poluramus

Is there any workflow for the new video model ?

> import torch > from app.sana_pipeline import SanaPipeline > from torchvision.utils import save_image > > device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") > generator = torch.Generator(device=device).manual_seed(0) > > sana =...

> Have you solved the problem? [@poluramus](https://github.com/poluramus) Unfortunately not :-(. Let me know if you need more info.

> Is it possible that you have run OOM when using the 4.8B model. How much VRAM do you have? Maybe you can also try to use VAE tiling before...

I got this code working now. It's weird, since i tried it a couple of days ago and I was getting an error: ``` # run `pip install git+https://github.com/huggingface/diffusers` before...

> What changes have you made? [@poluramus](https://github.com/poluramus) The code you provided in your first message still don't work (same problem as described in my previous messages). And i just found...

> > The code you provided in your first message still don't work (same problem as described in my previous messages). > > > And > > i just found...

@lawrence-cj Here is the code i use on my part: No PAG ```import torch from diffusers import SanaPipeline pipe = SanaPipeline.from_pretrained( "Efficient-Large-Model/SANA1.5_4.8B_1024px_diffusers", torch_dtype=torch.bfloat16 ) pipe.to("cuda") pipe.text_encoder.to(torch.bfloat16) pipe.vae.to(torch.bfloat16) prompt = 'a...

Did you figured it out @lawrence-cj ?

I have the same problem. Seems to works fine for 100 to 1000 steps, then the it/s goes up from 20 to 250 s/it. I recreated a fresh new venv...