Fabian Klemm
Senior Consultant
TNG Technology Consulting GmbH
Germany
About
Dr. Fabian Klemm completed his doctorate at the Technical University of Munich (TUM) in discrete mathematics and applied geometry. There, he worked on clustering problems under constraints in general geometric spaces before switching to IT in 2020 and joining TNG as a software consultant. In addition to his extensive DevOps experience, Fabian has developed a strong interest in AI, especially large language models (LLMs), with a particular focus on mechanistic interpretability and LLM architectures. Fabian is currently part of the TNG AI research team that published the DeepSeek Chimera models. He also contributes to the TNG Skainet team, which operates TNG’s internal AI server rack.
Talk
Fabian Klemm | Large Models, Small Resources: Customizing LLMs
LLMs, , Open Weight Models, Model Merging, Training, Finetuning
The Germans just Frankensteined DeepSeek’s R1 and V3 into something called R1T Chimera.”Beyond this post on X, the DeepSeek-R1T and R1T2 Chimera models published by TNG have gained significant attention, with daily usage exceeding 10 billion tokens on OpenRouter.So, what kind of “Frankensteining” is going on here? How can a small software consultancy such as TNG produce its own models?An internal AI research group within TNG has been experimenting with and publishing research on mixture-of-experts (MoE) large language models (LLMs).The team started by manipulating the way experts work within a model under the name “Mixture of Tunable Experts.” It then continued with an assembly-of-experts model-merging process, which resulted in the Chimera models. Since then, TNG’s research team has been working on various ways to adapt state-of-the-art LLMs.In this talk, Fabian Klemm provides technical insights and reviews the most important results. He demonstrates how the models were engineered and shares anecdotes about the successes and setbacks encountered along the way.