A static sparsity mask chosen by plain gradients matches or beats LoRA and second-order salience metrics across NLP fine-tuning benchmarks.
https://www.amd.com/en/products/accelerators/instinct/mi300.html
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Refining Salience-Aware Sparse Fine-Tuning Strategies for Language Models
A static sparsity mask chosen by plain gradients matches or beats LoRA and second-order salience metrics across NLP fine-tuning benchmarks.