Any way to run this on Intel Arc?

[HyunJae5463]

Guess it doesn't support IPEX? Tried it and it gives TypeError: 'NoneType' object is not iterable Error

[congdm]

It can run on IPEX. I haven't tested extensions but generating images is fine.

[HyunJae5463]

It can run on IPEX. I haven't tested extensions but generating images is fine.

As i said, when i try to generate anything it says "TypeError: 'NoneType' object is not iterable"

[congdm]

How about trying to reinstall it?

1. Delete the venv folder
2. Check if COMMANDLINE_ARGS=--use_ipex in webui-user.bat
3. Run webui-user.bat to reinstall
   If even reinstalling doesn't fix it then maybe it's another problem

[HyunJae5463]

How about trying to reinstall it?

1. Delete the venv folder
2. Check if COMMANDLINE_ARGS=--use_ipex in webui-user.bat
3. Run webui-user.bat to reinstall
   If even reinstalling doesn't fix it then maybe it's another problem

Tried reinstalling twice. Second time on a fresh Windows User. Doesnt work.

[redrum-llik]

Have the same issue on a clean install - the underlying exception is as following:

  File "C:\Forge\modules\rng.py", line 86, in create_generator
    generator = torch.Generator(device).manual_seed(int(seed))
RuntimeError: Device type XPU is not supported for torch.Generator() api.
Device type XPU is not supported for torch.Generator() api.

I have been able to work around the issue by amending the modules/rng.py script - since this Generator class seems to be used for the RNG only, I have amended the device checks to use CPU instead of GPU if Intel Arc (XPU) device is detected. Please find the unified diff below:

 modules/rng.py | 13 ++++++++-----
 1 file changed, 8 insertions(+), 5 deletions(-)

diff --git a/modules/rng.py b/modules/rng.py
index 4a3bbb20..1426a428 100644
--- a/modules/rng.py
+++ b/modules/rng.py
@@ -3,6 +3,8 @@ import torch
 from modules import devices, rng_philox, shared
 
 
+TORCH_UNSUPPORTED_DEVICES = ['mps', 'xpu']
+
 def randn(seed, shape, generator=None):
     """Generate a tensor with random numbers from a normal distribution using seed.
 
@@ -19,7 +21,7 @@ def randn(seed, shape, generator=None):
     if shared.opts.randn_source == "NV":
         return torch.asarray((generator or nv_rng).randn(shape), device=devices.device)
 
-    if shared.opts.randn_source == "CPU" or devices.device.type == 'mps':
+    if shared.opts.randn_source == "CPU" or devices.device.type in TORCH_UNSUPPORTED_DEVICES:
         return torch.randn(shape, device=devices.cpu, generator=generator).to(devices.device)
 
     return torch.randn(shape, device=devices.device, generator=generator)
@@ -34,7 +36,7 @@ def randn_local(seed, shape):
         rng = rng_philox.Generator(seed)
         return torch.asarray(rng.randn(shape), device=devices.device)
 
-    local_device = devices.cpu if shared.opts.randn_source == "CPU" or devices.device.type == 'mps' else devices.device
+    local_device = devices.cpu if shared.opts.randn_source == "CPU" or devices.device.type in TORCH_UNSUPPORTED_DEVICES else devices.device
     local_generator = torch.Generator(local_device).manual_seed(int(seed))
     return torch.randn(shape, device=local_device, generator=local_generator).to(devices.device)
 
@@ -47,7 +49,7 @@ def randn_like(x):
     if shared.opts.randn_source == "NV":
         return torch.asarray(nv_rng.randn(x.shape), device=x.device, dtype=x.dtype)
 
-    if shared.opts.randn_source == "CPU" or x.device.type == 'mps':
+    if shared.opts.randn_source == "CPU" or x.device.type in TORCH_UNSUPPORTED_DEVICES:
         return torch.randn_like(x, device=devices.cpu).to(x.device)
 
     return torch.randn_like(x)
@@ -61,7 +63,7 @@ def randn_without_seed(shape, generator=None):
     if shared.opts.randn_source == "NV":
         return torch.asarray((generator or nv_rng).randn(shape), device=devices.device)
 
