The latest update combines the WAN2GP installer with the WAN2GP-AMD installer.
It is compatible with GTX10 - RTX50 series and supports AMD on Windows and Linux.
WAN2GP can now be upgraded with python 3.11, pytorch 2.10 and NVFP4 support for RTX 50 series.
If you install WAN2GP from scratch you will automatically get the new version.
If you are still using WAN2GP you can now upgrade to Python 3.11 and Pytorch 2.10 with NVFP4 support.
To do so, click the update button to get the new version.
Return to the main page and click WAN2GP. You can now see an upgrade button that allows you to upgrade to the new version.
This upgrade is optional and won't delete your settings or models.
Like before this new WAN2GP version comes with all Attention Types: xformers, flash attention and sage attention 2 (for GTX 10 / 16 series sage attention 1).
Thanks to @gkartist and SAM the installer was tested on Linux and Windows for RTX 30 - RTX 50 series.
Please note that NVFP4 will fail if Compile Transformer Model is activated. This is no bug, this is a limitation of NVFP4.
To deactivate Compile Transformer Model go to Configuration - Performance and set Compile Transofmer Model to Off.
Save the Settings at the bottom and restart WAN2GP.
Feedback would be highly appreciated. Here as a comment or in pinokio-support.
If anything isn't working please report in pinokio-support on the Pinokio Discord server.
I'll take care of any issues and fix them.
Once we've received enough feedback, this may become part of the official WAN2GP installer.
The latest update combines the WAN2GP installer with the WAN2GP-AMD installer.
It is compatible with GTX10 - RTX50 series and supports AMD on Windows and Linux.
WAN2GP can now be upgraded with python 3.11, pytorch 2.10 and NVFP4 support for RTX 50 series.
If you install WAN2GP from scratch you will automatically get the new version.
If you are still using WAN2GP you can now upgrade to Python 3.11 and Pytorch 2.10 with NVFP4 support.
To do so, click the update button to get the new version.
Return to the main page and click WAN2GP. You can now see an upgrade button that allows you to upgrade to the new version.
This upgrade is optional and won't delete your settings or models.
Like before this new WAN2GP version comes with all Attention Types: xformers, flash attention and sage attention 2 (for GTX 10 / 16 series sage attention 1).
Thanks to @gkartist and SAM the installer was tested on Linux and Windows for RTX 30 - RTX 50 series.
Please note that NVFP4 will fail if Compile Transformer Model is activated. This is no bug, this is a limitation of NVFP4.
To deactivate Compile Transformer Model go to Configuration - Performance and set Compile Transofmer Model to Off.
Save the Settings at the bottom and restart WAN2GP.
Feedback would be highly appreciated. Here as a comment or in pinokio-support.
If anything isn't working please report in pinokio-support on the Pinokio Discord server.
I'll take care of any issues and fix them.
Once we've received enough feedback, this may become part of the official WAN2GP installer.