Research map / Architectures and efficiency / Pruning and efficient tuning
L2-LoRA: Improving Low-Rank Adaptation with Layer-Specific Regularization
Xiang Zhang, Rui Xie, Shikun Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add layer-specific L2 regularization to LoRA | LoRA (Llama-7B) · MMLU avg after Commonsense170K | accuracy: 23.1 -> 28.3 | Improved |
| add layer-specific L2 regularization to LoRA | LoRA (Llama-7B) · ARC-c | accuracy: 61.3 -> 65.6 | Improved |
| add layer-specific L2 regularization in continual learning | LoRA (Llama-7B) · MMLU avg, stage 1 | accuracy: 25.7 -> 28.3 | Improved |
| add layer-specific L2 regularization in continual learning | LoRA (Llama-7B) · MMLU avg, stage 3 | accuracy: 23.3 -> 23.6 | No clear effect |
| add layer-specific L2 regularization | LoRA (Llama2-7B) · PIQA | accuracy: 79.9 -> 85.2 | Improved |
| add layer-specific L2 regularization | LoRA (Llama2-7B) · BoolQ | accuracy: 69.8 -> 68.9 | Hurt |
| add layer-specific L2 regularization | LoRA (Llama2-7B) · GSM8K | accuracy: 60.58 -> 65.32 | Improved |
| add layer-specific L2 regularization | LoRA (Llama2-7B) · MATH | accuracy: 16.88 -> 18.12 | Improved |
| fine-tune only lowest layers vs highest layers (ablation) | LoRA (Llama2-7B) · MMLU | accuracy: 43.95 -> 41.71 (low, 6 layers); 43.88 (high) | Hurt |
| fine-tune highest layers instead of lowest layers (ablation) | LoRA (Llama2-7B) · AddSub | accuracy: 11.14 -> 17.22 | Improved |
Closest papers by meaning
- LoRASculpt: Sculpting LoRA for Harmonizing General and Specialized Knowledge in Multimodal Large Language Models (CVPR 2025 · similarity 0.92)
- Beyond LoRA vs. Full Fine-Tuning: Gradient-Guided Optimizer Routing for LLM Adaptation (NeurIPS 2026 · similarity 0.92)
- Learning Rate Matters: Vanilla LoRA May Suffice for LLM Fine-tuning (NeurIPS 2026 · similarity 0.92)
- LoRA vs Full Fine-tuning: An Illusion of Equivalence (NeurIPS 2025 · similarity 0.91)
- Uni-LoRA: One Vector is All You Need (NeurIPS 2025 · similarity 0.91)
- Train Small, Infer Large: Memory-Efficient LoRA Training for Large Language Models (ICLR 2025 · similarity 0.90)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