-    if shared.opts.randn_source == "CPU" or devices.device.type == 'mps':
+    if shared.opts.randn_source == "CPU" or devices.device.type in TORCH_UNSUPPORTED_DEVICES:
         return torch.randn(shape, device=devices.cpu, generator=generator).to(devices.device)
 
     return torch.randn(shape, device=devices.device, generator=generator)
@@ -82,7 +84,8 @@ def create_generator(seed):
     if shared.opts.randn_source == "NV":
         return rng_philox.Generator(seed)
 
-    device = devices.cpu if shared.opts.randn_source == "CPU" or devices.device.type == 'mps' else devices.device
+    device = devices.cpu if shared.opts.randn_source == "CPU" or devices.device.type in TORCH_UNSUPPORTED_DEVICES else devices.device
     generator = torch.Generator(device).manual_seed(int(seed))
     return generator

There is also a warning stating that Forge expects a higher version of torch:

You are running torch 2.0.0a0+gite9ebda2.
The program is tested to work with torch 2.1.2.

This is the same version I end up with when I run Forge with --reinstall-torch flag. @congdm, if it works for you without any manual changes, do you mind sharing your local version of torch? Perhaps the newer torch releases do support xpu devices; if they do not, I will create a PR for the above change.

[redrum-llik]

Apparently, yeah, xpu is supported by Torch 2.1+ but on Windows, Forge will run with 2.0.X custom build at the moment. We might want to wait until there is a newer build available and the installation script is updated, or test it out manually for a quicker turnaround.

[congdm]

@redrum-llik Hi, actually I also got that Generator error too, however Forge allowed us to change RNG type to CPU in Settings/Stable Diffusion tab (Random number generator source setting), no need to edit any source file.

@HyunJae5463 I just discovered that some sampler could cause TypeError: 'NoneType' exception. Have you tested it on various samplers or settings, maybe it could help.

[redrum-llik]

no need to edit any source file.

Ah cool, thank you for the notice!
I have run into one more issue that might be specific to Intel Arc GPUs, which is described here. IPEX is known to have a limit of 4 GBs for memory allocations (for performance reasons), and there is a piece of code that is meant to handle that, but it does not seem to be invoked in my case. Have you seen this one, and, if yes, is there any workaround you could recommend?

[redrum-llik]

A quick update on that - made a crude hack that seems to fix the allocation issue on the newer versions, you may find the diff here.

[Tolga077]

mine works fine install through stability matrix. https://github.com/LykosAI/StabilityMatrix

[them3rcury]

Launching Web UI with arguments: --use_ipex --skip-torch-cuda-test
Traceback (most recent call last):
  File "E:\stable-diffusion-webui-forge\launch.py", line 51, in <module>
    main()
  File "E:\stable-diffusion-webui-forge\launch.py", line 47, in main
    start()
  File "E:\stable-diffusion-webui-forge\modules\launch_utils.py", line 541, in start
    import webui
  File "E:\stable-diffusion-webui-forge\webui.py", line 17, in <module>
    initialize_forge()
  File "E:\stable-diffusion-webui-forge\modules_forge\initialization.py", line 50, in initialize_forge
    import ldm_patched.modules.model_management as model_management
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 122, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 91, in get_torch_device
    return torch.device(torch.cuda.current_device())
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\cuda\__init__.py", line 769, in current_device
    _lazy_init()
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\cuda\__init__.py", line 298, in _lazy_init
    torch._C._cuda_init()
RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx

After trying to use it on Windows with Intel Arc A750. I wonder what I'm doing wrong.

[redrum-llik]

Hello! You have --use_ipex in your args, but the correct one is --use-ipex (with hyphen). Please rerun the script with --use-ipex --reinstall-torch for the first run, and subsequent runs should be good with --use-ipex only. If that does not help for any reason, as @congdm suggested below, remove venv folder and run the script with just --use-ipex.

