: def torch_gc():

Old code:

if [torch.cuda.is](http://torch.cuda.is/)_available():

    torch.cuda.empty_cache()

    if torch_version > Version("2.6.0"):

        torch._C._host_emptyCache()

Replace with:

if [torch.cuda.is](http://torch.cuda.is/)_available():

    torch.cuda.empty_cache()

    if torch_version > Version("2.6.0"):

        # TODO: replace with a torch.cuda binding once that's available

        if hasattr(torch._C, '_host_emptyCache'):

            print("true")

            torch._C._host_emptyCache()

then i created conda environment with python3.11 and also installed

pip install -r requirements-global.txt

then installed xpu with

pip install torch==2.9.1 torchvision==0.24.1 torchaudio==2.9.1 --index-url https://download.pytorch.org/whl/xpu

morei info at https://docs.pytorch.org/docs/main/notes/get_start_xpu.html

Why this matters:

The hasattr check prevents crashes on XPU builds where _host_emptyCache might not exist
The ipex_init() hijacks CUDA API calls and redirects them to XPU (Intel GPUs)
Without this, training scripts think there's no GPU and fall back to CPU

0 replies

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

XPU Support #1072

Uh oh!

{{title}}

Uh oh!

Replies: 2 comments 4 replies

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{editor}}'s edit

{{editor}}'s edit

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{title}}

Uh oh!

Uh oh!

{{editor}}'s edit

{{editor}}'s edit

Uh oh!

Select a reply

Uh oh!

Uh oh!

XPU Support #1072

Uh oh!

alexfreyre Oct 22, 2025

Testing platform:

Hardware

Steps from Ubuntu 24.04 clean installation

Graphics Drivers & Compute Libraries Installation

OneTrainer installation

First Run

Second Run

Replies: 2 comments · 4 replies

Uh oh!

dxqb Oct 23, 2025 Collaborator

Uh oh!

alexfreyre Oct 23, 2025 Author

Uh oh!

Uh oh!

O-J1 Oct 23, 2025 Collaborator

Uh oh!

alexfreyre Oct 23, 2025 Author

Uh oh!

O-J1 Oct 23, 2025 Collaborator

Uh oh!

Uh oh!

kainatquaderee Apr 8, 2026

XPU (Intel GPU) Integration Notes — ontrainer

1. Clone IPEX bridge into module folder

2. Modify scripts/train_ui.py

3. Fix modules/util/torch_util.py

morei info at https://docs.pytorch.org/docs/main/notes/get_start_xpu.html

Why this matters:

alexfreyre
Oct 22, 2025

Replies: 2 comments 4 replies

dxqb
Oct 23, 2025
Collaborator

alexfreyre Oct 23, 2025
Author

O-J1 Oct 23, 2025
Collaborator

alexfreyre Oct 23, 2025
Author

O-J1 Oct 23, 2025
Collaborator

kainatquaderee
Apr 8, 2026