Error while installing Nunchaku

[Michoko92]

Hi,

I'm trying to use an SVDQuant model with Nunchaku. When I select the model, SwarmUI offers me to install the support files. I launch the install, but it fails. In the console, I get those errors :

14:35:02.727 [Warning] [ComfyUI-0/STDERR] Traceback (most recent call last):
14:35:02.738 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\dlbackend\comfy\ComfyUI\nodes.py", line 2124, in load_custom_node
14:35:02.762 [Warning] [ComfyUI-0/STDERR]     module_spec.loader.exec_module(module)
14:35:02.768 [Warning] [ComfyUI-0/STDERR]   File "<frozen importlib._bootstrap_external>", line 940, in exec_module
14:35:02.774 [Warning] [ComfyUI-0/STDERR]   File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
14:35:02.803 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\DLNodes\ComfyUI-nunchaku\__init__.py", line 2, in <module>
14:35:02.804 [Warning] [ComfyUI-0/STDERR]     from .nodes.lora import NunchakuFluxLoraLoader
14:35:02.804 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\DLNodes\ComfyUI-nunchaku\nodes\lora\__init__.py", line 1, in <module>
14:35:02.804 [Warning] [ComfyUI-0/STDERR]     from .flux import NunchakuFluxLoraLoader
14:35:02.805 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\src\BuiltinExtensions\ComfyUIBackend\DLNodes\ComfyUI-nunchaku\nodes\lora\flux.py", line 6, in <module>
14:35:02.805 [Warning] [ComfyUI-0/STDERR]     from nunchaku.lora.flux import to_diffusers
14:35:02.806 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\nunchaku\__init__.py", line 1, in <module>
14:35:02.806 [Warning] [ComfyUI-0/STDERR]     from .models import NunchakuFluxTransformer2dModel, NunchakuSanaTransformer2DModel, NunchakuT5EncoderModel
14:35:02.807 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\nunchaku\models\__init__.py", line 1, in <module>
14:35:02.807 [Warning] [ComfyUI-0/STDERR]     from .text_encoders.t5_encoder import NunchakuT5EncoderModel
14:35:02.807 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\nunchaku\models\text_encoders\t5_encoder.py", line 9, in <module>
14:35:02.808 [Warning] [ComfyUI-0/STDERR]     from .linear import W4Linear
14:35:02.808 [Warning] [ComfyUI-0/STDERR]   File "D:\AI\StableSwarmUI\dlbackend\comfy\python_embeded\Lib\site-packages\nunchaku\models\text_encoders\linear.py", line 8, in <module>
14:35:02.809 [Warning] [ComfyUI-0/STDERR]     from ..._C.ops import gemm_awq, gemv_awq
14:35:02.809 [Warning] [ComfyUI-0/STDERR] ImportError: DLL load failed while importing _C: La procédure spécifiée est introuvable.

(last message in French means that the specified procedure can't be found)

Would you have some suggestions, please ? Thank you!

[mcmonkey4eva]

This is probably something you'll have to ask on the nunchaku github

[Michoko92]

Thank you for your reply. Actually, I had to install the wheel and other files manually, and it seems to work now. I have two extra questions, please:

• Even if generating with the SVDQ models is currently working in my SwarmUI install, I'm not completely sure everything is installed properly. Is there a way to reset the nunchaku install in SwarmUI, so it tries again to install it when using an SVDQ model for the first time? I'd like to go though the regular install process again, to be sure everything's there.

• I'm seeing new models types posted a bit everywhere, using the new _r32 format (which is a single file, and is advertised as much better). Is there a way to use those models in SwarmUI? (example : https://civitai.com/models/646328/flux1-dev-fp8andnf4andgguf-6-steps-hybrid-4-steps-svdquant-int4-flux1-dev-lora-models)

Thank you for your amazing work, SwarmUI is definitely my favorite UI!