Failed with pytorch 2.8.0 and cu128

[madcrou]

Hello Community! Onetrainer stopped working and I decided to do a clean installation, now Onetrainer does not see pytorch and cuda.

---

_ERROR: Could not find a version that satisfies the requirement torch==2.8.0+cu128 (from versions: 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1, 2.5.0, 2.5.1, 2.6.0, 2.7.0, 2.7.1, 2.8.0, 2.9.0, 2.9.1)
ERROR: No matching distribution found for torch==2.8.0+cu128

ERROR: Dependencies install failed_

his appears at the end of the Onetrainer installation, I changed different versions of python (always installed between 3.10 and 3.12), always installed the 64 bit version,
rearranged pytorch with the command "pip install torch==2.8.0 torchvision==0.23.0 torcaudio==2.8.0" - - pytorch and components were installed correctly, but Onetrainer still doesn’t see it!

I tried to run the bat file to create a bug log, but alas:

---

activating venv C:\OneTrainer\venv
Using Python "C:\OneTrainer\venv\Scripts\python.exe" -X utf8
Generating debug report...
Traceback (most recent call last):
File "C:\OneTrainer\scripts\generate_debug_report.py", line 19, in
from deepdiff import DeepDiff
ModuleNotFoundError: No module named 'deepdiff'
Error: Debug report generation failed with code 1

Thanks in advance for your help! It might be simple, but I don't understand something. I tried a bunch of solutions from the Internet, but all to no avail.

[O-J1]

ModuleNotFoundError: No module named 'deepdiff'

Pretty sure you did not run install.bat, this all gets installed when you:

1. nuke your venv or conda
2. Run install.bat

[madcrou]

The result is the same. Onetrainer used to work on this computer, at some point it simply stopped updating with this error. I decided to reinstall and now I have this. I made a clone again, reainstalled Python and Pytorch again - the same errors

[madcrou]

Searching for a suitable Python installation...
Step 1: Checking for Python in PATH (to support Conda installs)...
  Testing Python from PATH: "C:\Program Files\Python311\python.exe"
  SELECTED Python from PATH: "C:\Program Files\Python311\python.exe"

Using Python: C:\Program Files\Python311\python.exe

Managing virtual environment...
Creating venv at "C:\OneTrainer\\venv"...
Activating virtual environment...
Virtual environment activated.
Checking for Tkinter availability...
Tkinter is available, proceeding ...

Upgrading pip and installing dependencies from requirements.txt...
Executing: python -m pip install --upgrade pip
Requirement already satisfied: pip in c:\onetrainer\venv\lib\site-packages (23.2.1)
Collecting pip
  Obtaining dependency information for pip from https://files.pythonhosted.org/packages/44/3c/d717024885424591d5376220b5e836c2d5293ce2011523c9de23ff7bf068/pip-25.3-py3-none-any.whl.metadata
  Using cached pip-25.3-py3-none-any.whl.metadata (4.7 kB)
Using cached pip-25.3-py3-none-any.whl (1.8 MB)
Installing collected packages: pip
  Attempting uninstall: pip
    Found existing installation: pip 23.2.1
    Uninstalling pip-23.2.1:
      Successfully uninstalled pip-23.2.1
Successfully installed pip-25.3
Executing: python -m pip install -r requirements.txt
Looking in indexes: https://pypi.org/simple, https://download.pytorch.org/whl/cu128
Ignoring nvidia-nccl-cu12: markers 'sys_platform == "linux"' don't match your environment
Obtaining diffusers from git+https://github.com/huggingface/diffusers.git@256e010#egg=diffusers (from -r C:\OneTrainer\requirements-global.txt (line 23))
  Cloning https://github.com/huggingface/diffusers.git (to revision 256e010) to c:\onetrainer\venv\src\diffusers
  Running command git clone --filter=blob:none --quiet https://github.com/huggingface/diffusers.git 'C:\OneTrainer\venv\src\diffusers'
  Did not find branch or tag '256e010', assuming revision or ref.
  Running command git checkout -q 256e010
  Resolved https://github.com/huggingface/diffusers.git to commit 256e010
  Installing build dependencies ... done
  Checking if build backend supports build_editable ... done
  Getting requirements to build editable ... done
  Preparing editable metadata (pyproject.toml) ... done
Obtaining mgds from git+https://github.com/Nerogar/mgds.git@385578f#egg=mgds (from -r C:\OneTrainer\requirements-global.txt (line 35))
  Cloning https://github.com/Nerogar/mgds.git (to revision 385578f) to c:\onetrainer\venv\src\mgds
  Running command git clone --filter=blob:none --quiet https://github.com/Nerogar/mgds.git 'C:\OneTrainer\venv\src\mgds'
  Did not find branch or tag '385578f', assuming revision or ref.
  Running command git checkout -q 385578f
  Resolved https://github.com/Nerogar/mgds.git to commit 385578f
  Installing build dependencies ... done
  Checking if build backend supports build_editable ... done
  Getting requirements to build editable ... done
  Preparing editable metadata (pyproject.toml) ... done
Obtaining muon-optimizer from git+https://github.com/KellerJordan/Muon.git@f90a42b#egg=muon-optimizer (from -r C:\OneTrainer\requirements-global.txt (line 45))
  Cloning https://github.com/KellerJordan/Muon.git (to revision f90a42b) to c:\onetrainer\venv\src\muon-optimizer
  Running command git clone --filter=blob:none --quiet https://github.com/KellerJordan/Muon.git 'C:\OneTrainer\venv\src\muon-optimizer'
  Did not find branch or tag 'f90a42b', assuming revision or ref.
  Running command git checkout -q f90a42b
  Resolved https://github.com/KellerJordan/Muon.git to commit f90a42b
  Installing build dependencies ... done
  Checking if build backend supports build_editable ... done
  Getting requirements to build editable ... done
  Preparing editable metadata (pyproject.toml) ... done
ERROR: Could not find a version that satisfies the requirement torch==2.8.0+cu128 (from versions: 2.0.0, 2.0.1, 2.1.0, 2.1.1, 2.1.2, 2.2.0, 2.2.1, 2.2.2, 2.3.0, 2.3.1, 2.4.0, 2.4.1, 2.5.0, 2.5.1, 2.6.0, 2.7.0, 2.7.1, 2.8.0, 2.9.0, 2.9.1)
ERROR: No matching distribution found for torch==2.8.0+cu128

