# When Gradient Importance Lies: Adaptive LoRA Rank Allocation Fails Under GRPO

arXiv: https://arxiv.org/abs/2605.07366 · HTML: https://mlsanity.com/paper/2605.07366/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 6 of its 11 references that are in the map.
- First posted on arXiv 8 May 2026.

## Builds on (most cited first, 6 of 6)

- [Training Verifiers to Solve Math Word Problems](https://mlsanity.com/paper/2110.14168/) (2021)
- [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://mlsanity.com/paper/2402.03300/) (2024)
- [Understanding Layer Significance in LLM Alignment](https://mlsanity.com/paper/2410.17875/) (2024)
- [IGU-LoRA: Adaptive Rank Allocation via Integrated Gradients and Uncertainty-Aware Scoring](https://mlsanity.com/paper/2603.13792/) (2026)
- [Aletheia: Gradient-Guided Layer Selection for Efficient LoRA Fine-Tuning Across Architectures](https://mlsanity.com/paper/2604.15351/) (2026)
- [Why Is RLHF Alignment Shallow? A Gradient Analysis](https://mlsanity.com/paper/2603.04851/) (2026)

## Built on by (most cited first, 1 of 1)

- [How Many Ranks Does a Token Need? A Noise-Floor Squelch for Rank-One LoRA Experts](https://mlsanity.com/paper/2610.19033/) (2026)

## Abstract (arXiv)

Adaptive rank allocation for LoRA - allocating more parameters to important layers and fewer to unimportant ones - consistently improves efficiency under supervised fine-tuning (SFT). We test whether this success transfers to reinforcement learning, specifically Group Relative Policy Optimization (GRPO). Using gradient-magnitude profiling on Qwen 2.5 1.5B with GSM8K, we find that, in our setting, it does not: proportional rank allocation degrades accuracy by 4.5 points compared to uniform allocation (70.0% vs. 74.5%), despite using identical parameter budgets. We identify two mechanisms behind this failure. First, the gradient landscape under GRPO is fundamentally flatter than under SFT: the max-to-min layer importance ratio is only 2.17x, whereas the layer concentration reported by Shi et al. (2024) for SFT (top 30% of layers carrying >80% of the gradient signal) implies a max/min ratio well above 10x. All layers carry meaningful gradient signal; none are truly idle. Second, we observe a gradient amplification effect: non-uniform allocation widens the importance spread from 2.17x to 3.00x, creating a positive feedback loop where high-rank layers absorb more gradient while low-rank layers are progressively silenced. A random-allocation control yields the same amplification (r=0.972 correlation between assigned rank and resulting gradient share), indicating that rank causally determines gradient importance rather than the reverse. The negative result is single-seed and single-task; we present it as preliminary evidence that gradient importance does not predict capacity requirements under RL, and that naive transfer of SFT-era rank allocation strategies to alignment training should be evaluated cautiously.