How Much Can a LoRA Adapter Memorize? Measuring Adapter Capacity in Bits

arXiv:2607.21351 · cs.LG · Submitted 2026-07-23 · Read on arXiv

cs.LG

Submitted: 2026-07-23

Updated: 2026-10-04

Terminology

• Fine-Tuning — 3× in this paper · explained in 5 episodes
• Reinforcement Learning — 1× in this paper · explained in 4 episodes
• Verifiable Rewards — 1× in this paper · explained in 2 episodes

Sources

• A Little Rank Goes a Long Way: Random Scaffolds with LoRA Adapters Are All You Need
• Learning to Reason in 13 Parameters
• Qwen2.5 Technical Report
• DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models
• Leaner Training, Lower Leakage: Revisiting Memorization in LLM Fine-Tuning with LoRA
• How LoRA Remembers? A Parametric Memory Law for LLM Finetuning

Related papers

• Polynomial-Augmented Neural Networks (PANNs) with Weak Orthogonality Constraints for Enhanced Function and PDE Approximation
• AIRL-S: Unifying Reinforcement Learning and Search-Based Test-Time Scaling via Adversarial Inverse Reinforcement Learning
• Transformers as Bayesian In-Context Experimenters: Smoothness-Adaptive Efficient ATE Estimation
• Convergence issues in Relational Concept Analysis based on AOC-posets
• Beliefs Beyond Posteriors: The Predictive Cost of ELBO Binding
• Understanding Diffusion Models via Ratio-Based Function Approximation with SignReLU Networks