Installation error: missing fbgemm.dll #363
Description
When running install.bat on Windows with python version 3.11.9 and a 50 series card.
Are you using this locally? (y/n): y
Submodule 'backend' (https://github.com/derrian-distro/LoRA_Easy_Training_scripts_Backend) registered for path 'backend'
Cloning into 'F:/repos/LoRA_Easy_Training_Scripts/backend'...
Submodule path 'backend': checked out '162e56de18e4d4ab96b066380606d657a68c66fa'
Submodule 'LyCORIS' (https://github.com/67372a/LyCORIS) registered for path 'lycoris'
Submodule 'sd_scripts' (https://github.com/kohya-ss/sd-scripts) registered for path 'sd_scripts'
Cloning into 'F:/repos/LoRA_Easy_Training_Scripts/backend/lycoris'...
Cloning into 'F:/repos/LoRA_Easy_Training_Scripts/backend/sd_scripts'...
Submodule path 'lycoris': checked out 'ca9f47d238bb67266acd7354ce31a72eea37bdd2'
Submodule path 'sd_scripts': checked out '5a18a03ffcc2a21c6e884a25d041076911a79a2a'
setting execution policy to unrestricted
creating venv and installing requirements
NVIDIA GeForce RTX 5090
Looking in indexes: https://download.pytorch.org/whl/nightly/cu128
Collecting torch
Downloading https://download.pytorch.org/whl/nightly/cu128/torch-2.12.0.dev20260303%2Bcu128-cp311-cp311-win_amd64.whl.metadata (32 kB)
Collecting torchvision
Downloading https://download-r2.pytorch.org/whl/nightly/cu128/torchvision-0.26.0.dev20260221%2Bcu128-cp311-cp311-win_amd64.whl.metadata (5.6 kB)
Collecting filelock (from torch)
Downloading filelock-3.20.3-py3-none-any.whl.metadata (2.1 kB)
Collecting typing-extensions>=4.10.0 (from torch)
Downloading https://download.pytorch.org/whl/nightly/typing_extensions-4.15.0-py3-none-any.whl.metadata (3.3 kB)
Requirement already satisfied: setuptools<82 in f:\repos\lora_easy_training_scripts\backend\sd_scripts\venv\lib\site-packages (from torch) (65.5.0)
Collecting sympy>=1.13.3 (from torch)
Using cached sympy-1.14.0-py3-none-any.whl.metadata (12 kB)
Collecting networkx>=2.5.1 (from torch)
Using cached networkx-3.6.1-py3-none-any.whl.metadata (6.8 kB)
Collecting jinja2 (from torch)
Downloading https://download.pytorch.org/whl/nightly/jinja2-3.1.6-py3-none-any.whl.metadata (2.9 kB)
Collecting fsspec>=0.8.5 (from torch)
Using cached fsspec-2026.2.0-py3-none-any.whl.metadata (10 kB)
Collecting numpy (from torchvision)
Downloading numpy-2.4.2-cp311-cp311-win_amd64.whl.metadata (6.6 kB)
Collecting pillow!=8.3.*,>=5.3.0 (from torchvision)
Downloading pillow-12.1.0-cp311-cp311-win_amd64.whl.metadata (9.0 kB)
Collecting mpmath<1.4,>=1.1.0 (from sympy>=1.13.3->torch)
Using cached mpmath-1.3.0-py3-none-any.whl.metadata (8.6 kB)
Collecting MarkupSafe>=2.0 (from jinja2->torch)
Downloading https://download.pytorch.org/whl/nightly/MarkupSafe-3.0.2-cp311-cp311-win_amd64.whl.metadata (4.1 kB)
Downloading https://download.pytorch.org/whl/nightly/cu128/torch-2.12.0.dev20260303%2Bcu128-cp311-cp311-win_amd64.whl (2796.9 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 2.8/2.8 GB 4.3 MB/s eta 0:00:00
Downloading https://download-r2.pytorch.org/whl/nightly/cu128/torchvision-0.26.0.dev20260221%2Bcu128-cp311-cp311-win_amd64.whl (9.5 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 9.5/9.5 MB 35.7 MB/s eta 0:00:00
