Position: Federated Foundation Language Model Post-Training Should Focus on Open-Source Models

2025-05-29

AI Summary

This paper identifies a critical misalignment: current federated learning (FL) for large language model (LLM) post-training predominantly relies on black-box foundation models, violating FL’s core tenets—data privacy and client autonomy. To address this, the work presents the first systematic critique of the black-box paradigm in federated post-training and introduces the “open-first” principle, formally defining model openness across architecture, weights, and training logic within FL. Methodologically, it integrates federated theory, model interpretability analysis, privacy-utility trade-off modeling, and open-source ecosystem evaluation. The key contribution is establishing open-source models as a necessary precondition for trustworthy federated LLM post-training—thereby laying a theoretical foundation and practical framework for building verifiable, collaborative, and regulation-compliant federated language models.

Technology Category

• Machine Learning: Distributed Machine Learning & Federated Learning
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Post-training of foundation language models has emerged as a promising research domain in federated learning (FL) with the goal to enable privacy-preserving model improvements and adaptations to user's downstream tasks. Recent advances in this area adopt centralized post-training approaches that build upon black-box foundation language models where there is no access to model weights and architecture details. Although the use of black-box models has been successful in centralized post-training, their blind replication in FL raises several concerns. Our position is that using black-box models in FL contradicts the core principles of federation such as data privacy and autonomy. In this position paper, we critically analyze the usage of black-box models in federated post-training, and provide a detailed account of various aspects of openness and their implications for FL.

Problem

• Federated post-training should prioritize open-source models over black-box ones
• Black-box models in FL risk violating data privacy and autonomy principles
• Analyzing openness aspects is crucial for effective federated language model adaptation

Innovation

• Focus on open-source models in FL
• Avoid black-box models for privacy
• Analyze openness implications in FL