Inference-Time Algorithms for Large Language Models

Dear Colleagues,

The rapid advancements in large language models (LLMs) have revolutionized natural language processing, enabling a wide array of applications, from conversational AI to content generation and beyond. However, as these models grow in size and complexity, the challenges associated with their deployment at inference time become increasingly significant. This Special Issue seeks to explore novel algorithms, techniques, and frameworks that address these challenges, focusing specifically on improving the efficiency, robustness, and adaptability of LLMs during inference.

Inference-time algorithms play a critical role in the practical deployment of LLMs, impacting their performance in real-world applications. This Special Issue aims to bring together cutting-edge research that advances our understanding of how LLMs can be optimized at inference time, making them more effective and accessible for a broader range of users and tasks.

This Special Issue’s scope encompasses a broad array of topics related to inference-time algorithms for LLMs, including but not limited to the following:

• Methods for improving efficiency during inference, such as quantization, pruning, and distillation;
• Enhancing robustness and reliability in inference-time decision-making processes;
• Adaptive algorithms that tailor model behavior dynamically based on task-specific requirements or input characteristics;
• Strategies for reducing latency and computational costs in real-time applications;
• Addressing security and privacy concerns, including defenses against adversarial attacks at inference time;
• Comprehensive benchmarking and evaluation of inference-time algorithms across diverse real-world scenarios.

This Special Issue aims to position itself within the broader literature by highlighting the intersection of inference-time challenges with the ongoing development of LLMs. It seeks contributions that not only push the boundaries of current methodologies but also address the pressing need for the scalable and sustainable deployment of these powerful models. By fostering dialogue between researchers focused on model training and those working on deployment, this Special Issue will contribute to a more comprehensive understanding of the lifecycle of LLMs, ensuring their effective harnessing in practical, real-world settings.

We invite researchers, practitioners, and industry experts to submit their original contributions that address these key challenges, offering insights and innovations which will shape the future of LLM applications.

Dr. Yiwei Wang
Dr. Yujun Cai
Guest Editors

Manuscript Submission Information

Manuscripts should be submitted online at www.mdpi.com by registering and logging in to this website. Once you are registered, click here to go to the submission form. Manuscripts can be submitted until the deadline. All submissions that pass pre-check are peer-reviewed. Accepted papers will be published continuously in the journal (as soon as accepted) and will be listed together on the special issue website. Research articles, review articles as well as short communications are invited. For planned papers, a title and short abstract (about 100 words) can be sent to the Editorial Office for announcement on this website.

Submitted manuscripts should not have been published previously, nor be under consideration for publication elsewhere (except conference proceedings papers). All manuscripts are thoroughly refereed through a single-blind peer-review process. A guide for authors and other relevant information for submission of manuscripts is available on the Instructions for Authors page. Electronics is an international peer-reviewed open access semimonthly journal published by MDPI.

Please visit the Instructions for Authors page before submitting a manuscript. The Article Processing Charge (APC) for publication in this open access journal is 2400 CHF (Swiss Francs). Submitted papers should be well formatted and use good English. Authors may use MDPI's English editing service prior to publication or during author revisions.

• inference-time algorithms
• large language models (LLMs)
• efficiency
• robustness
• benchmarking
• security and privacy