noise output
#2
by
Luke100000
- opened
Running the example code results in noise:
Run on an 5600x, python 3.9. I expect a loss if performance, but it should generally work on AMD, right?
Hi, it should work on AMD as well. I assume that the problem in the compatibility of the current version of Transformer libraries and Optimum-Intel with the model. I would recommend regenerating the model from the source using export option in the pipeline class, like pipe = OVStableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5", export=True).
Please take a look at the notebooks we have in OpenVINO, e.g. https://github.com/openvinotoolkit/openvino_notebooks/blob/main/notebooks/276-stable-diffusion-torchdynamo-backend/276-stable-diffusion-torchdynamo-backend.ipynb