mirror of
https://github.com/invoke-ai/InvokeAI.git
synced 2026-01-14 20:38:08 -05:00
d85733f22b
Each version of torch is only available for specific versions of CUDA and ROCm. The Invoke installer and dockerfile try to install torch 2.4.1 with ROCm 5.6 support, which does not exist. As a result, the installation falls back to the default CUDA version so AMD GPUs aren't detected. This commits fixes that by bumping the ROCm version to 6.1, as suggested by the PyTorch documentation. [1] The specified CUDA version of 12.4 is still correct according to [1] so it does need to be changed. Closes #7006 Closes #7146 [1]: https://pytorch.org/get-started/previous-versions/#v241
InvokeAI
Project homepage: https://github.com/invoke-ai/InvokeAI
Preparations:
You will need to install Python 3.10 or higher for this installer
to work. Instructions are given here:
https://invoke-ai.github.io/InvokeAI/installation/INSTALL_AUTOMATED/
Before you start the installer, please open up your system's command
line window (Terminal or Command) and type the commands:
python --version
If all is well, it will print "Python 3.X.X", where the version number
is at least 3.10.*, and not higher than 3.11.*.
If this works, check the version of the Python package manager, pip:
pip --version
You should get a message that indicates that the pip package
installer was derived from Python 3.10 or 3.11. For example:
"pip 22.0.1 from /usr/bin/pip (python 3.10)"
Long Paths on Windows:
If you are on Windows, you will need to enable Windows Long Paths to
run InvokeAI successfully. If you're not sure what this is, you
almost certainly need to do this.
Simply double-click the "WinLongPathsEnabled.reg" file located in
this directory, and approve the Windows warnings. Note that you will
need to have admin privileges in order to do this.
Launching the installer:
Windows: double-click the 'install.bat' file (while keeping it inside
the InvokeAI-Installer folder).
Linux and Mac: Please open the terminal application and run
'./install.sh' (while keeping it inside the InvokeAI-Installer
folder).
The installer will create a directory of your choice and install the
InvokeAI application within it. This directory contains everything you need to run
invokeai. Once InvokeAI is up and running, you may delete the
InvokeAI-Installer folder at your convenience.
For more information, please see
https://invoke-ai.github.io/InvokeAI/installation/INSTALL_AUTOMATED/