AutoGen
Framework for multi-agent conversations where agents critique each other.
Обзор
AutoGen runs several agents in a conversation, so one can propose and another can check, with the exchange defined rather than improvised. It solves the quality problem of a single pass: a model reviewing its own output tends to agree with itself, while a separate reviewer with a different instruction does not. It exists because some tasks improve more from structured disagreement than from a larger model.
Граф технологий
Как AutoGen связана с остальным стеком.
Инструменты
ПО, которое реализует AutoGen или интегрируется с ним.