Mergenetic: a Simple Evolutionary Model Merging Library

Abstract: Model merging allows combining the capabilities of existing models into a new one - post hoc, without additional training. This has made it increasingly popular thanks to its low cost and the availability of libraries that support merging on consumer GPUs. Recent work shows that pairing merging with evolutionary algorithms can boost performance, but no framework currently supports flexible experimentation with such strategies in LLMs. We introduce Mergenetic, an open-source library for evolutionary model merging. Mergenetic enables easy composition of merging methods and evolutionary algorithms while incorporating lightweight fitness estimators to reduce evaluation costs. We describe its design and demonstrate that Mergenetic produces competitive results across tasks and languages using modest hardware.

Overview of LuaLaTeX and XeLaTeX Template for *ACL Style Files

The document titled "LuaLaTeX and XeLaTeX Template for *ACL Style Files" presents an example of the utilization of *ACL style files with LuaLaTeX and XeLaTeX typesetting systems. The primary aim of the paper is to provide an exemplary structure for including foreign language text in academic publications, especially those adhering to the *ACL standard formatting guidelines.

Introduction

The introduction serves as a reference point to the file acl_latex.tex, which encapsulates general instructions for formatting adherence in *ACL publications. Within this context, the authors demonstrate the inclusion of foreign languages, specifically Hindi and Arabic, using the babel package. This capability is essential for researchers aiming to incorporate multilingual elements within their conference and journal submissions without compromising the structural integrity of the document.

Technical Implementation

The core implementation leverages babel provides to display text in various scripts, notably using Lohit Devanagari for Hindi and Noto Sans Arabic for Arabic script. The integration of these fonts into a LuaLaTeX or XeLaTeX document exemplifies the versatility and powerful capabilities of modern LaTeX systems in handling diverse character sets. Such functionality is increasingly significant in a globalized research environment, where scholars often need to present research findings in multiple languages.

The document concludes with a comprehensive reference section. This section underscores the importance of citation management and integration within LaTeX documents, aligning with the *ACL standards. Various citation formats are demonstrated, promoting proper academic referencing practices in computational linguistics and aligning with the *ACL's emphasis on rigorous citation standards.

Implications and Future Directions

The implications of the research are substantial for the computational linguistics community, particularly for those involved with the *ACL conferences. The document template effectively addresses challenges associated with multilingual content presentation in technical papers—facilitating broader dissemination of research across linguistic boundaries.

In terms of future directions, evolving LaTeX capabilities suggest potential for enhancing integration with other language scripts and novel typesetting functionalities to address the needs of an expansive research domain. As the field progresses, improvements in multilingual support within LaTeX frameworks will further streamline the creation of inclusive and accessible academic materials.

Overall, while the document itself primarily serves as a template guide, it demonstrates the practical application and integration of multilingual text within *ACL style files—a necessity for the diversification and dissemination of research outputs in contemporary academia.