Cannot load model: ggml.dll WinError 127 after installing llama_cpp_python CUDA build

[Aski-d]

Hi, I'm very new to this, and I recently installed both of these versions:

llama_cpp_python-0.3.40+cu124-cp312-cp312-win_amd64 and also llama_cpp_python-0.3.45+cu124-cp312-cp312-win_amd64

My setup:

CUDA: 12.4
Python: 3.12.7 (the embedded Python included with ComfyUI)
OS: Windows
ComfyUI portable installation

Previously, I was using llama_cpp_python 0.3.4 (auto-installed by another extension), and it worked normally. However, it seems that it cannot run with Qwen 3.6.

I tried manually installing llama_cpp_python-0.3.40 & 0.3.45 using pip, and I checked the installed version with:

PS D:\ComfyUI_windows_portable_nvidia> python_embeded\python.exe -c "import llama_cpp; print(llama_cpp.version)"

However, I got this error related to ggml:

PS D:\ComfyUI_windows_portable_nvidia> python_embeded\python.exe -c "import llama_cpp; print(llama_cpp.__version__)"
Traceback (most recent call last):
  File "<string>", line 1, in <module>
  File "D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\__init__.py", line 1, in <module>
    from .llama_cpp import *
  File "D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\llama_cpp.py", line 8, in <module>
    from ._ggml import (
  File "D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\_ggml.py", line 30, in <module>
    libggml = load_shared_library("ggml", libggml_base_paths)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\_ctypes_extensions.py", line 149, in load_shared_library
    raise RuntimeError(
RuntimeError: Failed to load 'ggml' from [WindowsPath('D:/ComfyUI_windows_portable_nvidia/python_embeded/Lib/site-packages/llama_cpp/lib'), WindowsPath('D:/ComfyUI_windows_portable_nvidia/python_embeded/Lib/site-packages/llama_cpp/bin')]
D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\bin\ggml.dll: [WinError 127] The specified procedure could not be found
D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\lib\ggml.dll: [WinError 127] The specified procedure could not be found
D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\bin\ggml.dll: [WinError 127] The specified procedure could not be found
Library search path contents:
D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\lib:
  - ggml-base.dll
  - ggml-cpu-alderlake.dll
  - ggml-cpu-cannonlake.dll
  - ggml-cpu-cascadelake.dll
  - ggml-cpu-cooperlake.dll
  - ggml-cpu-haswell.dll
  - ggml-cpu-icelake.dll
  - ggml-cpu-ivybridge.dll
  - ggml-cpu-piledriver.dll
  - ggml-cpu-sandybridge.dll
  - ggml-cpu-sapphirerapids.dll
  - ggml-cpu-skylakex.dll
  - ggml-cpu-sse42.dll
  - ggml-cpu-x64.dll
  - ggml-cpu-zen4.dll
  - ggml-cuda.dll
  - ggml.dll
  - libomp140.x86_64.dll
  - llama-common.dll
  - llama.dll
  - mtmd.dll
D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\bin:
  - ggml-base.dll
  - ggml-cpu-alderlake.dll
  - ggml-cpu-cannonlake.dll
  - ggml-cpu-cascadelake.dll
  - ggml-cpu-cooperlake.dll
  - ggml-cpu-haswell.dll
  - ggml-cpu-icelake.dll
  - ggml-cpu-ivybridge.dll
  - ggml-cpu-piledriver.dll
  - ggml-cpu-sandybridge.dll
  - ggml-cpu-sapphirerapids.dll
  - ggml-cpu-skylakex.dll
  - ggml-cpu-sse42.dll
  - ggml-cpu-x64.dll
  - ggml-cpu-zen4.dll
  - ggml-cuda.dll
  - ggml.dll
  - llama-common.dll
  - llama.dll
  - mtmd.dll

I also get a Windows popup saying that an entry point could not be found.

I'm not very familiar with the technical details, so I would really appreciate any help.

Thank you very much for any advice.

UPDATED: Version 0.3.38 works.

[JamePeng]

Try deleting the folder D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\bin, or copy libomp140.x86_64.dll from the lib folder into this bin folder.

[Aski-d]

Try deleting the folder D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\bin, or copy libomp140.x86_64.dll from the lib folder into this bin folder.

I tried copying the files from the lib folder into the bin folder, but it still didn't work. I noticed that the pop-up error disappeared, and I was able to launch the ComfyUI web interface using PowerShell. However, the encoder still couldn't find ggml.

In the end, I downgraded to llama_cpp_python 0.3.38, and everything worked without any issues.

Thank you for your response and for taking the time to help.

[JamePeng]

@Aski-d You can also copy libomp140.x86_64.dll from the lib directory to C:\Windows\System32. The main issue is that your system has an OpenMP library other than the VC143 version.

Update: You could try replacing _ggml.py to see if that works. （path: D:\ComfyUI_windows_portable_nvidia\python_embeded\Lib\site-packages\llama_cpp\_ggml.py）
62d3ae5