[congdm]

Launching Web UI with arguments: --use_ipex --skip-torch-cuda-test
Traceback (most recent call last):
  File "E:\stable-diffusion-webui-forge\launch.py", line 51, in <module>
    main()
  File "E:\stable-diffusion-webui-forge\launch.py", line 47, in main
    start()
  File "E:\stable-diffusion-webui-forge\modules\launch_utils.py", line 541, in start
    import webui
  File "E:\stable-diffusion-webui-forge\webui.py", line 17, in <module>
    initialize_forge()
  File "E:\stable-diffusion-webui-forge\modules_forge\initialization.py", line 50, in initialize_forge
    import ldm_patched.modules.model_management as model_management
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 122, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 91, in get_torch_device
    return torch.device(torch.cuda.current_device())
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\cuda\__init__.py", line 769, in current_device
    _lazy_init()
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\cuda\__init__.py", line 298, in _lazy_init
    torch._C._cuda_init()
RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx

After trying to use it on Windows with Intel Arc A750. I wonder what I'm doing wrong.

Delete venv folder in webui and run again, it should work, I think

[them3rcury]

So after changing it to --use-ipex and reinstalling the webui, it gave out the error TypeError: 'NoneType' object is not iterable

Changing the sampler didnt change the outcome. Then I went on and changed the RNG source to CPU and voilà it works like a charm.

512x512 with Euler A getting 8.80it/s and 768x1024 at 4.40it/s with Euler A. Thanks all, appreciate the help.

[redrum-llik]

One more hack to collection, this time related to the issue with GFPGAN/Codeformer not running on IPEX which affects ReActor. If you use it (or if you run face restoration with your generations), consider changing the device that is used to run those models to CPU using this approach. We probably could benefit from being able to select this device via settings or CMD flags.

[them3rcury]

I know I said it fixed the issue, but I still get the error now and then randomly while generating images, only fix is to restart the webui and hope it doesnt do it again. It's more common when using SDXL or SDXL sometimes doesnt work at all. Pretty random. But when it works its making wonders.

[redrum-llik]

Can you share the full stacktrace? Perhaps you are seeing the allocation issue which I have mentioned above.

[them3rcury]

Traceback (most recent call last):
  File "E:\stable-diffusion-webui-forge\modules_forge\main_thread.py", line 37, in loop
    task.work()
  File "E:\stable-diffusion-webui-forge\modules_forge\main_thread.py", line 26, in work
    self.result = self.func(*self.args, **self.kwargs)
  File "E:\stable-diffusion-webui-forge\modules\txt2img.py", line 111, in txt2img_function
    processed = processing.process_images(p)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 752, in process_images
    res = process_images_inner(p)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 922, in process_images_inner
    samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 1275, in sample
    samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
  File "E:\stable-diffusion-webui-forge\modules\sd_samplers_kdiffusion.py", line 251, in sample
    samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "E:\stable-diffusion-webui-forge\modules\sd_samplers_common.py", line 263, in launch_sampling
    return func()
  File "E:\stable-diffusion-webui-forge\modules\sd_samplers_kdiffusion.py", line 251, in <lambda>
    samples = self.launch_sampling(steps, lambda: self.func(self.model_wrap_cfg, x, extra_args=self.sampler_extra_args, disable=False, callback=self.callback_state, **extra_params_kwargs))
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\utils\_contextlib.py", line 115, in decorate_context
    return func(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\repositories\k-diffusion\k_diffusion\sampling.py", line 145, in sample_euler_ancestral
    denoised = model(x, sigmas[i] * s_in, **extra_args)
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\modules\sd_samplers_cfg_denoiser.py", line 182, in forward
    denoised = forge_sampler.forge_sample(self, denoiser_params=denoiser_params,
  File "E:\stable-diffusion-webui-forge\modules_forge\forge_sampler.py", line 88, in forge_sample
    denoised = sampling_function(model, x, timestep, uncond, cond, cond_scale, model_options, seed)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\samplers.py", line 289, in sampling_function
    cond_pred, uncond_pred = calc_cond_uncond_batch(model, cond, uncond_, x, timestep, model_options)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\samplers.py", line 258, in calc_cond_uncond_batch
    output = model.apply_model(input_x, timestep_, **c).chunk(batch_chunks)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_base.py", line 90, in apply_model
    model_output = self.diffusion_model(xc, t, context=context, control=control, transformer_options=transformer_options, **extra_conds).float()
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\diffusionmodules\openaimodel.py", line 867, in forward
    h = forward_timestep_embed(module, h, emb, context, transformer_options, time_context=time_context, num_video_frames=num_video_frames, image_only_indicator=image_only_indicator)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\diffusionmodules\openaimodel.py", line 55, in forward_timestep_embed
    x = layer(x, context, transformer_options)
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\attention.py", line 620, in forward
    x = block(x, context=context[i], transformer_options=transformer_options)
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\attention.py", line 447, in forward
    return checkpoint(self._forward, (x, context, transformer_options), self.parameters(), self.checkpoint)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\diffusionmodules\util.py", line 194, in checkpoint
    return func(*inputs)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\attention.py", line 507, in _forward
    n = self.attn1(n, context=context_attn1, value=value_attn1, transformer_options=extra_options)
  File "E:\stable-diffusion-webui-forge\venv\lib\site-packages\torch\nn\modules\module.py", line 1501, in _call_impl
    return forward_call(*args, **kwargs)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\attention.py", line 399, in forward
    out = optimized_attention(q, k, v, self.heads)
  File "E:\stable-diffusion-webui-forge\ldm_patched\ldm\modules\attention.py", line 332, in attention_pytorch
    out = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=mask, dropout_p=0.0, is_causal=False)
RuntimeError: Native API failed. Native API returns: -997 (Command failed to enqueue/execute) -997 (Command failed to enqueue/execute)
Native API failed. Native API returns: -997 (Command failed to enqueue/execute) -997 (Command failed to enqueue/execute)
*** Error completing request
    Traceback (most recent call last):
      File "E:\stable-diffusion-webui-forge\modules\call_queue.py", line 57, in f
        res = list(func(*args, **kwargs))
    TypeError: 'NoneType' object is not iterable

