Add OpenAI (LLM) procedures

[jexp]

Fixes #3574

One sentence summary of the change.

Proposed Changes (Mandatory)

Similar to the NLP procedures, we can provide a number of procedures that make working with LLM easier

Starting with OpenAI API, possibly extending it to other providers

• generate embeddings
• text & chat completions

Thanks to @tomasonjo for the initial PR

[jexp]

Want to change the procedure results

columns for completion:

created, model, id,completion_tokens, prompt_tokens total_tokens, object->type

not sure how to handle: choices=[{finish_reason="stop", index=0, text="Blue", logprobs=null}]
perhaps we can turn that into one row per choice? and then add the columns for finish_reason, index, text, logprobs

(not sure what logprobs is)

same for chat-completion

[conker84]

created, model, id,completion_tokens, prompt_tokens total_tokens, object->type

I would leave it as is as currently we don't have a clear view about how the developer will use it; @tomasonjo wdyt?

[tomasonjo]

I would also leave it as we return full responses and allow users to do whatever they want.

I am guessing in 98%, the developer usage will be:

YIELD value
WITH value.choices[0].text AS response
do something

[jexp]

Ok then we can release it as is.

[conker84]

@jexp @tomasonjo thank you so much, I just have one question

[conker84]

@jexp @tomasonjo does make sense to make this and the other models configurable, via configuration?

[tomasonjo]

not sure exactly if I understand what you mean, but we could add model parameter under the configuration map. I would leave the prompts as a separate parameter.

[jexp]

The model is something important, so it cannot be missed. For the user it's already in the configuration. But internally I think we should make it explicit.

[conker84]

Yeah I mean something like this

Suggested change

	Stream<Object> resultStream = executeRequest(apiKey, configuration, "embeddings", "text-embedding-ada-002", "input", texts, "$.data");
	String model = configuration.getOrDefault("model", "text-embedding-ada-002");
	Stream<Object> resultStream = executeRequest(apiKey, configuration, "embeddings", model, "input", texts, "$.data");

[jexp]

Not sure we need to do this on every call, as it's already done inside of executeRequest.

[conker84]

ok

[conker84]

@jexp @tomasonjo thank you!