The advent of large language models (LLMs) like GPT-4 has catalyzed the
exploration of multi-task learning (MTL), in which a single model demonstrates
proficiency across diverse tasks. Task arithmetic has emerged as a
cost-effective approach for MTL. It enables performance enhancement across
multiple tasks by adding their corresponding task vectors to a pre-trained
model. However, the current lack of a method that can simultaneously achieve
optimal performance, computational efficiency, and data privacy limits their
application to LLMs. In this paper, we propose \textbf{M}odel
\textbf{E}xclusive \textbf{T}ask \textbf{A}rithmetic for merging
\textbf{GPT}-scale models, which formalizes the objective of model merging into
a multi-task learning framework, aiming to minimize the average loss difference
between the merged model and each individual task model. Since data privacy
limits the use of multi-task training data, we leverage LLMs' local linearity
and task vectors' orthogonality to separate the data term and scaling
coefficients term and derive a model-exclusive task arithmetic method. Our
proposed MetaGPT is data-agnostic and bypasses the heavy search process, making
it cost-effective and easy to implement for LLMs.Extensive experiments
demonstrate that MetaGPT leads to improvements in task arithmetic and achieves
state-of-the-art performance on multiple tasks.

本文介绍了一种用于合并大型语言模型的模型专属任务算法，该算法通过优化合并模型和每个个体任务模型之间的平均损失差异来实现多任务学习，该方法不受隐私限制，具有成本效益和易于实施的优势，并在多个任务上取得了最先进的性能。

MetaGPT: 使用模型专属任务算法合并大型语言模型

MetaGPT: Merging Large Language Models Using Model Exclusive Task  Arithmetic

Recently, remarkable progress has been made in automated task-solving through
the use of multi-agents driven by large language models (LLMs). However,
existing works primarily focuses on simple tasks lacking exploration and
investigation in complicated tasks mainly due to the hallucination problem.
This kind of hallucination gets amplified infinitely as multiple intelligent
agents interact with each other, resulting in failures when tackling
complicated problems.Therefore, we introduce MetaGPT, an innovative framework
that infuses effective human workflows as a meta programming approach into
LLM-driven multi-agent collaboration. In particular, MetaGPT first encodes
Standardized Operating Procedures (SOPs) into prompts, fostering structured
coordination. And then, it further mandates modular outputs, bestowing agents
with domain expertise paralleling human professionals to validate outputs and
reduce compounded errors. In this way, MetaGPT leverages the assembly line work
model to assign diverse roles to various agents, thus establishing a framework
that can effectively and cohesively deconstruct complex multi-agent
collaborative problems. Our experiments conducted on collaborative software
engineering tasks illustrate MetaGPT's capability in producing comprehensive
solutions with higher coherence relative to existing conversational and
chat-based multi-agent systems. This underscores the potential of incorporating
human domain knowledge into multi-agents, thus opening up novel avenues for
grappling with intricate real-world challenges. The GitHub repository of this
project is made publicly available on: this https URL

MetaGPT 是一种创新的框架，将有效的人类工作流作为元编程方法融入到 LLM 驱动的多智能体协作中，以便有效地解决复杂的多智能体协作问题，并在协作软件工程任务上实现了高一致性的全面解决方案。