[FEATURE] Deploy Remote Model for Integration tests

[owaiskazi19]

Is your feature request related to a problem?

Currently, as part of integration, bwc etc tests we register a local model here. This issue talks about supporting integration tests for remote model as those are the ones used in production environment.

What solution would you like?

A clear and concise description of what you want to happen.

What alternatives have you considered?

No change

Do you have any additional context?

Add any other context or screenshots about the feature request here.

[will-hwang]

which model should be used for integ test? For integ test, I think testing it against local model should be sufficient for our test cases

[owaiskazi19]

@will-hwang ml-commons uses remote model for integration tests, we could use the same. Agreed, it is enough to test against a local model but on a production system we deploy a remote model so for our integ tests to behave the same on the CI, suggested to use remote model instead.

[heemin32]

Should we do entire of our test using remote model or only the basic funtionality?

[owaiskazi19]

Should we do entire of our test using remote model or only the basic funtionality?

We can use the remote model for basic functionality

[martin-gaievski]

Should we do entire of our test using remote model or only the basic funtionality?

We can use the remote model for basic functionality

+1 to this, we need to test if remote model works with neural search in general,, like an ultimate version of end to end test, so smoke test is enough. Checking of complete scenario set should be done with local model.

[andrross]

Catch All Triage - 1 2

[owaiskazi19]

Created a request to add env variables in neural search repo opensearch-project/.github#372
Once this is done, next task would be to add a require approval CI for external users https://github.com/opensearch-project/ml-commons/blob/main/.github/workflows/require-approval.yml to avoid generating aws credentials on every request

[owaiskazi19]

For remote integ test using a local LLM we can follow the same as search relevance to run OLLAMA opensearch-project/search-relevance#173