Gradient-based proportional LoRA rank allocation under GRPO reduces accuracy by 4.5 points versus uniform allocation because GRPO gradients are flatter across layers and non-uniform ranks amplify importance differences.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Gradient-Based LoRA Rank Allocation Under GRPO: An Empirical Study
Gradient-based proportional LoRA rank allocation under GRPO reduces accuracy by 4.5 points versus uniform allocation because GRPO gradients are flatter across layers and non-uniform ranks amplify importance differences.