`newbyteorder` was removed from the ndarray class in NumPy 2.0.

[leszekhanusz]

I got the following error:

Error occurred when executing UnetLoaderGGUF:

`newbyteorder` was removed from the ndarray class in NumPy 2.0. Use `arr.view(arr.dtype.newbyteorder(order))` instead.

 File "/mnt/tera/git-repos/ComfyUI/execution.py", line 152, in recursive_execute
 output_data, output_ui = get_output_data(obj, input_data_all)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 File "/mnt/tera/git-repos/ComfyUI/execution.py", line 82, in get_output_data
 return_values = map_node_over_list(obj, input_data_all, obj.FUNCTION, allow_interrupt=True)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 File "/mnt/tera/git-repos/ComfyUI/execution.py", line 75, in map_node_over_list
 results.append(getattr(obj, func)(**slice_dict(input_data_all, i)))
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 File "/mnt/tera/git-repos/ComfyUI/custom_nodes/ComfyUI-GGUF/nodes.py", line 53, in load_unet
 sd = gguf_sd_loader(unet_path)
 ^^^^^^^^^^^^^^^^^^^^^^^^^
 File "/mnt/tera/git-repos/ComfyUI/custom_nodes/ComfyUI-GGUF/nodes.py", line 19, in gguf_sd_loader
 reader = gguf.GGUFReader(path)
 ^^^^^^^^^^^^^^^^^^^^^
 File "/home/leszek/miniconda3/envs/comfyui/lib/python3.11/site-packages/gguf/gguf_reader.py", line 94, in __init__
 if self._get(offs, np.uint32, override_order = '<')[0] != GGUF_MAGIC:
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 File "/home/leszek/miniconda3/envs/comfyui/lib/python3.11/site-packages/gguf/gguf_reader.py", line 151, in _get
 .newbyteorder(override_order or self.byte_order)
 ^^^^^^^^^^^^

It seems to be this bug

As a very temporary workaround, you can simply delete the .newbyteorder(override_order or self.byte_order) line in site-packages/gguf/gguf_reader.py and it should work.

[jetjodh]

I think you should rollback your numpy version to <2.0.0

[M1kep]

This seems to break with a fresh Comfy install

[superskirv]

I ran this to fix the issue.

Linux Mint(Ubuntu 22.04), 64gb ram, 3090ti

pip install numpy==1.26.3

[city96]

I just ran into this too, sorry I forgot about this issue. I've added numpy<2.0.0 to the requirements which should hopefully fix it when installing.

[erasurecom]

Those who want to stay on NumPy>=2.0, e.g. to avoid "Dependency Hell", could change the code in gguf_reader.py file:
ggml-org/llama.cpp#9021 (comment)

Works stable on ComfyUI with GGUF loaders e.g. "Unet Loader (GGUF)" and "DualCLIPLoader (GGUF)" nodes in the workflows:

### Loading: ComfyUI-Manager (V2.51.2)
### ComfyUI Revision: 2737 [d985d1d7] | Released on '2024-10-02'

Import times for custom nodes:
0.0 seconds: E:\NEURAL\ComfyUI\Current\ComfyUI\custom_nodes\websocket_image_save.py
0.0 seconds: E:\NEURAL\ComfyUI\Current\ComfyUI\custom_nodes\ComfyUI-GGUF
0.2 seconds: E:\NEURAL\ComfyUI\Current\ComfyUI\custom_nodes\ComfyUI-Manager

The file located in "your_python_venv_dir\Lib\site-packages\gguf\gguf_reader.py".

[m18coppola]

upstream fix is waiting to be merged: link

[Xeverian]

I just ran into this too, sorry I forgot about this issue. I've added numpy<2.0.0 to the requirements which should hopefully fix it when installing.

doesn't seem to work for me. After updating comfyUI today it got broken. I've tried to remove the node and install it again but to no avail

[metamountain]

how do i downgrade in the standalone ? have a similar problem ....

AttributeError: newbyteorder was removed from the ndarray class in NumPy 2.0. Use arr.view(arr.dtype.newbyteorder(order)) instead.

[Xeverian]

how do i downgrade in the standalone ? have a similar problem ....

AttributeError: newbyteorder was removed from the ndarray class in NumPy 2.0. Use arr.view(arr.dtype.newbyteorder(order)) instead.

just follow this answer. It helped me fix this
#7 (comment)

[JumpLink]

Those who want to stay on NumPy>=2.0, e.g. to avoid "Dependency Hell", could change the code in gguf_reader.py file: ggerganov/llama.cpp#9021 (comment)

...

For Linux users, the file is located at ~/.local/lib/pythonX.X/site-packages/gguf/gguf_reader.py
(replace pythonX.X with your Python version, in my case python3.12).
And this is what the change should look like: https://github.com/ggerganov/llama.cpp/pull/9772/files