Routing strategy based on feature of models (i.e. function calling)

[shuther]

Many applications uses gpt4 for everything they do, but I would like to leverage other models based on particular conditions; as an example:

• use llama3 if no function calls are requested from the prompt
• use a model that supports it if there are up to 3 proposed functions
• use gpt4 if there are more than 4 in the prompt

Could it be done using: enable_pre_call_checks; I am looking for advice on how to build such routing mechanism.