Add PromptSelector plugin to dynamically select the prompt & options to use based on context

[Stevenic]

There are a cases where you might want to change the model (or even prompt) used for a given request based on the current context. Adding a new PromptSelector plugin will let a wave evaluate the current context and select the appropriate prompt and model to use. Here are a few scenarios this would enable:

• Choosing an alternate model when giving feedback (see issue Add support for calling an alternate model when giving feedback #2.) Use a low cost model for the primary task but use GPT-4 to repair hallucinations.
• Choosing an alternate model for longer prompts. Use a low cost model for short tasks but switch to GPT-4 when you need the added tokens.
• Choosing an alternate model based on query complexity. You can use a classifier to select the model to use based off the complexity of the current query.

There would be 2 default implementations provided: The DefaultPromptSelector would just select the configured prompt every time. The FeedbackPromptSelector would chose an alternate prompt when giving the model feedback.

[rmolinamir]

Hi @Stevenic, this project is awesome. I've been using it at work for a little R&D, the appliance is in the media industry so I've been using it to generate automated translations of SRT files, it's been a great first step, the idea is to increase output and automate the operational workflow as much as possible. I've been thinking of using it for some side projects as well, I was wondering if you were open to contributors, I'd love to pitch in.

[Stevenic]

Of course @rmolinamir... I'd love contributors... Let me setup discussions for the project and I'll invite you as a collaborator. I'm in the middle of adding support for OpenAI functions so quite a few moving pieces right now.