Using cached fsspec-2026.2.0-py3-none-any.whl (202 kB)
Using cached networkx-3.6.1-py3-none-any.whl (2.1 MB)
Downloading pillow-12.1.0-cp311-cp311-win_amd64.whl (7.0 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 7.0/7.0 MB 30.0 MB/s eta 0:00:00
Using cached sympy-1.14.0-py3-none-any.whl (6.3 MB)
Downloading https://download.pytorch.org/whl/nightly/typing_extensions-4.15.0-py3-none-any.whl (44 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 44.6/44.6 kB ? eta 0:00:00
Downloading filelock-3.20.3-py3-none-any.whl (16 kB)
Downloading https://download.pytorch.org/whl/nightly/jinja2-3.1.6-py3-none-any.whl (134 kB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 134.9/134.9 kB ? eta 0:00:00
Downloading numpy-2.4.2-cp311-cp311-win_amd64.whl (12.6 MB)
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 12.6/12.6 MB 40.9 MB/s eta 0:00:00
Downloading https://download.pytorch.org/whl/nightly/MarkupSafe-3.0.2-cp311-cp311-win_amd64.whl (15 kB)
Using cached mpmath-1.3.0-py3-none-any.whl (536 kB)
Installing collected packages: mpmath, typing-extensions, sympy, pillow, numpy, networkx, MarkupSafe, fsspec, filelock, jinja2, torch, torchvision
Successfully installed MarkupSafe-3.0.2 filelock-3.20.3 fsspec-2026.2.0 jinja2-3.1.6 mpmath-1.3.0 networkx-3.6.1 numpy-2.4.2 pillow-12.1.0 sympy-1.14.0 torch-2.12.0.dev20260303+cu128 torchvision-0.26.0.dev20260221+cu128 typing-extensions-4.15.0
50 series GPU doesn't have xformers support!
Checking if installed pytorch has libomp issue...
Traceback (most recent call last):
File "F:\repos\LoRA_Easy_Training_Scripts\backend\fix_torch.py", line 19, in <module>
with open(test_file, "rb") as f:
^^^^^^^^^^^^^^^^^^^^^
FileNotFoundError: [Errno 2] No such file or directory: 'F:\\repos\\LoRA_Easy_Training_Scripts\\backend\\sd_scripts\\venv\\Lib\\site-packages\\torch\\lib\\fbgemm.dll'
Traceback (most recent call last):
File "F:\repos\LoRA_Easy_Training_Scripts\backend\installer.py", line 218, in <module>
main()
File "F:\repos\LoRA_Easy_Training_Scripts\backend\installer.py", line 211, in main
setup_venv(pip)
File "F:\repos\LoRA_Easy_Training_Scripts\backend\installer.py", line 123, in setup_venv
subprocess.check_call("venv\\Scripts\\python.exe ..\\fix_torch.py")
File "C:\Users\night\.pyenv\pyenv-win\versions\3.11.9\Lib\subprocess.py", line 413, in check_call
raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command 'venv\Scripts\python.exe ..\fix_torch.py' returned non-zero exit status 1.
Traceback (most recent call last):
File "F:\repos\LoRA_Easy_Training_Scripts\install.py", line 60, in <module>
main()
File "F:\repos\LoRA_Easy_Training_Scripts\install.py", line 56, in main
subprocess.check_call(f"{python} installer.py local", shell=sys.platform == "linux")
File "C:\Users\night\.pyenv\pyenv-win\versions\3.11.9\Lib\subprocess.py", line 413, in check_call
raise CalledProcessError(retcode, cmd)
subprocess.CalledProcessError: Command 'C:\Users\night\.pyenv\pyenv-win\versions\3.11.9\python.exe installer.py local' returned non-zero exit status 1.
