Learning to Collaborate: An Orchestrated-Decentralized Framework for Peer-to-Peer LLM Federation

TL;DR

This paper presents KNEXA-FL, a novel decentralized framework for collaborative fine-tuning of LLMs that uses a learned matchmaking policy to improve performance and stability without central aggregation.

Contribution

Introduces KNEXA-FL, a framework that employs a contextual bandit approach for orchestrated peer-to-peer collaboration among heterogeneous LLMs via secure distillation.

Findings

Achieves approximately 50% improvement in Pass@1 over random P2P collaboration.

Demonstrates stable convergence unlike centralized distillation which collapses.

Establishes adaptive learning-based orchestration as key for decentralized AI ecosystems.

Abstract

Fine-tuning Large Language Models (LLMs) for specialized domains is constrained by a fundamental challenge: the need for diverse, cross-organizational data conflicts with the principles of data privacy and sovereignty. While Federated Learning (FL) provides a framework for collaboration without raw data exchange, its classic centralized form introduces a single point of failure and remains vulnerable to model inversion attacks. Decentralized FL (DFL) mitigates this risk by removing the central aggregator but typically relies on inefficient, random peer-to-peer (P2P) pairings, forming a collaboration graph that is blind to agent heterogeneity and risks negative transfer. This paper introduces KNEXA-FL, a novel framework for orchestrated decentralization that resolves this trade-off. KNEXA-FL employs a non-aggregating Central Profiler/Matchmaker (CPM) that formulates P2P collaboration as a contextual bandit problem, using a LinUCB algorithm on abstract agent profiles to learn an optimal matchmaking policy. It orchestrates direct knowledge exchange between heterogeneous, PEFT-based LLM agents via secure distillation, without ever accessing the models themselves. Our comprehensive experiments on a challenging code generation task show that KNEXA-FL yields substantial gains, improving Pass@1 by approx. 50% relative to random P2P collaboration. Critically, our orchestrated approach demonstrates stable convergence, in stark contrast to a powerful centralized distillation baseline which suffers from catastrophic performance collapse. Our work establishes adaptive, learning-based orchestration as a foundational principle for building robust and effective decentralized AI ecosystems.