Wanda- or magnitude-ordered fixed sparse supports, alone or hybridized with LoRA under a matched budget, can outperform tested PEFT baselines on Math17K arithmetic fine-tuning.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Super-Tuning: From Activation-Aware Pruning to Sparse Fine-Tuning
Wanda- or magnitude-ordered fixed sparse supports, alone or hybridized with LoRA under a matched budget, can outperform tested PEFT baselines on Math17K arithmetic fine-tuning.