Research map / Architectures and efficiency / Pruning and efficient tuning
LIFT: Language-Interfaced Fine-Tuning for Non-language Machine Learning Tasks
Tuan Dinh, Yuchen Zeng, Ruisu Zhang, Ziqian Lin, Michael Gira, Shashank Rajput, Jy-yong Sohn, Dimitris Papailiopoulos, Kangwook Lee
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 |
|---|---|---|---|
| fine-tune pretrained GPT-J on non-language tasks via sentence conversion (LIFT) vs logistic regression | LIFT/GPT-3 · circles classification | accuracy: 48.58 -> 81.17 | Improved |
| increase number of classes to 100 | LIFT/GPT-J · Margin | accuracy: 50.23 vs 81.98 (RBF-SVM) | Hurt |
| fine-tune on full dataset instead of in-context learning with few prompts (ablation) | LIFT/GPT-J · Vehicle | accuracy: 25.49 -> 64.31 | Improved |
| fine-tune rather than in-context learning at same sample count (ablation) | LIFT/GPT-3 · Vehicle | accuracy: 28.82 -> 23.73 | No clear effect |
| use pretrained GPT-J instead of random initialization (ablation) | LIFT/GPT-J · Wine | accuracy: 22.22 -> 93.52 | Improved |
| use LM pretrained on natural language instead of code (CodeGen) (ablation) | LIFT · Two Circles | accuracy: 53.02 -> 75.92 | Improved |
| fine-tune GPT-J on gibberish data first (ablation) | LIFT/GPT-J · Two Circles | accuracy: 75.92 -> 68.67 | Hurt |
| add correct feature names to prompt (ablation) | LIFT/GPT-3 · TAE | accuracy: 65.59 -> 69.89 | Improved |
| add correct feature names to prompt (ablation) | LIFT/GPT-3 · CMC | accuracy: 57.74 -> 57.40 | No clear effect |
| use shuffled feature names instead of correct ones (ablation) | LIFT/GPT-3 · TAE | accuracy: 69.89 -> 60.22 | Hurt |
| add Gaussian-noise data augmentation (eps=0.1) to training (ablation) | LIFT/GPT-J · MNIST Gaussian noise eps=0.1 | accuracy: 56.14 -> 94.98 | Improved |
| add Gaussian-noise data augmentation (eps=0.1) to training (ablation) | LIFT/GPT-J · MNIST clean | accuracy: 96.88 -> 93.78 | Hurt |
Closest papers by meaning
- LISA: Layerwise Importance Sampling for Memory-Efficient Large Language Model Fine-Tuning (NeurIPS 2024 · similarity 0.88)
- Tuning LayerNorm in Attention: Towards Efficient Multi-Modal LLM Finetuning (ICLR 2024 · similarity 0.88)
- Linearization Explains Fine-Tuning in Large Language Models (NeurIPS 2025 · similarity 0.88)
- Get more for less: Principled Data Selection for Warming Up Fine-Tuning in LLMs (ICLR 2024 · similarity 0.88)
- Cross-model Control: Improving Multiple Large Language Models in One-time Training (NeurIPS 2024 · similarity 0.87)
- Understanding Linear Probing then Fine-tuning Language Models from NTK Perspective (NeurIPS 2024 · similarity 0.87)
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.
