Biography

Kazuki Fujii is a second-year Master’s student at the Institute of Science Tokyo, specializing in the intersection of High-Performance Computing (HPC) and Machine Learning. His research centers on distributed training of large-scale models and low-precision training techniques, including FP8 and Block-wise FP8, to optimize computational efficiency and model performance. As a core contributor to the Swallow Project, a leading initiative for developing open-source Japanese large language models (LLMs), Kazuki plays a pivotal role in maintaining pre-training libraries and conducting experiments to advance LLM training.

With extensive experience as a research intern and engineer, Kazuki has contributed to cutting-edge projects across multiple organizations in Tokyo, Japan. His work includes developing LLM and vision-language model (VLM) training libraries, orchestrating cluster deployments on Google Cloud, and managing internal cluster environments.