Standardizing the LLM parameters ?

[AbdealiLoKo]

Hi, I am trying to out aisuite - and was wondering if there is any interest in standardizing LLM params like temperature, top_k, etc.
I notice that currently every provider has it's own independent set of params

Bedrock -

aisuite/aisuite/providers/aws_provider.py

Line 13 in cf9df9a

INFERENCE_PARAMETERS = ["maxTokens", "temperature", "topP", "stopSequences"]

Ollama -

aisuite/aisuite/providers/ollama_provider.py

Line 29 in cf9df9a

def chat_completions_create(self, model, messages, **kwargs):

etc.

[rohitprasad15]

Standardizing the parameters was not our goal.
We wanted to provide a way to switch between models & providers easily.. and since OpenAI API format was most prevalent, we chose to use that API (and parameter names) - and chose to convert everything transparently to the downstream provider.

If you have a proposal on how we can cleanly help clear confusions around model specific parameters, please do so. I am happy to discuss approaches and figure out a solution.