---

This is with SDXL, ADetailer/HiresFix disabled, hypertile enabled

Total progress: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:20<00:00,  1.01s/it]
NeverOOM Enabled for UNet (always maximize offload)████████████████████████████████████| 20/20 [00:20<00:00,  1.03it/s]
To load target model SDXL
Begin to load 1 model
ERROR diffusion_model.output_blocks.0.1.transformer_blocks.0.ff.net.0.proj.weight [enforce fail at C:\Users\vfirs\projects\ipex-standalone\pytorch\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 52428800 bytes.
ERROR diffusion_model.output_blocks.0.1.transformer_blocks.1.ff.net.0.proj.weight [enforce fail at C:\Users\vfirs\projects\ipex-standalone\pytorch\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 52428800 bytes.
Traceback (most recent call last):
  File "E:\stable-diffusion-webui-forge\modules_forge\main_thread.py", line 37, in loop
    task.work()
  File "E:\stable-diffusion-webui-forge\modules_forge\main_thread.py", line 26, in work
    self.result = self.func(*self.args, **self.kwargs)
  File "E:\stable-diffusion-webui-forge\modules\txt2img.py", line 111, in txt2img_function
    processed = processing.process_images(p)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 752, in process_images
    res = process_images_inner(p)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 922, in process_images_inner
    samples_ddim = p.sample(conditioning=p.c, unconditional_conditioning=p.uc, seeds=p.seeds, subseeds=p.subseeds, subseed_strength=p.subseed_strength, prompts=p.prompts)
  File "E:\stable-diffusion-webui-forge\modules\processing.py", line 1275, in sample
    samples = self.sampler.sample(self, x, conditioning, unconditional_conditioning, image_conditioning=self.txt2img_image_conditioning(x))
  File "E:\stable-diffusion-webui-forge\modules\sd_samplers_kdiffusion.py", line 207, in sample
    sampling_prepare(self.model_wrap.inner_model.forge_objects.unet, x=x)
  File "E:\stable-diffusion-webui-forge\modules_forge\forge_sampler.py", line 105, in sampling_prepare
    model_management.load_models_gpu(
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 494, in load_models_gpu
    loaded_model.model_load(async_kept_memory)
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 332, in model_load
    raise e
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_management.py", line 328, in model_load
    self.real_model = self.model.patch_model(device_to=patch_model_to) #TODO: do something with loras and offloading to CPU
  File "E:\stable-diffusion-webui-forge\ldm_patched\modules\model_patcher.py", line 216, in patch_model
    out_weight = self.calculate_weight(self.patches[key], temp_weight, key).to(weight.dtype)
RuntimeError: [enforce fail at C:\Users\vfirs\projects\ipex-standalone\pytorch\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 26214400 bytes.
[enforce fail at C:\Users\vfirs\projects\ipex-standalone\pytorch\c10\core\impl\alloc_cpu.cpp:72] data. DefaultCPUAllocator: not enough memory: you tried to allocate 26214400 bytes.
*** Error completing request
    Traceback (most recent call last):
      File "E:\stable-diffusion-webui-forge\modules\call_queue.py", line 57, in f
        res = list(func(*args, **kwargs))
    TypeError: 'NoneType' object is not iterable

