Towards Harnessing the Collaborative Power of Large and Small Models for Domain Tasks

2025-04-24

AI Summary

To address the high adaptation cost and deployment difficulty of large language models (LLMs) in private domains, this paper proposes a novel paradigm of collaborative adaptation between large and small models: leveraging a lightweight small model as an intelligent proxy to orchestrate efficient domain transfer of LLMs. Methodologically, we design a dynamic collaboration strategy, plug-and-play domain-adaptation modules, a multi-objective evaluation benchmark grounded in real-world private data, and a privacy-aware data modeling pipeline. Theoretically, we establish the first analytical framework for collaborative adaptation. Practically, we deliver an end-to-end industrial-grade roadmap, validated across vertical domains—including finance and healthcare—demonstrating substantial improvements: 2.3× average inference speedup, 76% reduction in fine-tuning GPU memory consumption, and preservation of domain-specific task performance. This work advances the development of trustworthy, production-ready domain-specialized AI ecosystems.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: (Large) Language Models
• Computer Vision: Large Vision Models

Application Category

• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large language models (LLMs) have demonstrated remarkable capabilities, but they require vast amounts of data and computational resources. In contrast, smaller models (SMs), while less powerful, can be more efficient and tailored to specific domains. In this position paper, we argue that taking a collaborative approach, where large and small models work synergistically, can accelerate the adaptation of LLMs to private domains and unlock new potential in AI. We explore various strategies for model collaboration and identify potential challenges and opportunities. Building upon this, we advocate for industry-driven research that prioritizes multi-objective benchmarks on real-world private datasets and applications.

Problem

• Leveraging large and small models for domain tasks
• Accelerating LLM adaptation to private domains
• Exploring collaborative strategies for model synergy

Innovation

• Collaborative approach combining large and small models
• Multi-objective benchmarks on private datasets
• Industry-driven research for domain adaptation