Clean and fresh install on linux - AMD - backend crash nvidia - no vars used

[scryptio]

Expected Behavior

SwarmUI and backend ComfyUI up and running

Actual Behavior

The backend crashes because it tries to load nvidia drivers and libs

Steps to Reproduce

$ which pip
/usr/bin/pip
$ which python3
/usr/bin/python3
$ python3.12 -m pip --version
python3.12         python3.12-config
$ wget https://github.com/mcmonkeyprojects/SwarmUI/releases/download/0.6.5-Beta/install-linux.sh -O install-linux.sh
$ chmod +x install-linux.sh
$ ./install-linux.sh

Debug Logs

https://paste.denizenscript.com/View/137937

Other

$ apt list --installed | grep -o "^[^/]*" | egrep "^(rocm|amd)"
amd-smi-lib
amd64-microcode
amdgpu-core
amdgpu-dkms-firmware
amdgpu-dkms
amdgpu-insecure-instinct-udev-rules
amdgpu-install
rocm-cmake
rocm-core
rocm-dbgapi
rocm-debug-agent
rocm-developer-tools
rocm-device-libs
rocm-gdb
rocm-hip
rocm-language-runtime
rocm-llvm
rocm-opencl-dev
rocm-opencl-sdk
rocm-opencl
rocm-openmp
rocm-smi-lib
rocm
rocminfo

[scryptio]

@mcmonkey4eva
The reason is that torchaudio is missing here because it tries to use the version from requirements, which requires lower torch and torchvision (and therefore other dependencies)

SwarmUI/launchtools/comfy-install-linux.sh

Line 70 in b8a731e

$python -s -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/rocm7.1

-> ComfyUI: https://github.com/comfyanonymous/ComfyUI/blob/master/README.md?plain=1#L229

Introduced with this: 0acc3a7

[madias123]

same error on windows 11 (with RX-9070XT) today

[RobbieNeko]

Can confirm issue on Linux with the 9060 XT 16GB

[madias123]

This is the error log on fresh install/update (11.12.2025, 21:00 MEZ):
After pushing "Generate" (equal what model, settings...):
"Error - ComfyUI execution error: Falscher Parameter." (=wrong parameter)

20:53:52.958 [Info] User local requested 1 image with model 'OfficialStableDiffusion/sd_xl_base_1.0.safetensors'...
20:54:04.803 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last):
20:54:04.803 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\execution.py", line 515, in execute
20:54:04.803 [Warning] [ComfyUI-0/STDERR]     output_data, output_ui, has_subgraph, has_pending_tasks = await get_output_data(prompt_id, unique_id, obj, input_data_all, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
20:54:04.804 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\execution.py", line 329, in get_output_data
20:54:04.804 [Warning] [ComfyUI-0/STDERR]     return_values = await _async_map_node_over_list(prompt_id, unique_id, obj, input_data_all, obj.FUNCTION, allow_interrupt=True, execution_block_cb=execution_block_cb, pre_execute_cb=pre_execute_cb, v3_data=v3_data)
20:54:04.804 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\execution.py", line 303, in _async_map_node_over_list
20:54:04.805 [Warning] [ComfyUI-0/STDERR]     await process_inputs(input_dict, i)
20:54:04.805 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\execution.py", line 291, in process_inputs
20:54:04.805 [Warning] [ComfyUI-0/STDERR]     result = f(**inputs)
20:54:04.806 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\nodes.py", line 346, in encode
20:54:04.806 [Warning] [ComfyUI-0/STDERR]     t = vae.encode(pixels[:,:,:,:3])
20:54:04.806 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\comfy\sd.py", line 859, in encode
20:54:04.807 [Warning] [ComfyUI-0/STDERR]     out = self.first_stage_model.encode(pixels_in).to(self.output_device).float()
20:54:04.807 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\comfy\ldm\models\autoencoder.py", line 207, in encode
20:54:04.807 [Warning] [ComfyUI-0/STDERR]     z = self.encoder(x)
20:54:04.808 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl
20:54:04.808 [Warning] [ComfyUI-0/STDERR]     return self._call_impl(*args, **kwargs)
20:54:04.808 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl
20:54:04.808 [Warning] [ComfyUI-0/STDERR]     return forward_call(*args, **kwargs)
20:54:04.809 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\comfy\ldm\modules\diffusionmodules\model.py", line 659, in forward
20:54:04.809 [Warning] [ComfyUI-0/STDERR]     h1 = conv_carry_causal_3d(x1, self.conv_in, conv_carry_in, conv_carry_out)
20:54:04.809 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\comfy\ldm\modules\diffusionmodules\model.py", line 78, in conv_carry_causal_3d
20:54:04.810 [Warning] [ComfyUI-0/STDERR]     out = op(x)
20:54:04.810 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1553, in _wrapped_call_impl
20:54:04.810 [Warning] [ComfyUI-0/STDERR]     return self._call_impl(*args, **kwargs)
20:54:04.811 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\module.py", line 1562, in _call_impl
20:54:04.811 [Warning] [ComfyUI-0/STDERR]     return forward_call(*args, **kwargs)
20:54:04.811 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\ComfyUI\comfy\ops.py", line 200, in forward
20:54:04.812 [Warning] [ComfyUI-0/STDERR]     return super().forward(*args, **kwargs)
20:54:04.812 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\conv.py", line 458, in forward
20:54:04.813 [Warning] [ComfyUI-0/STDERR]     return self._conv_forward(input, self.weight, self.bias)
20:54:04.813 [Warning] [ComfyUI-0/STDERR]   File "G:\SwarmUI\SwarmUI\dlbackend\comfy\python_embeded\lib\site-packages\torch\nn\modules\conv.py", line 454, in _conv_forward
20:54:04.813 [Warning] [ComfyUI-0/STDERR]     return F.conv2d(input, weight, bias, self.stride,
20:54:04.813 [Warning] [ComfyUI-0/STDERR] RuntimeError: Falscher Parameter.
20:54:04.814 [Warning] [ComfyUI-0/STDERR]

