In this paper, we introduce a novel approach for large language model merging
via black-box multi-objective optimization algorithms. The goal of model
merging is to combine multiple models, each excelling in different tasks, into
a single model that outperforms any of the individual source models. However,
model merging faces two significant challenges: First, existing methods rely
heavily on human intuition and customized strategies. Second, parameter
conflicts often arise during merging, and while methods like DARE [1] can
alleviate this issue, they tend to stochastically drop parameters, risking the
loss of important delta parameters. To address these challenges, we propose the
MM-MO method, which automates the search for optimal merging configurations
using multi-objective optimization algorithms, eliminating the need for human
intuition. During the configuration searching process, we use estimated
performance across multiple diverse tasks as optimization objectives in order
to alleviate the parameter conflicting between different source models without
losing crucial delta parameters. We conducted comparative experiments with
other mainstream model merging methods, demonstrating that our method
consistently outperforms them. Moreover, our experiments reveal that even task
types not explicitly targeted as optimization objectives show performance
improvements, indicating that our method enhances the overall potential of the
model rather than merely overfitting to specific task types. This approach
provides a significant advancement in model merging techniques, offering a
robust and plug-and-play solution for integrating diverse models into a
unified, high-performing model.

介绍了一种通过黑盒多目标优化算法进行大型语言模型合并的新方法，通过自动化配置搜索的过程，使用多个多样化任务的性能估计作为优化目标，以消除不同源模型之间的参数冲突，同时不丢失重要的增量参数，提供了模型合并技术的重要进展，为将多个模型集成为统一的高性能模型提供了强大而易于使用的解决方案。

多目标优化：通过多个 LLM 的潜力释放形态时间

It's Morphing Time: Unleashing the Potential of Multiple LLMs via  Multi-objective Optimization

Large Language Models (LLMs) have garnered considerable attention owing to
their remarkable capabilities, leading to an increasing number of companies
offering LLMs as services. Different LLMs achieve different performance at
different costs. A challenge for users lies in choosing the LLMs that best fit
their needs, balancing cost and performance. In this paper, we propose a
framework for addressing the cost-effective query allocation problem for LLMs.
Given a set of input queries and candidate LLMs, our framework, named OptLLM,
provides users with a range of optimal solutions to choose from, aligning with
their budget constraints and performance preferences, including options for
maximizing accuracy and minimizing cost. OptLLM predicts the performance of
candidate LLMs on each query using a multi-label classification model with
uncertainty estimation and then iteratively generates a set of non-dominated
solutions by destructing and reconstructing the current solution. To evaluate
the effectiveness of OptLLM, we conduct extensive experiments on various types
of tasks, including text classification, question answering, sentiment
analysis, reasoning, and log parsing. Our experimental results demonstrate that
OptLLM substantially reduces costs by 2.40% to 49.18% while achieving the same
accuracy as the best LLM. Compared to other multi-objective optimization
algorithms, OptLLM improves accuracy by 2.94% to 69.05% at the same cost or
saves costs by 8.79% and 95.87% while maintaining the highest attainable
accuracy.

我们提出了一个针对大型语言模型的成本效益查询分配问题的框架，名为 OptLLM，通过使用多标签分类模型进行性能预测，生成一系列优化解决方案，旨在满足用户的预算限制和性能偏好，包括最大化准确性和最小化成本。OptLLM 在各种类型的任务上进行了广泛的实验，包括文本分类、问答、情感分析、推理和日志解析，实验证明 OptLLM 在降低成本 2.40% 至 49.18% 的同时实现与最佳大型语言模型相同的准确性，相比其他多目标优化算法，OptLLM 在相同成本下提高 2.94% 至 69.05% 的准确性或节省 8.79% 至 95.87% 的成本并保持最高可达准确性。

OptLLM: 大型语言模型的最佳查询分配

OptLLM: Optimal Assignment of Queries to Large Language Models

The significance of transportation efficiency, safety, and related services
is increasing in urban vehicular networks. Within such networks, roadside units
(RSUs) serve as intermediates in facilitating communication. Therefore, the
deployment of RSUs is of utmost importance in ensuring the quality of
communication services. However, the optimization objectives, such as time
delay and deployment cost, are commonly developed from diverse perspectives. As
a result, it is possible that conflicts may arise among the objectives.
Furthermore, in urban environments, the presence of various obstacles, such as
buildings, gardens, lakes, and other infrastructure, poses challenges for the
deployment of RSUs. Hence, the deployment encounters significant difficulties
due to the existence of multiple objectives, constraints imposed by obstacles,
and the necessity to explore a large-scale optimization space. To address this
issue, two versions of multi-objective optimization algorithms are proposed in
this paper. By utilizing a multi-population strategy and an adaptive
exploration technique, the methods efficiently explore a large-scale
decision-variable space. In order to mitigate the issue of an overcrowded
deployment of RSUs, a calibrating mechanism is adopted to adjust RSU density
during the optimization procedures. The proposed methods also take care of data
offloading between vehicles and RSUs by setting up an iterative best response
sequence game (IBRSG). By comparing the proposed algorithms with several
state-of-the-art algorithms, the results demonstrate that our strategies
perform better in both high-density and low-density urban scenarios. The
results also indicate that the proposed solutions substantially improve the
efficiency of vehicular networks.

提出了两个多目标优化算法以解决城市车辆网络中路侧单元 (RSU) 部署的难题，结果表明这些方法在高密度和低密度城市场景中均表现出更好的性能，从而大大提高了车辆网络的效率。