ON compile LCM-SD broken with cannot import NeuronLatentConsistencyModelPipeline error

[kct22aws]

System Info

optimum                       1.17.0
optimum-neuron                0.0.13
Python 3.10
Ubuntu 22.04
Neuron 2.17.0

Who can help?

@JingyaHuang Official code snippet in https://huggingface.co/docs/optimum-neuron/tutorials/stable_diffusion#compile-lcm is broken with following error:

    from optimum.neuron import NeuronLatentConsistencyModelPipeline
ImportError: cannot import name 'NeuronLatentConsistencyModelPipeline' from 'optimum.neuron' (/home/ubuntu/aws_neuron_venv_pytorch/lib/python3.10/site-packages/optimum/neuron/__init__.py)

Information

• The official example scripts
• My own modified scripts

Tasks

• An officially supported task in the examples folder (such as GLUE/SQuAD, ...)
• My own task or dataset (give details below)

Reproduction (minimal, reproducible, runnable)

Use AWS inf2 instance, install Neuron 2.17.0, then install current version of transformers, accelerate, diffusers, evaluate, datasets. Then install ON via pip install optimum[neuronx]. Then run code as a Python script.

Expected behavior

Expect the code to complete the run.

[JingyaHuang]

Hi @kct22aws, thanks fore reporting the issue. The error is introduced by the refactoring of the model registration mechanism in Optimum 1.17.0, it is fixed by #414, and we are planning for a release for it. In the future we will also pin optimum version to avoid breaking changes from Optimum's side.

If you are testing something, you could use optimum < 1.17 until the new release of Optimum Neuron.

[JingyaHuang]

Optimum Neuron 0.0.19 is out, could you try if the issue is solved? Thanks!
https://pypi.org/project/optimum-neuron/0.0.19/

[JingyaHuang]

Close as it should have been solved with the latest release. Feel free to reopen if there are any follow-up questions.