---

And I would believe this is the allocation error that you have mentioned. It doesnt happen when I generate but happens after a couple generations. Also if I dont have Never OOM UNet enabled, SDXL doesnt work as "stable" as with UNet and it crashes still with UNet.
This is with SDXL, ADetailer/HiresFix disabled, hypertile disabled and UNet enabled.
Edit: I have deleted my prompts from the logs for making it easier to read.

[redrum-llik]

RuntimeError: Native API failed. Native API returns: -997 - I am seeing this once I had an OOM at least once, until the process is restarted. Was this the first exception you saw, or a subsequent one?

The second sounds like the model was being offloaded from VRAM (to regular RAM, I suppose), but allocation has failed. Do you have any swap enabled on your system? I had some issues until I have set my swap to be twice as large as my RAM - even though it is situated on HDD, it seems to help.

[HyunJae5463]

I know I said it fixed the issue, but I still get the error now and then randomly while generating images, only fix is to restart the webui and hope it doesnt do it again. It's more common when using SDXL or SDXL sometimes doesnt work at all. Pretty random. But when it works its making wonders.

Sounds like the same issue I have. I just get "Error" but no further console log. Just have to restart SD. This only happens when I use any Face Restoration. It doesn't happen on SD.Next tho. Only on Forge and A1111.

[martin-afh]

Hi,

I recently switched from an NVIDIA GPU to an Intel Arc B580 GPU, and now WebUI Forge is no longer working. The issue is related to the fact that Intel GPUs don't support CUDA.

I’ve tried running WebUI Forge, but it fails to initialize. I’m looking for guidance on how to configure it to use the Intel Arc B580, as I understand CUDA-specific features need to be replaced with alternatives for Intel GPUs.
I tried to update webui-user.bat as follows:
set COMMANDLINE_ARGS=--use_ipex --disable-gpu-warning --skip-torch-cuda-test
And I installed Intel oneAPI

What are the necessary steps to make WebUI Forge compatible with Intel Arc GPUs? Do I need to enable Intel-specific extensions (like Intel® Extension for PyTorch*) or modify any scripts/configuration files?

Any help would be greatly appreciated!

Thanks in advance.

[hajimeg3]

I got a workaround for B580.
I'm not sure why it works.
Just for your information.
(Some steps may not necessary)

---

1. Install oneAPI, version 2025.0.1 for now.
   https://www.intel.com/content/www/us/en/developer/tools/oneapi/base-toolkit-download.html
2. setup python 3.10 with conda (I'm using Miniforge Prompt as admin)
   conda create -n sd-forge-ipex python==3.10
conda activate sd-forge-ipex
conda install libuv
3. Setup oneAPI variables
   "C:\Program Files (x86)\Intel\oneAPI\setvars.bat"
4. Install Intel pytorch extension with pip
   https://pytorch-extension.intel.com/installation?platform=gpu&version=v2.5.10%2Bxpu&os=windows&package=pip
   python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/bmg/us/
5. get forge repo
6. modify webui-user.bat as
   set COMMANDLINE_ARGS=--use-ipex --skip-torch-cuda-test --skip-version-check
7. run webui-user.bat from miniforge, and error dialog will appear
   close dialog and continue;
8. update pytorch in from pytorch official
   https://pytorch.org/docs/stable/notes/get_start_xpu.html
   pip install --upgrade torch --index-url https://download.pytorch.org/whl/test/xpu
9. run webui-user.bat again, and error dialog shows again.
10. override pytorch again from intel-repo.
    python -m pip install torch==2.5.1+cxx11.abi torchvision==0.20.1+cxx11.abi torchaudio==2.5.1+cxx11.abi intel-extension-for-pytorch==2.5.10+xpu --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/bmg/us/
11. run webui-user.bat, and it will work now.