Backend refuses to install. Here are the contents of that directory F:\\repos\\LoRA_Easy_Training_Scripts\\backend\\sd_scripts\\venv\\Lib\\site-packages\\torch\\lib\\ for debugging purposes:
ls F:\\repos\\LoRA_Easy_Training_Scripts\\backend\\sd_scripts\\venv\\Lib\\site-packages\\torch\\lib\\
Directory: F:\repos\LoRA_Easy_Training_Scripts\backend\sd_scripts\venv\Lib\site-packages\torch\lib
Mode LastWriteTime Length Name
---- ------------- ------ ----
d---- 3/3/2026 7:38 PM libshm
d---- 3/3/2026 7:38 PM libshm_windows
-a--- 3/3/2026 7:38 PM 1924 _C.lib
-a--- 3/3/2026 7:38 PM 408576 c10_cuda.dll
-a--- 3/3/2026 7:38 PM 59324 c10_cuda.lib
-a--- 3/3/2026 7:38 PM 1086976 c10.dll
-a--- 3/3/2026 7:38 PM 829698 c10.lib
-a--- 3/3/2026 7:38 PM 17920 caffe2_nvrtc.dll
-a--- 3/3/2026 7:38 PM 1850 caffe2_nvrtc.lib
-a--- 3/3/2026 7:38 PM 117988 cpuinfo.lib
-a--- 3/3/2026 7:38 PM 113716224 cublas64_12.dll
-a--- 3/3/2026 7:38 PM 674667520 cublasLt64_12.dll
-a--- 3/3/2026 7:38 PM 573952 cudart64_12.dll
-a--- 3/3/2026 7:38 PM 282445872 cudnn_adv64_9.dll
-a--- 3/3/2026 7:38 PM 4618272 cudnn_cnn64_9.dll
-a--- 3/3/2026 7:38 PM 513926688 cudnn_engines_precompiled64_9.dll
-a--- 3/3/2026 7:38 PM 20201008 cudnn_engines_runtime_compiled64_9.dll
-a--- 3/3/2026 7:38 PM 2420256 cudnn_graph64_9.dll
-a--- 3/3/2026 7:38 PM 56823328 cudnn_heuristic64_9.dll
-a--- 3/3/2026 7:38 PM 126508576 cudnn_ops64_9.dll
-a--- 3/3/2026 7:38 PM 266288 cudnn64_9.dll
-a--- 3/3/2026 7:38 PM 276121600 cufft64_11.dll
-a--- 3/3/2026 7:38 PM 163328 cufftw64_11.dll
-a--- 3/3/2026 7:38 PM 4483664 cupti64_2025.1.1.dll
-a--- 3/3/2026 7:38 PM 71955968 curand64_10.dll
-a--- 3/3/2026 7:38 PM 225683456 cusolver64_11.dll
-a--- 3/3/2026 7:38 PM 157071360 cusolverMg64_11.dll
-a--- 3/3/2026 7:38 PM 379535872 cusparse64_12.dll
-a--- 3/3/2026 7:38 PM 1457288 fmt.lib
-a--- 3/3/2026 7:38 PM 1614184 libiomp5md.dll
-a--- 3/3/2026 7:38 PM 43880 libiompstubs5md.dll
-a--- 3/3/2026 7:38 PM 287624 libittnotify.lib
-a--- 3/3/2026 7:38 PM 2885048 microkernels-prod.lib
-a--- 3/3/2026 7:38 PM 77860352 nvJitLink_120_0.dll
-a--- 3/3/2026 7:38 PM 21641792 nvperf_host.dll
-a--- 3/3/2026 7:38 PM 6356480 nvrtc-builtins64_128.dll
-a--- 3/3/2026 7:38 PM 86794240 nvrtc64_120_0.alt.dll
-a--- 3/3/2026 7:38 PM 86728192 nvrtc64_120_0.dll
-a--- 3/3/2026 7:38 PM 48128 nvToolsExt64_1.dll
-a--- 3/3/2026 7:38 PM 185288 pthreadpool.lib
-a--- 3/3/2026 7:38 PM 15360 shm.dll
-a--- 3/3/2026 7:38 PM 3790 shm.lib
-a--- 3/3/2026 7:38 PM 8811892 sleef.lib
-a--- 3/3/2026 7:38 PM 266793472 torch_cpu.dll
-a--- 3/3/2026 7:38 PM 29671498 torch_cpu.lib
-a--- 3/3/2026 7:38 PM 800235008 torch_cuda.dll
-a--- 3/3/2026 7:38 PM 1815822 torch_cuda.lib
-a--- 3/3/2026 7:38 PM 9728 torch_global_deps.dll
-a--- 3/3/2026 7:38 PM 19086336 torch_python.dll
-a--- 3/3/2026 7:38 PM 289248 torch_python.lib
-a--- 3/3/2026 7:38 PM 9728 torch.dll
-a--- 3/3/2026 7:38 PM 1832 torch.lib
-a--- 3/3/2026 7:38 PM 195072 uv.dll
-a--- 3/3/2026 7:38 PM 2702626 XNNPACK.lib
-a--- 3/3/2026 7:38 PM 89088 zlibwapi.dll
As can be seen, fbgemm.dll is not there.
The issue is upstream:
https://github.com/derrian-distro/LoRA_Easy_Training_scripts_Backend/blob/162e56de18e4d4ab96b066380606d657a68c66fa/fix_torch.py#L14
A workaround is to comment out the with block on lines 19-27. Installation will proceed without it, and UI/training appears to work fine. Recommended to patch this one.