CAS Distinguished Lecture - J. K. Eshraghian UC Santa Cruz - Neuromorphic LLMs

Distinguished Lecture by Jason K. Eshraghian

Neuromorphic Large Language Models (LLMs)

Abstract: The brain is the perfect place to look for inspiration to develop more efficient neural networks. Inspired by the recurrent dynamics of biological neurons, this talk will present several frontier reasoning LLMs developed in my lab, from software to device deployments. Trained end-to-end in an academic lab on a full production pipeline (data curation, pre-training, to post-training and alignment) these models surpass all leading LLMs from Meta, Google and every other over-resourced company in the ~10-billion parameter regime, despite being ~5x smaller. We have deployed several of our models on neuromorphic hardware at 2-watts, bringing SoTA-level reasoning from the datacenter to the edge.