Shrinking LLMs with Self-Compression
In this work, we take the idea further still with a block-based sparsity pattern. The sections that follow describe how Self-Compression is woven into the baseline model, examine the weight patterns it produces, and discuss implications for resource-constrained deployment. Our…