Remove top_p default value

[xdite]

Pratically, we should not use top_p and temperture at the same time, however, we may override temperture in some cases, but since DSPy predefine top_p = 1, it's impossible to leave top_p as blank.

this weired behavior is due to there are very diffrent results from direct openai call V.S. DSPy.

[arnavsinghvi11]

Thanks @xdite . Yes ideally, we can remove these definitions and have users define all the model kwargs directly (some models are not compatible with frequency_penalty, presence_penalty, etc.) but tagging @okhat to verify this won't break any existing caches with GPT3?

[okhat]

This will break existing caches. Maybe this can somehow be marked as "for next breaking release" and then merged in the future.

But I don't understand the need for this. Can't the user just pass None? Why can't we specify top_p and temperature at once? What's the effect of this PR?

[ammirsm]

hey folks, just a follow up on this one, should we close this pull request?