Associate prompts to models and inference parameters

[synw]

It would be nice to directly associate a model to a given prompt. As Ollama can switch models at runtime it would be very convenient to automate this instead of having to manually pick a model each time.

The custom prompts specification could include more information:

• System prompt
• Main prompt
• Model to use
• Inference parameters to use
• Few shots (option)

This would make the custom prompts more precise. For example using a few shots helps a lot with small models, or temperature changes between the tasks. To have different inference params and system prompts for each task would be really nice.

I have made a Typescript spec and library that implements yaml defined prompts with all these features, if it can help: check the doc and code

[n4ze3m]

Yes, this is a nice idea! I'll look into it. :)