thx, could you please change pytorch version to 2.7.1 as i have spotted some RAM issues

Changed Torch Version To 2.7.1+cu128

Note: The embedded Python of WanGP Portable did not come with certain Python development headers and libraries that are needed for Triton. Therefore, users who want to use Sage Attention in the portable version need to obtain those files by downloading and installing the full Python installer from the Python site. The Python version they should be downloading is the exact same Python version that portable WanGP uses(Currently : 3.10.9). After installing, copy the Include and libs folders from the Python installation folder and paste them inside the embedded Python folder of WanGP. Then, install the triton-windows and sage-attention packages.

Example of This Workflow

Sorry, but your "copy and paste" suggestion won't work in most cases, because people won't have installed trition and sage attention in exactly this python version (folder). So the portable version is more or less useless, because you need to install globally(!) everything for 3.10.9 again - ended up in python version dependences conflict hell. I tried it over an hour without success, so I will drop WanGP.
Understand me correctly: with all the Python stuff, it's absolutely essential to create local portable versions like you did, thank you for that!, especially if you already have several programs with specific Python dependencies. However, everything really needs to be included, or there needs to be a way to install it directly locally using PIP.

I haven't tried this, but it's a fantastic idea. One suggestion is to detect the gpu, and install packages needed, sageattention 1/2/3 sdpa, etc, as well as cuda and pytorch

Grateful feedback: The latest version with Sage attention works fabulously with my RTX-5060Ti-16GB!