Initializing LoRA adapters from the SVD of the first full fine-tuning gradient yields subspace alignment and fast convergence in theory, and the resulting LoRA-One method beats standard LoRA on several LLM benchmarks.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
stat.ML 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
LoRA-One: One-Step Full Gradient Could Suffice for Fine-Tuning Large Language Models, Provably and Efficiently
Initializing LoRA adapters from the SVD of the first full fine-tuning gradient yields subspace alignment and fast convergence in theory, and the resulting LoRA-One method beats standard LoRA on several LLM benchmarks.