Closed endymion closed 1 month ago
The base class required it. I brought this up previously but didn't get an actionable response.
https://github.com/AnthusAI/Plexus/commit/092e04415b8dcb299472345f4697e73744713f06
I implemented a pass
on train_model
and gave the method a more appropriate name create_validation_agent
.
Ah, okay, that makes sense. I was just worried about errors from that function getting called when we call plexus train ...
on a scorecard with a mixture of those scores and ML scores.
There is a
train_model()
function in theAgenticValidator
class, but does that make sense?This needs better documentation, but
train_model()
is a standard lifecycle function for a PlexusScore
that's vital for ML models that require training weights or whatever and then saving a model artifact file for use during inference. But theAgenticValidator
operates on retrained models through APIs, and doesn't require doing anything like that at training time, as far as I know?Is that functionality in the function right now important for something? I'm worried that it will never be invoked.
We probably need a standard
document_score()
lifecycle function for PlexusScore
s that will generate model diagrams for ML models graph diagrams for LangGraph agenticScore
s, etc.