SwiftPrune prunes LLMs in seconds using a Hessian-free importance metric plus an EWMA threshold, with an O(n) pruning pass and strong 2:4 structured sparsity results.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
SwiftPrune: Hessian-Free Weight Pruning for Large Language Models
SwiftPrune prunes LLMs in seconds using a Hessian-free importance metric plus an EWMA threshold, with an O(n) pruning pass and strong 2:4 structured sparsity results.