pruning

Efficient Pruning of Large Language Models

Ivan Ilin, Ph.D. Student, Computer, Electrical and Mathematical Sciences and Engineering

Mar 10, 12:00 - 13:00

B9 L2 R2325

LLM pruning machine learning

Thanos, a novel weight-pruning algorithm, efficiently reduces the size and improves performance of large language models by removing redundant weights using a block-wise, adaptively masked strategy that supports flexible sparsity patterns and achieves state-of-the-art results.