A Scalable Measure of Loss Landscape Curvature for Analyzing the Training Dynamics of LLMs - Dayal Singh Kalra, Jean-Christophe Gagnon-Audet, Andrey Gromov, Ishita Mediratta, Kelvin Niu

A Scalable Measure of Loss Landscape Curvature for Analyzing the Training Dynamics of LLMs - Dayal Singh Kalra, Jean-Christophe Gagnon-Audet, Andrey Gromov, Ishita Mediratta, Kelvin Niu | Arena