o3-mini: unsupported parameters listed in docs/usage.md

[bocytko]

2a355ea#diff-72376d0b487d7231cf31959bf266f6aea098e899743bae02e36d176cef4db476R528 added a copy of the parameter list from o1 model.

OpenAI returns exceptions if at least one of the following is used (I may not have tested all combinations):

• temperature
• max_tokens (instead max_completion_tokens should be used as mentioned in OpenAI o1 models require max_completion_tokens instead of max_tokens #724)
• top_p
• frequency_penalty
• presence_penalty

Model capability registry

I don't know if a crowdsourced source of truth for all model capabilities exist that can be easily used as model registry.
litellm ships with an endpoint that serves supported model lists that can conveniently be used in tools. Ideally, we'd have sth. similar, hosted on S3/CDN.

---

Example errors

Note: truncated output for extra models integrated via litellm.

$ llm -m o3-mini --option temperature 0.5 "write a poem"
OpenAIException - Error code: 400 - {\'error\': {\'message\': "Unsupported parameter: \'temperature\' is not supported with this model.", \'type\': \'invalid_request_error\', \'param\': \'temperature\', \'code\': \'unsupported_parameter\'}}

$ llm -m o3-mini --option max_tokens 50 "write a poem"
OpenAIException - Error code: 400 - {\'error\': {\'message\': "Unsupported parameter: \'max_tokens\' is not supported with this model. Use \'max_completion_tokens\' instead.", \'type\': \'invalid_request_error\', \'param\': \'max_tokens\', \'code\': \'unsupported_parameter\'}}