Is it possible to provide a virtual_vram_gb slider or Expert Mode (more suitable for multiple older GPUs) for WanVideoModelLoaderMultiGPU to use distributed torch?

[Meaolock]

Can I load the models (e.g., wan2.2, lora, extra_model, inifniteTalk) required by the WanVideoModelLoaderMultiGPU node using Distorch2?

I want to do this because my GPU architecture (Pascal) is relatively old and doesn't support any attention mechanisms (e.g., SageAttention, flash_attention).

However, (UNet, CLIP, VAE) LoaderDisTorch2MultiGPU works perfectly. I really like this project. It makes many of my older, high-memory GPUs no longer useless (like a bloated vase without flowers).

So, Is it possible to provide a virtual_vram_gb slider or Expert Mode (more suitable for multiple older GPUs) for WanVideoModelLoaderMultiGPU use distributed torch?

[an80sPWNstar]

Bump on this. Since I have no idea if it is possible, I would love to have the distorch2 options for WanVideoModelLoader and other WanVideo nodes. Kijai loves to use those nodes and it forces us multi gpu peeps to offload onto ram.

Your contributions have been sooooo hepful. Thank you for all that you do!