Conversation
| agent: Agent | None, | ||
| include_reasoning: bool = False, | ||
| ) -> CategoricalPrediction: | ||
| agent = agent or Agent(model="gpt-3.5-turbo-0125", model_settings=ModelSettings(temperature=0.7)) |
There was a problem hiding this comment.
Just copy pasted the defaults there were previously from Agentcoin's team.
Note that in PMA, model is overridden, so it doesn't really matter for us.
| class CategoricalPrediction(TypedDict): | ||
| decision: str | ||
| probabilities: Dict[str, Probability] | ||
| confidence: float | ||
| info_utility: float | ||
| reasoning: Optional[str] |
There was a problem hiding this comment.
Rhetorical question: would it be somehow possible to start using pyedantic here? (understand why you did not used it)
There was a problem hiding this comment.
Yeah, I just followed Agentcoin's convention in this repo. But if we agree on refactoring this, I don't think they will have any problems with that (given that much bigger changes were implemented already anyway 😄)
Also PydanticAI isn't used to its full potential here, we could do something like:
agent = Agent(
model,
output=CategoricalPrediction
)Instead of parsing it manually, but I didn't want to break the current flow (just to be sure to not break performance of these agents).
But again if we agree on refactoring, and test out that performance remains +/- the same, we can.
3 participants
No description provided.