Research map / Architectures and efficiency / Pruning and efficient tuning
Learning Rate Scaling across LoRA Ranks and Transfer to Full Finetuning
Nan Chen, Soledad Villar, Soufiane Hayou
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 |
|---|---|---|---|
| scale LoRA alpha as r^-1 instead of 1 under Init[A] (ablation) | LoRA Init[A] · SFT across five models | optimal learning rate vs rank | Improved |
| use Init[B] with alpha=1 instead of Init[A] (ablation) | LoRA · SFT, RLVR, diffusion | LoRA-to-FFT optimal learning rate alignment | Improved |
| increase learning rate 2x beyond optimum (ablation) | LoRA Init[A] alpha=1 · Qwen2.5-3B-Instruct, RoBERTa-large | training loss | Hurt |
| use Init[B] alpha=1 versus Init[A] alpha=1 for large-LR tolerance (ablation) | LoRA · SFT | learning rate stability | Improved |
| choose suboptimal learning rate (ablation) | LoRA/FFT GRPO · Llama-3.1-8B GSM8k | wall-clock training time: 2--4x longer | Hurt |
| use LoRA r=256 instead of full finetuning (ablation) | Llama-3.1-8B GRPO · Llama-3.1-8B GSM8k | peak GPU memory: FFT over 2x LoRA memory | Improved |
| alpha in {2, 4} instead of 1 under Init[A] (ablation) | LoRA Init[A] · SFT | scaling rule r^-1/2 | No clear effect |
Closest papers by meaning
- LoRA+: Efficient Low Rank Adaptation of Large Models (ICML 2024 · similarity 0.93)
- SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation (NeurIPS 2026 · similarity 0.92)
- Flat-LoRA: Low-Rank Adaptation over a Flat Loss Landscape (ICML 2025 · similarity 0.92)
- LoRA Training Provably Converges to a Low-Rank Global Minimum Or It Fails Loudly (But it Probably Won't Fail) (ICML 2025 · similarity 0.92)
- Learning Rate Matters: Vanilla LoRA May Suffice for LLM Fine-tuning (NeurIPS 2026 · similarity 0.91)
- The Expressive Power of Low-Rank Adaptation (ICLR 2024 · similarity 0.91)
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.