[RobbieNeko]

Running pip install --pre -U torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm7.1 (with a sufficiently enlarged tmpfs because torch is huge) fixed the issue for me

[madias123]

ok, errors exists anyway it "works", but how do I activate ROCM under windows for SwarmUI?
Server-> Backends _> EsxtraArgs: instead --directml using -rocm leads to nothing. what is the exact arguments for using rocm inside SwarmUI?

BTW: ComfyUI_windows_portable_amd https://github.com/comfyanonymous/ComfyUI/releases/tag/v0.4.0) works very well on my RX-9700XT, but I have no clue how to integrate it into SwarmUi

here is my actual output:

Looking in indexes: https://download.pytorch.org/whl/nightly/rocm7.1
Requirement already satisfied: torch in c:\program files\python312\lib\site-packages (2.9.0+rocmsdk20251116)
Requirement already satisfied: torchvision in c:\program files\python312\lib\site-packages (0.24.0+rocmsdk20251116)
Requirement already satisfied: torchaudio in c:\program files\python312\lib\site-packages (2.9.0+rocmsdk20251116)
Requirement already satisfied: filelock in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (3.20.0)
Requirement already satisfied: typing-extensions>=4.10.0 in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (4.15.0)
Requirement already satisfied: setuptools in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (69.0.2)
Requirement already satisfied: sympy>=1.13.3 in c:\program files\python312\lib\site-packages (from torch) (1.14.0)
Requirement already satisfied: networkx>=2.5.1 in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (3.6)
Requirement already satisfied: jinja2 in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (3.1.2)
Requirement already satisfied: fsspec>=0.8.5 in c:\users\test\appdata\roaming\python\python312\site-packages (from torch) (2025.10.0)
Requirement already satisfied: rocm==0.1.dev0 in c:\program files\python312\lib\site-packages (from rocm[libraries]==0.1.dev0->torch) (0.1.dev0)
Requirement already satisfied: rocm-sdk-core==0.1.dev0 in c:\program files\python312\lib\site-packages (from rocm==0.1.dev0->rocm[libraries]==0.1.dev0->torch) (0.1.dev0)
Requirement already satisfied: rocm-sdk-libraries-custom==0.1.dev0 in c:\program files\python312\lib\site-packages (from rocm[libraries]==0.1.dev0->torch) (0.1.dev0)
Requirement already satisfied: numpy in c:\users\test\appdata\roaming\python\python312\site-packages (from torchvision) (1.26.2)
Requirement already satisfied: pillow!=8.3.*,>=5.3.0 in c:\users\test\appdata\roaming\python\python312\site-packages (from torchvision) (10.1.0)
Requirement already satisfied: mpmath<1.4,>=1.1.0 in c:\users\test\appdata\roaming\python\python312\site-packages (from sympy>=1.13.3->torch) (1.3.0)
Requirement already satisfied: MarkupSafe>=2.0 in c:\users\test\appdata\roaming\python\python312\site-packages (from jinja2->torch) (2.1.3)

[madias123]

Okay, I solved it the hard way:

Download the latest ComfyUI (portable) version from:
https://docs.comfy.org/installation/comfyui_portable_windows
In the directory: xxxx:\SwarmUI\SwarmUI\dlbackend
replaced the entire comfy folder
Removed all arguments and auto-updates in the "Backends"

[scryptio]

snip random hostility

Automated Install

• Make sure prerequisites are met
• wget https://github.com/mcmonkeyprojects/SwarmUI/releases/download/0.6.5-Beta/install-linux.sh -O install-linux.sh
• chmod +x install-linux.sh
• ./install-linux.sh
• Once you get the error, terminate SwarmUI
• cd SwarmUI
• nano launchtools/comfy-install-linux.sh and edit line 70: Add torchaudio to the amd section
• rm -rf dlbackend/
• bash launchtools/comfy-install-linux.sh amd
• ./launch-linux.sh

Manual Install
Clone the repository and edit line 70

• nano launchtools/comfy-install-linux.sh and edit line 70: Add torchaudio to the amd section
• Continue manual install

Feel free to create your own virtual environment or don't use a virtual environment using install args in combination with i.e. conda.