---

[congdm]

Updated:
Nowadays with Windows wheel build provided by Intel, it's pretty easy to get updated IPEX.
First, clone a clean Forge, or delete old venv, then just run webui-user.bat with --use-ipex as usual. There will be error, of course, just press any key to exit, then install IPEX as instructed by https://pytorch-extension.intel.com/installation?platform=gpu&version=v2.7.10%2Bxpu&os=windows&package=pip

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

After that, run webui-user again and enjoy.

Btw, the reason I use Forge again is to checkout the latest Chroma model. This patch https://github.com/croquelois/forgeChroma works perfectly on latest Forge and IPEX. I get around 7.0s/it with standard 1024x1024, 7000 MB VRAM reserved for model, Chroma Q4_1 gguf

[mahiro21h]

did any linux user encounter forge crashing with AssertionError: Torch not compiled with CUDA enabled when using ipex 2.7.10? it looks like newer versions of glibc break the loading of an intel library. i'll leave the steps i did to fix it just in case.

you can verify that you're running into the same problem by running this command python -c 'import torch; import intel_extension_for_pytorch'. if you get ImportError: libintel-ext-pt-cpu.so: cannot enable executable stack as shared object requires: Invalid argument then you are running into it.

to fix it, run find venv/ -name 'libintel-ext-pt-cpu.so' to find the library. then do sudo patchelf --clear-execstack path/to/lib/libintel-ext-pt-cpu.so

[mahiro21h]

also, i ran into two other odd issues. the first had to do with MoeModelOutputWithPast not being defined for whatever reason so i had to patch an import in venv/lib/python3.10/site-packages/intel_extension_for_pytorch/transformers/models/reference/models.py. the second was RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cpu and xpu:0! (when checking argument for argument index in method wrapper_XPU__index_select). i got around it by adding --always-gpu to my command line arguments. did anyone encounter these issues? im using latest branch

[djg26]

With torch 2.7 I still get the 4GB allocation error, just worded differently. It happens every time I try and use hiresfix. I'm not sure if anyone else has this problem too? I have webui-user.sh as COMMANDLINE_ARGS=--use-ipex --disable-gpu-warning --skip-torch-cuda-test

RuntimeError: XPU out of memory. Tried to allocate 7.45 GiB (GPU 0; 15.11 GiB total capacity; 774.24 MiB already allocated; 1.67 GiB reserved in total by PyTorch)
XPU out of memory. Tried to allocate 7.45 GiB (GPU 0; 15.11 GiB total capacity; 774.24 MiB already allocated; 1.67 GiB reserved in total by PyTorch)

[mahiro21h]

weird, i tried to follow your steps and when i changed the xpu_available line to fix the error i get this error instead:

Check TCMalloc: libtcmalloc_minimal.so.4
libtcmalloc_minimal.so.4 is linked with libc.so,execute LD_PRELOAD=/usr/lib/libtcmalloc_minimal.so.4
Python 3.10.18 (main, Jun 11 2025, 02:42:11) [GCC 15.1.1 20250425]
Version: f2.0.1v1.10.1-previous-665-gae278f79
Commit hash: ae278f794069a69b79513e16207efc7f1ffdf406
Legacy Preprocessor init warning: Unable to install insightface automatically. Please try run `pip install insightface` manually.
Launching Web UI with arguments: --use-ipex --always-gpu
/home/void/home/stable-diffusion-webui-forge-main/venv-arc-nightly/lib/python3.10/site-packages/torch/xpu/__init__.py:60: UserWarning: XPU device count is zero! (Triggered internally at /pytorch/c10/xpu/XPUFunctions.cpp:115.)
  return torch._C._xpu_getDeviceCount()
Traceback (most recent call last):
  File "/home/void/home/stable-diffusion-webui-forge-main/launch.py", line 54, in <module>
    main()
  File "/home/void/home/stable-diffusion-webui-forge-main/launch.py", line 50, in main
    start()
  File "/home/void/home/stable-diffusion-webui-forge-main/modules/launch_utils.py", line 546, in start
    import webui
  File "/home/void/home/stable-diffusion-webui-forge-main/webui.py", line 21, in <module>
    initialize_forge()
  File "/home/void/home/stable-diffusion-webui-forge-main/modules_forge/initialization.py", line 58, in initialize_forge
    from backend import memory_management
  File "/home/void/home/stable-diffusion-webui-forge-main/backend/memory_management.py", line 133, in <module>
    total_vram = get_total_memory(get_torch_device()) / (1024 * 1024)
  File "/home/void/home/stable-diffusion-webui-forge-main/backend/memory_management.py", line 98, in get_torch_device
    return torch.device("xpu", torch.xpu.current_device())
  File "/home/void/home/stable-diffusion-webui-forge-main/venv-arc-nightly/lib/python3.10/site-packages/torch/xpu/__init__.py", line 267, in current_device
    _lazy_init()
  File "/home/void/home/stable-diffusion-webui-forge-main/venv-arc-nightly/lib/python3.10/site-packages/torch/xpu/__init__.py", line 125, in _lazy_init
    torch._C._xpu_init()
RuntimeError: No XPU devices are available.

