LoRA Rank Error Bounds for Transformer Attention
How Much Rank Does LoRA Need? Rank-Error Bounds for Transformer Attention
The theoretical bounds help practitioners choose optimal LoRA ranks for efficient Transformer fine-tuning, reducing computational costs while maintaining performance.
This paper provides a task-dependent theory of approximation error achievable at each LoRA rank for Transformer attention. The researchers bound the smallest expected KL error for rank-r query LoRA updates, proving lower bounds proportional to ψ(||d||₂) when target attention probabilities are bounded away from zero. They also establish unconditional upper bounds of min{||d||₂²/4,√2||d||₂} and extend analysis to fused multi-head LoRA and joint query/key updates.
How Much Rank Does LoRA Need? Rank-Error Bounds for Transformer Attention
Choosing the rank of a low-rank adaptation (LoRA) update is usually an empirical task. In this paper, we provide a task-dependent theory of the approximation error achievable at each LoRA rank for Transformer attention. We fix a pretrained attention head, a target attention function, and a distribution over inputs from the downstream task, and bound the smallest expected Kullback--Leibler (KL) error achievable by a rank-$r$ query LoRA update. When target attention probabilities are bounded away from zero, we prove a lower bound of the error proportional to $ψ(\|d\|_2)$, where $d$ is the difference between candidate and target attention scores and $ψ(t)=\min\{t^2,t\}$. We also prove an unconditional upper bound $\min\{\|d\|_2^2/4,\sqrt2\|d\|_2\}$. Under explicit realizability, geometry, and moment conditions, we then bound the best rank-$r$ error between an explicit multiple of $ψ(\sqrt{T_r})$ and $\min\{T_r/4,\sqrt{2T_r}\}$, where $T_r$ is the downstream-weighted tail energy of the target update. We also provide target-Fisher bounds when candidate scores remain within a fixed range of the target scores, and an unrestricted lower bound when a subset of tokens carries most of the probability mass. These spectral bounds describe finite-score approximation. We then construct explicit families in which softmax saturation makes the rank required to match the attention function strictly smaller than the rank required to match the finite logits. Finally, we extend the analysis to fused multi-head LoRA and joint query/key updates, exposing the effects of rank sharing and query/key factorization constraints.
