Microscaling Floating Point Formats for Large Language Models

Marco Cococcioni, Dario Pagani, Federico Rossi

The increasing computational and memory demands of large language models (LLMs) necessitate innovative approaches to optimize resource usage without compromising performance. This…