openvinotoolkit
Sana-sprint with gray blank image when using GPU
when I using sana-sprint with CPU, works fine but using GPU, shows blank gray images How can I fix that?
when i inferencing float32, works fine. but spend too much memory ans slower than float16...
Can you provide more details about your environment, please? Do you refer to this Jupyter-Notebook https://github.com/openvinotoolkit/openvino_notebooks/blob/latest/notebooks/sana-image-generation/sana-image-generation.ipynb ?
Which HW are you using, what about of system memory do you have?
Which operating system, which Python version, which version of this repository do you use?
Have you changed something in the notebook?
Which exact model variant have you selected?
When using the model variant "Efficient-Large-Model/Sana_Sprint_0.6B_1024px_diffusers" (default, the first model in the drop-down menu), keeping the checkbox for weight-compression selected, using MS-Win-11-Pro on an Intel Core Ultra 7 155H laptop, using Python 3.12.4 and 64GB system memory, using CPU for inference is working (using the prompt "A software engineer experimenting with a generative artificial intelligence for the first time and experiencing something unexpected"):
Set timesteps: tensor([1.5708, 1.3000, 0.0000]) 100%  2/2 [00:06<00:00,  3.37s/it]
Then I have shutdown the Jupyter server and restarted it (because the memory consumption didn't went back after running the inference cell; I have not used the interactive demo!).
Then selecting "GPU". Inference completed faster than CPU: Set timesteps: tensor([1.5708, 1.3000, 0.0000]) 100%  2/2 [00:05<00:00,  2.14s/it]
However, the generated picture was completely grey:
We received the same output when running with the GPU. I'm checking with the development team on this problem.
