Research map / Architectures and efficiency / Pruning and efficient tuning
LoFT: Low-Rank Adaptation That Behaves Like Full Fine-Tuning
Nurbek Tastan, Stefanos Laskaridis, Martin Takáč, Karthik Nandakumar, Samuel Horváth
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 |
|---|---|---|---|
| use LoFT (r=16) instead of LoRA (r=16) | LLaMA-7B · Commonsense reasoning avg | accuracy: 73.57 -> 76.08 | Improved |
| use LoFT (r=16) instead of LoRA (r=16) | LLaMA2-7B · Commonsense reasoning avg | accuracy: 68.69 -> 80.46 | Improved |
| use LoFT (r=16) instead of DoRA (r=16) | LLaMA2-7B · Commonsense reasoning avg | accuracy: 79.71 -> 80.46 | Improved |
| use LoFT (r=16) instead of LoRA (r=16) | LLaMA3-8B · Commonsense reasoning avg | accuracy: 84.97 -> 85.63 | Improved |
| reduce LoFT rank from 16 to 1 | LLaMA-7B · Commonsense reasoning avg | accuracy: 76.08 -> 72.17 | Hurt |
| reduce LoFT rank from 16 to 1 | LLaMA3-8B · Commonsense reasoning avg | accuracy: 85.63 -> 82.42 | Hurt |
| use LoFT (r=16) instead of LoRA (r=16) | ViT-Base · 4 image datasets avg | accuracy: 75.46 -> 76.12 | Improved |
| use LoFT (r=16) instead of full fine-tuning | ViT-Base · 4 image datasets avg | accuracy: 75.86 -> 76.12 | No clear effect |
| reduce LoFT rank from 16 to 4 | ViT-Base · 4 image datasets avg | accuracy: 76.12 -> 74.27 | Hurt |
| add optimizer state calibration (ablation) | LoFT on GPT-2 · WikiText-2 | perplexity | Improved |
| add alternating updates (ablation) | LoFT on GPT-2 · WikiText-2 | perplexity | Hurt |
| add second-moment calibration (ablation) | LoFT · commonsense reasoning memory and accuracy | accuracy: ~0.1% gain; memory up to 25.65% vs less than 6% over LoRA | No clear effect |
Closest papers by meaning
- 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 0.91)
- LoRA-GA: Low-Rank Adaptation with Gradient Approximation (NeurIPS 2024 · similarity 0.91)
- Asymmetry in Low-Rank Adapters of Foundation Models (ICML 2024 · similarity 0.91)
- LoRA-Pro: Are Low-Rank Adapters Properly Optimized? (ICLR 2025 · similarity 0.90)
- The Expressive Power of Low-Rank Adaptation (ICLR 2024 · 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.
