PhD Thesis: Optimization of LLM / FM Inference in Industrial Edge Networks

Aufgaben:

The rise of Large Language Models (LLMs) and, more broadly, Foundation Models (FMs), has transformed the landscape of artificial intelligence (AI). In industrial settings, FMs can support use cases such as product quality control, or operational decision support. These models have primarily been deployed in cloud environments, which introduces latency, network traffic, cost, and privacy issues. Hence, they are increasingly finding their way into industrial edge systems. However, compared to data-center environments, industrial edge environments are characterized by heterogeneous machines, hindered by memory/compute limits and the communication overhead when a model is split across multiple devices. This motivates the need for network-aware, distributed inference strategies (e.g., efficiently partitioning models across multiple nodes [1] based on strategies such as pipeline or tensor parallelism [2], potentially integrating with existing frameworks [3], supporting profiling driven scheduling [4], or compression approaches for transformers [5]). Hence, the goal of this thesis is to optimize the inference of distributed LLMs / FMs on the edge. The selected PhD student will be co-supervised by a Siemens researcher and a professor of a European university. The work on the PhD includes multiple iterations of a) investigation of state of the art and related work, b) clear definition and scoping of problem space, c) development of novel approach, d) implementation of an industrial demonstrator to apply the approach, and e) the evaluation of the implemented approach against a known baseline. [1] https://arxiv.org/pdf/2405.14371 [2] https://docs.pytorch.org/tutorials/beginner/dist_overview.html [3] https://github.com/ggml-org/llama.cpp [4] https://dl.acm.org/doi/pdf/10.1145/3812836.3814999 [5] https://arxiv.org/pdf/2507.12145

Qualifikation:

Education : You have completed or are about to complete a Master's degree in Computer Science or a related field Experience and Skills : You possess excellent programming skills in Python and expertise in C++ is a plus You have a strong understanding of Large Language Models (LLMs) and Foundation Models (FMs), including their underlying principles You are familiar with Linux environments and network management, and have a strong interest in AI and communication networks Ways of Working : You work independently as well as collaboratively within a team, demonstrating initiative and reliability Languages : Very good English skills are required

Wir bieten:

Attractive remuneration package 30 leave days and a variety of flexible working models that allow time off for yourself and your family Share matching programs to become a shareholder of Siemens AG Continuous Learning: Benefit from specialized training and daily challenges to keep your expert knowledge up-to-date Innovative Environment: Be part of a team that values innovation and continuous improvement

Über Uns:

Since each of over 300,000 employees feels that other benefits are particularly important, and we cannot list our entire benefit portfolio here, you can find more information  here . At Siemens, we believe that feeling valued and included is the foundation for doing great work. That's why we aim to create an inclusive workplace where everyone feels a sense of belonging, and where individual perspectives and experiences are celebrated. Our commitment to fairness and respect extends to every applicant. As an equal opportunity employer , we welcome applications from individuals of all backgrounds and particularly encourage applications from persons with disabilities. The world never stands still. And new challenges arise every day. With a passion for questioning things, for supplying ideas, and intelligently driving things forward we are helping society move towards a smarter tomorrow. Be it with technologies that reduce carbon emissions in cities or hyperintelligent robots. This is how we are able, to tackle the most important projects and push them forward together. Help us shape the future.