I am loading a dll in python using following code:
if os.path.exists(dll_path):
my_dll = ctypes.cdll.LoadLibrary(dll_path)
But I am continuously getting the following error
WindowsError: [Error 126] The specified module could not be found
dll is present at the specified path, but I didn't understand why I'm getting the error.
On the off chance anyone else ever runs into this extremely specific issue..
Something inside PyTorch breaks DLL loading. Once you run import torch, any further DLL loads will fail. So if you're using PyTorch and loading your own DLLs you'll have to rearrange your code to import all DLLs first. Confirmed w/ PyTorch 1.5.0 on Python 3.7
If you are using GCC to compile it for Windows, it's possible that the error is because dependent libraries can't be found.
Using the -static flag if linking with GCC might fix that.
Tried to specify dll path in different ways (proposed by @markm), but nothing has worked for me. Fixed the problem by copying dll into script folder. It's not a good solution, but ok for my purposes.
for me install Microsoft Visual C++ 2015 Redistributable Update 3 from https://www.microsoft.com/en-us/download/details.aspx?id=53587 solved it.
if you come across this error when you try running PyTorch related libraries you may have to consider installing PyTorch with CPU only version i.e. if you don't have Nvidia GPU in your system.
Pytorch with CUDA worked in Nvidia installed systems but not in others.
There is a promising answer at Problem updating bokeh: [WinError 126] The specified module could not be found.
It hints at https://github.com/conda/conda/issues/9313.
There, you find:
It's a library load issue. More details at github.com/conda/conda/issues/8836 You probably have a broken conda right now. You can use a standalone conda from repo.anaconda.com/pkgs/misc/conda-execs to repair it: standalone-conda.exe update -p C:\ProgramData\Anaconda3 conda-package-handling You should get version 1.6.0, and the problems should go away.
Thus, it might simply be a conda issue. Reinstalling standalone conda might repair the error. Please comment whoever can confirm this.
problem solved for me. I changed version from pytorch=1.5.1 to pytorch=1.4 and typed the below command in anaconda prompt window
conda install pytorch==1.4.0 torchvision==0.5.0 -c pytorch
NestedCaveats solution worked for me.
Imported my .dll files before importing torch and gpytorch, and all went smoothly.
So I just want to add that its not just importing pytorch but I can confirm that torch and gpytorch have this issue as well. I'd assume it covers any other torch-related libraries.
This is probably because a runtime dependency of one of the DLLs was not found on your system. I think that the expected Microsoft Visual C runtime DLL is missing from your system.
Install this:
https://www.microsoft.com/en-US/download/details.aspx?id=40784