Research map / Large language models / Language, safety and interpretability
SparseEval: Efficient Evaluation of Large Language Models by Sparse Optimization
Taolin Zhang, Hang Guo, Wang Lu, Tao Dai, Shu-Tao Xia, Jindong Wang
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 anchor refinement (AIS/CIS swaps) over k-means initialization (ablation) | SparseEval · ARC (20 anchors) | MAE (%): 1.945 -> 1.778 | Improved |
| add anchor refinement over random initialization (ablation) | SparseEval · ARC (20 anchors) | MAE (%): 3.131 -> 1.778 | Improved |
| add anchor refinement over k-means initialization (ablation) | SparseEval · GSM8K (20 anchors) | MAE (%): 3.544 -> 3.305 | Improved |
| add anchor refinement over k-means initialization (ablation) | SparseEval · HellaSwag (20 anchors) | MAE (%): 1.631 -> 1.477 | Improved |
| add anchor refinement over random/k-means initialization (ablation) | SparseEval · MMLU (20 anchors) | MAE (%): 2.152 -> 1.718 | Improved |
| add anchor refinement over k-means initialization (ablation) | SparseEval · TruthfulQA (20 anchors) | MAE (%): 2.107 -> 1.756 | Improved |
| add anchor refinement over random/k-means initialization (ablation) | SparseEval · Winogrande (20 anchors) | MAE (%): 2.617 -> 2.088 | Improved |
| add anchor refinement over k-means initialization with 100 anchors (ablation) | SparseEval · HellaSwag (100 anchors) | MAE (%): 0.772 -> 0.827 | Hurt |
| use MLP weight estimator instead of linear weights (ablation) | SparseEval · GSM8K | MAE and Kendall tau | Improved |
| reduce training data to 20% of models (ablation) | SparseEval · LLM benchmarks | MAE and Kendall tau: MAE under 1% and tau higher than 0.90 | Mixed |
| increase anchors from 80 to 100 for Anchor Points | Anchor Points · ARC | MAE (%): 2.289 -> 10.620 | Hurt |
| use SparseEval instead of TailoredBench on unseen DeepSeek models | SparseEval · Hold-out DeepSeek families | average MAE (%): 4.216 -> 2.090 | Improved |
Closest papers by meaning
- SparseLLM: Towards Global Pruning of Pre-trained Language Models (NeurIPS 2024 · similarity 0.91)
- Adaptive Layer Sparsity for Large Language Models via Activation Correlation Assessment (NeurIPS 2024 · similarity 0.91)
- Pruning Large Language Models with Semi-Structural Adaptive Sparse Training (AAAI 2025 · similarity 0.90)
- Weak-to-Strong Search: Align Large Language Models via Searching over Small Language Models (NeurIPS 2024 · similarity 0.90)
- SPP: Sparsity-Preserved Parameter-Efficient Fine-Tuning for Large Language Models (ICML 2024 · similarity 0.90)
- SoLA: Leveraging Soft Activation Sparsity and Low-Rank Decomposition for Large Language Model Compression (AAAI 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.
