This is a model compiled for optimized inference on AWS inf2 instances. The overall performance can achieve 4it/s for a 1024x1024 picture.
NOTE: In order to load and infer on inf2.xlarge instances, a minimum of 8 GB SWAP memory is required. Other instance sizes do not require this.
Assuming you're using the official DLAMI or installed the drivers and libraries required for neuron devices, install the latest optimum_neuronx library:
1from optimum.neuron import NeuronStableDiffusionXLPipeline
2pipe = NeuronStableDiffusionXLPipeline.from_pretrained("paulkm/sdxl_neuron_pipe", device_ids=[0,1])
3img = pipe("a cute black cat").images[0]
