fix: check for latest OpenAI model names

[ArzelaAscoIi]

Related Issues

• TBD

Proposed Changes:

• check for latest models: https://platform.openai.com/docs/models/continuous-model-upgrades

How did you test it?

• not tested yet

Checklist

• I have read the contributors guidelines and the code of conduct
• I have updated the related issue with new insights and changes
• I added unit tests and updated the docstrings
• I've used one of the conventional commit types for my PR title: fix:, feat:, build:, chore:, ci:, docs:, style:, refactor:, perf:, test:.
• I documented my code
• I ran pre-commit hooks and fixed any issue

[mathislucka]

Does this work for the snapshot models (e.g. gpt-3.5-turbo-0613)?

[ArzelaAscoIi]

No. :/ I just added the one that was missing for us. Basically we need to come up with a rule to match all OpenAI models or somehow with a hard coded rule or implement some kind of dynamic validation by e.g. fetching all available models and check if the chosen one is one of them.

Happy to add extra model names

[anakin87]

Hello!

A general note.
Currently, in Haystack, OpenAI models are supported using two different invocation layers:

• OpenAI for old GPT-3 models and gpt-3.5-turbo-instruct
• ChatGPT for gpt-3.5-turbo and gpt-4 families of models

Check out the two different supports methods:

haystack/haystack/nodes/prompt/invocation_layer/open_ai.py

Lines 272 to 276 in 3803d23

	def supports(cls, model_name_or_path: str, **kwargs) -> bool:
	valid_model = model_name_or_path in ["ada", "babbage", "davinci", "curie", "gpt-3.5-turbo-instruct"] or any(
	m in model_name_or_path for m in ["-ada-", "-babbage-", "-davinci-", "-curie-"]
	)
	return valid_model and not has_azure_parameters(**kwargs)

haystack/haystack/nodes/prompt/invocation_layer/chatgpt.py

Lines 106 to 111 in 3803d23

	def supports(cls, model_name_or_path: str, **kwargs) -> bool:
	valid_model = (
	any(m for m in ["gpt-3.5-turbo", "gpt-4"] if m in model_name_or_path)
	and not "gpt-3.5-turbo-instruct" in model_name_or_path
	)
	return valid_model and not has_azure_parameters(**kwargs)

My impression is that these last models are supported by the ChatGPT layer.

[wochinge]

how about making this configurable via env variables?

[ArzelaAscoIi]

Okay! Makes sense. We will add an extra step to validate also the ChatGPT invocation layer. Thanks for you help!