PML-302 MerLinProcessor export_config and typing

[LF-Vigneux]

Summary

Reinforced typing around the layer's export_config method necessity and its output dictionary

Related Issue

PML-302

Type of change

• Bug fix
• New feature
• Documentation update
• Refactor / Cleanup
• Performance improvement
• CI / Build / Tooling
• Breaking change (requires migration notes)

Proposed changes

• Created the runtime checkable SupportsExportConfig that makes sure that layers have a export_config function. The error message clearly indicates what should be the output of this method as well
• Created the class ValidatedLayerConfig that validates and keeps the typed checked values of the output of export_config.

How to test / How to run

pytest -q

Documentation

• User docs updated (Sphinx)
• Examples / notebooks updated
• Docstrings updated
• Updated the API

[LF-Vigneux]

I am not sure that I fullfill this criteria perfectly:

There is one obvious typed place to look to understand what MerlinProcessor expects from offloadable layers.

However, When there is no export_config method, I clearly state what the output of this method should look like. Also, a specific definition of types is generated for the errors of the output dictionary. I want your opinion on this point.

[ben9871]

Token validation is too extreme and stops the branches tests from running. Other notes covered in a PR to this fork unrelated to the ticket(only visible now following tests with scaleway token)

[LF-Vigneux]

From your markdown fixes

1. Ownership of Session: From what I understand, I removed the session paths to enter and exit in the corresponding methods

[LF-Vigneux]

For 3, changed in next commit

[LF-Vigneux]

For 2, commented the probs test, right now there is no session that has probs.
Here are all the platforms
SAMPLES ONLY: EMU-SAMPLING-4H100SXM ['samples', 'sample_count']
SAMPLES ONLY: QPU-BELENOS-12PQ ['sample_count', 'samples', 'entropy']
FAILED: QPU-ALTAIR-10PQ {'help_message': 'platform QPU in maintenance, wait for platform to be available or use another platform', 'message': 'precondition is not respected', 'precondition': 'resource_not_usable', 'type': 'precondition_failed'}
SAMPLES ONLY: EMU-ALTAIR-10PQ ['sample_count', 'samples']
SAMPLES ONLY: EMU-BELENOS-12PQ ['sample_count', 'samples']
SAMPLES ONLY: EMU-ASCELLA-6PQ ['sample_count', 'samples']
SAMPLES ONLY: EMU-SAMPLING-L4 ['samples', 'sample_count']
SAMPLES ONLY: EMU-SAMPLING-8H100SXM ['samples', 'sample_count']
SAMPLES ONLY: EMU-SAMPLING-P100 ['samples', 'sample_count']
SAMPLES ONLY: EMU-SAMPLING-2L4 ['samples', 'sample_count']
FAILED: QPU-ASCELLA-6PQ {'help_message': 'platform QPU in maintenance, wait for platform to be available or use another platform', 'message': 'precondition is not respected', 'precondition': 'resource_not_usable', 'type': 'precondition_failed'}
SAMPLES ONLY: EMU-SAMPLING-4L4 ['samples', 'sample_count']
SAMPLES ONLY: EMU-SAMPLING-H100 ['samples', 'sample_count']

So the test on scaleway is not useful yet but, it is tested locally to work on artifiical backends

[LF-Vigneux]

All tests are now passing scaleway and local

[ben9871]

feedback taken into account. good to go