Support passing metadata to cheval annotators.

[superdosh]

See related PR: https://github.com/mlcommons/sugar/pull/568

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅

[bkorycki]

Looks good! But why do we need to update both the ChevalAnnotator and DagAnnotator? Are they both used by security?

[rogthefrog]

👍🏻

[superdosh]

Looks good! But why do we need to update both the ChevalAnnotator and DagAnnotator? Are they both used by security?

ChevalAnnotator makes sure the request sent by modelbench to cheval has the metadata. Then the change to DAGAnnotator is used within cheval to make sure the metadata is read into the EvalContext.

@bkorycki

[bkorycki]

@superdosh That makes sense!

[wpietri]

I think this is going the wrong direction in a way that makes the domain model muddier and harder to maintain. If it's something we need to do for expediency, go for it, but please add a card for doing it right that we tackle as soon as the emergency is past.

[superdosh]

I think this is going the wrong direction in a way that makes the domain model muddier and harder to maintain. If it's something we need to do for expediency, go for it, but please add a card for doing it right that we tackle as soon as the emergency is past.

@wpietri I think there's a few changes I can make quickly that should alleviate many of the concerns. Will do a quick update / test and then ping you!

[wpietri]

This makes the problem clearer, which I think is a good step forward, but let's definitely come back to clean this up when we're past the rush.

[superdosh]

This makes the problem clearer, which I think is a good step forward, but let's definitely come back to clean this up when we're past the rush.

Added #1555 with details as follow up!