fdt_sparse_single_component
Published in Figshare • Jan 1, 2024
NobleIDNI2P45W43R02S43
Authors:
Anonymous Anonymouw
Abstract
Boxplot evaluation of FDT (Y-Axis) when pruning lowest weights of components [(2xMLP+4xAttention)*40 layers] individually on LLama2-13B (X-Axis). It can be observed that the 75% quantile contains the most variance.
Finding related papers...
Discussions
(0)No comments yet
Be the first to share your thoughts!