Make LLM-powered apps Smarter and Cheaper, Reliable, Auditable changing 1 LOC (github.com)

A new tool called Consortium has been announced, designed to optimize large language model (LLM) applications by facilitating complex, reliable workflows that go beyond single model calls. This system enables durable directed acyclic graph (DAG) execution, layered reasoning approaches, and detailed cost/token tracking. By incorporating ensemble and fusion methods, Consortium allows for multiple models to operate concurrently, comparing outputs, generating audit trails, and ensuring redundancy, ultimately enhancing decision-making reliability across fields like healthcare and finance. For the AI/ML community, the significance of Consortium lies in its ability to streamline model orchestration while reducing costs. It leverages established theoretical frameworks, such as Condorcet's Jury Theorem, to allow diverse models to work together effectively. The introduction of a user-friendly interface for workflow management, alongside an OpenAI-compatible API, allows existing applications to integrate seamlessly without modification. Ensemble methods, including majority voting and self-consistency checks, are built-in, providing developers with versatile tools tailored to different tasks, thus promoting innovation and cost-efficiency within AI deployments.