ERROR: Dependencies install failed

Press any key to continue . . .

Checking CUDA availability...
Traceback (most recent call last):
  File "<string>", line 1, in <module>
ModuleNotFoundError: No module named 'torch'
CUDA not found via torch.cuda.is_available.
AMD GPU? (y/n): n

ERROR: CUDA unavailable and not an AMD GPU setup - aborting. Please check PyTorch and NVIDIA driver compatibility.

Press any key to continue . . .

[O-J1]

At this point I've tried everything I can think of on my end to reproduce it, no dice.

Couple of thoughts:

• Torch is available for Python 3.11 + your GPU with CUDA 12.8, so no clue why its saying its not
• Its not a pip issue because pip automatically gets upgraded and it was successful
• You have python in program files, this is non standard and there was issues in the past with this
• We fixed the actual other concurrent issue which was the av library yanking binarys for some reason.

All I can suggest for now is

1. uninstalling python 3.11
2. downloading and installing python 3.12.10 or 3.12.9
3. Making sure to not "Install for all users" nor add for all users
4. git clone https://github.com/Nerogar/OneTrainer.git
5. Then once again try running install.bat

I have no estimation of whether this will or wont solve it, but we've tested on multiple different machines now, macs, linux, windows 11 and 10, cant recreate 😕

[madcrou]

Today I didn't wait for your answer and reinstalled Windows - new clean installation, version 25H2 pro, new git, pytorch, etc. The result is the same and the errors are the same. After I saw your answer, I reinstalled Python, pytorch with components and followed all your instructions again - same errors

[carlo0000]

do you have an older gpu? cu128 is the older one right ?

then install an older onetrainer (before the new pytorch update)

because me i can't update sins last week, my gpu is 1080ti and if i update it won't run anymore
i use stable matrix , so not sure it s a stable matrix issue or onetrainer , (because it share folder) and when i lauch comfyui i got a warning also but then it say it found an other version of pytorch and continue working

i force install manually old pytorch version but it won't use that

so i reverse to old onetrainer for now

last version i can use in stable matrix is master@aa8d802 not sure if this help if that is version for stable matrix only

[O-J1]

Pytorch dropped support of those GPUs.

[carlo0000]

ok but we should be able to run onetrainer with an old version of pytorch right? or we gonna be stuck with an older version of onetrainer? the good thing for me, the last working version is only 2-3 weeks old so Z-image can be trained

i already made a backup , otherwise i would be unable to get the right version working if i had to reinstall

[O-J1]

We actually need 2.8 for torch compilation which already dropped support for those GPUs. Its time to upgrade the GPU or you will be stuck

[madcrou]

Today I managed to get Onetrainer working - I removed venv, turned off goodbyeDPI, ran install.bat and it worked!