i checked your error but i don't have the ipex import in stable-diffusion-webui-forge/modules/processing.py despite being on the main branch just like you. i would like to recommend that you remove the import, but i don't think that's going to help judging by the error i got. i'm out of ideas

[djg26]

I forgot that I tried adding ipex manually to some of the files, removed those and it starts up fine. But I get a different error when I try to generate something:

[Memory Management] Target: JointTextEncoder, Free GPU: 15473.49 MB, Model Require: 1559.68 MB, Previously Loaded: 0.00 MB, Inference Require: 1024.00 MB, Remaining: 12889.82 MB, All loaded to GPU.
Traceback (most recent call last):
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules_forge/main_thread.py", line 30, in work
    self.result = self.func(*self.args, **self.kwargs)
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/txt2img.py", line 131, in txt2img_function
    processed = processing.process_images(p)
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/processing.py", line 842, in process_images
    res = process_images_inner(p)
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/processing.py", line 962, in process_images_inner
    p.setup_conds()
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/processing.py", line 1601, in setup_conds
    super().setup_conds()
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/processing.py", line 503, in setup_conds
    self.uc = self.get_conds_with_caching(prompt_parser.get_learned_conditioning, negative_prompts, total_steps, [self.cached_uc], self.extra_network_data)
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/processing.py", line 474, in get_conds_with_caching
    cache[1] = function(shared.sd_model, required_prompts, steps, hires_steps, shared.opts.use_old_scheduling)
  File "/mnt/sdf1/stable-diffusion-webui-forge/modules/prompt_parser.py", line 189, in get_learned_conditioning
    conds = model.get_learned_conditioning(texts)
  File "/mnt/sdf1/stable-diffusion-webui-forge/venv/lib/python3.10/site-packages/torch/utils/_contextlib.py", line 120, in decorate_context
    return func(*args, **kwargs)
  File "/mnt/sdf1/stable-diffusion-webui-forge/backend/diffusion_engine/sdxl.py", line 87, in get_learned_conditioning
    memory_management.load_model_gpu(self.forge_objects.clip.patcher)
  File "/mnt/sdf1/stable-diffusion-webui-forge/backend/memory_management.py", line 694, in load_model_gpu
    return load_models_gpu([model])
  File "/mnt/sdf1/stable-diffusion-webui-forge/backend/memory_management.py", line 684, in load_models_gpu
    loaded_model.model_load(model_gpu_memory_when_using_cpu_swap)
  File "/mnt/sdf1/stable-diffusion-webui-forge/backend/memory_management.py", line 521, in model_load
    self.real_model = torch.xpu.optimize(self.real_model.eval(), inplace=True, auto_kernel_selection=True, graph_mode=True)
AttributeError: module 'torch.xpu' has no attribute 'optimize'
module 'torch.xpu' has no attribute 'optimize'

From what I can tell, it's trying to use an api in ipex and failing cause ipex isn't installed.

[mahiro21h]

yeah i also made a mistake on my machine. it works now but you have to pass --disable-ipex-hijack to get rid of this error about optimize

[djg26]

Yeah passing --disable-ipex-hijack fixes it and I can now use it with hiresfix without it erroring out now.

[congdm]

Yeah passing --disable-ipex-hijack fixes it and I can now use it with hiresfix without it erroring out now.

Haha, what a nice solution. you should post it as a separate comment from this thread for more exposure :)