Research map / Architectures and efficiency / Attention and transformers
Log-Linear Attention
Han Guo, Songlin Yang, Tarushii Goel, Eric P. Xing, Tri Dao, Yoon Kim
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 log-linear hierarchical state | Mamba-2 · MQAR (dimension 16) | accuracy: 46.9 -> 55.9 | Improved |
| add log-linear hierarchical state | Gated DeltaNet · MQAR (dimension 32) | accuracy: 79.0 -> 84.4 | Improved |
| add log-linear hierarchical state | Gated DeltaNet · MQAR (dimension 64) | accuracy: ≥ 99 -> ≥ 99 | No clear effect |
| add log-linear hierarchical state | Mamba-2 · LAMBADA | perplexity: 24.14 -> 21.86 | Improved |
| add log-linear hierarchical state | Mamba-2 · WikiText | perplexity: 22.44 -> 22.11 | Improved |
| add log-linear hierarchical state | Gated DeltaNet · LAMBADA | perplexity: 19.71 -> 18.09 | Improved |
| add log-linear hierarchical state | Gated DeltaNet · LM-Eval commonsense average | accuracy: 45.0 -> 45.5 | Improved |
| add log-linear hierarchical state | Mamba-2 · LM-Eval commonsense average | accuracy: 44.8 -> 44.9 | No clear effect |
| add log-linear hierarchical state | Mamba-2 · S-NIAH-1 (8K) | accuracy: 56.8 -> 99.8 | Improved |
| add log-linear hierarchical state | Mamba-2 · S-NIAH-2 (16K) | accuracy: 18.6 -> 12.8 | Hurt |
| add log-linear hierarchical state | Gated DeltaNet · MK-NIAH-1 (4K) | accuracy: 23.0 -> 49.4 | Improved |
| add log-linear hierarchical state | Gated DeltaNet · S-NIAH-3 (4K) | accuracy: 66.2 -> 48.8 | Hurt |
Closest papers by meaning
- ZeroS: Zero‑Sum Linear Attention for Efficient Transformers (NeurIPS 2025 · similarity 0.89)
- Gated Linear Attention Transformers with Hardware-Efficient Training (ICML 2024 · similarity 0.89)
- Learning Linear Attention in Polynomial Time (NeurIPS 2025 · similarity 0.88)
- Memorization Capacity of Multi-Head Attention in Transformers (ICLR 2024 · similarity 0.88)
- Linear Log-Normal Attention with Unbiased Concentration (ICLR 2024 · similarity 0.88)
- Rethinking Attention: Exploring Shallow Feed-Forward Neural Networks as an Alternative to Attention Layers in Transformers (Student Abstract) (AAAI 2024 · similarity 0.88)
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.
