Improve the installation of optimum-neuron through optimum extras

[JingyaHuang]

What does this PR do?

Fixes huggingface/optimum-neuron#490

Make it possible to install the latest optimum-neuron, with neuron extra optimum[neuron] and optimum[neuronx].

Before submitting

• This PR fixes a typo or improves the docs (you can dismiss the other checks if that's the case).
• Did you make sure to update the documentation with your changes?
• Did you write any new necessary tests?

[JingyaHuang]

Could you review? @philschmid @shub-kris @dacorvo @michaelbenayoun

Tested with testpypi: https://test.pypi.org/manage/project/optimum/release/1.17.2/, should work:

pip install -i https://testpypi.python.org/pypi --upgrade-strategy eager optimum[neuronx]

Installed:

(optimum_setup_test) ubuntu@ip-xxx-xx-xx-xx:~$ pip show optimum
Name: optimum
Version: 1.17.0
Summary: Optimum Library is an extension of the Hugging Face Transformers library, providing a framework to integrate third-party libraries from Hardware Partners and interface with their specific functionality.
Home-page: https://github.com/huggingface/optimum
Author: HuggingFace Inc. Special Ops Team
Author-email: hardware@huggingface.co
License: Apache
Location: /home/ubuntu/pyvenv/optimum_setup_test/lib/python3.8/site-packages
Requires: coloredlogs, datasets, huggingface-hub, numpy, packaging, sympy, torch, transformers
Required-by: optimum-neuron
(optimum_setup_test) ubuntu@ip-xxx-xx-xx-xx:~$ pip show optimum-neuron
Name: optimum-neuron
Version: 0.0.20
Summary: Optimum Neuron is the interface between the Hugging Face Transformers and Diffusers libraries and AWS Tranium and Inferentia accelerators. It provides a set of tools enabling easy model loading, training and inference on single and multiple neuron core settings for different downstream tasks.
Home-page: https://huggingface.co/hardware/aws
Author: HuggingFace Inc. Special Ops Team
Author-email: hardware@huggingface.co
License: Apache
Location: /home/ubuntu/pyvenv/optimum_setup_test/lib/python3.8/site-packages
Requires: accelerate, huggingface-hub, numpy, optimum, protobuf, transformers
Required-by: 

[michaelbenayoun]

The issue with that is that when we update transformers and / or accelerate requirements we will need to make an optimum release as well.

[JingyaHuang]

The other possibility is to remove transformers from REQUIRED_PKGS of optimum, but in this way every subpackage need to add transformers in their requirement, and we need to sync a release... @mfuntowicz

[JingyaHuang]

Remove accelerate since it's not in REQUIRED_PKGS so irrelevant.

[michaelbenayoun]

Alright, then let's do it!

[JingyaHuang]

Will merge this once optimum neuron 0.0.21 is out to test the combination works as expected.

[JingyaHuang]

FYI @shub-kris

The installation of Optimum Neuron should be correct with

pip install --upgrade-strategy eager optimum[neuronx]

(the latest optimum 1.18.1 and optimum-neuron 0.0.21 shall be installed)