ECHO-LLaMA: Efficient Caching for High-Performance LLaMA Training

Maryam Dialameh, Rezaul Karim, Hossein Rajabzadeh +5

This paper introduces ECHO-LLaMA, an efficient LLaMA architecture designed to improve both the training speed and inference throughput of LLaMA architectures while maintaining its…