Enable azure openai engines

[ecatkins]

Love the library!

I had to make modifications to get it to work with the Azure instances of OpenAI models that I use.

Primarily, I made it possible to pass an engine parameter to the LLMClassifier as well/instead of the model parameter which is what the Azure instance requires.

I tried to do it in a way that minimized the # of edits on the existing codebase - if this is something you'd be interested in including, happy to make any changes to fit with your design principles.

import openai
openai.api_type = "azure"
openai.api_version = "2023-07-01-preview"
openai.api_base = "https://azure-openai-xxxxx.openai.azure.com/"  # Your Azure OpenAI resource's endpoint value.
openai.api_key = "<AZURE-OPENAI_API-KEY>"
evaluator = Factuality(engine="<Azure-Deployment-Name>", model=None)

[ankrgyl]

This looks great! Just left one comment. Really appreciate the contribution 🙏

[ankrgyl]

We may want to just remove this check altogether — we definitely support more than the listed models, and "non-chat" flavored models are all deprecated at this point.

[ecatkins]

Removed

[ankrgyl]

One more thing -- can you make the equivalent change in the JS implementation?

[ecatkins]

One more thing -- can you make the equivalent change in the JS implementation?

JS isn't really my forte - I took a brief look - but chance I would do more harm than good :)

[ankrgyl]

One more thing -- can you make the equivalent change in the JS implementation?

JS isn't really my forte - I took a brief look - but chance I would do more harm than good :)

No worries! I can follow up and do that.

[ankrgyl]

Hopefully last thing -- I ran the linter on your change and it appears to be failing (missing a comma after engine). I think you can fix this by running make fixup, or just add the comma in directly.

[ecatkins]

Hopefully last thing -- I ran the linter on your change and it appears to be failing (missing a comma after engine). I think you can fix this by running make fixup, or just add the comma in directly.

Done - but it looked like the pytests were also failing?

[ankrgyl]

Ah yep -- I think the way we cache parameters will be invalidated everytime a new one is added. Let me update the cached test results and push to your branch