L1RA uses L1 regularization to prune and reallocate LoRA ranks across adapters during fine-tuning, achieving small perplexity gains on two 7B to 8B models at nearly the same training time.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
L1RA: Dynamic Rank Assignment in LoRA Fine-Tuning
L1RA uses L1 regularization to prune and reallocate LoRA ranks across adapters during fine-tuning, achieving small perplexity gains on two 7B to 8B models at nearly the same training time.