Research map / Architectures and efficiency / RNNs and normalization
Singular Proxies for Adaptive Caching in Diffusion Language Models
Wenhao Sun, Rong-Cheng Tu, Yifu Ding, Zhao Jin, Jingyi Liao, Yongcheng Jing, Dacheng Tao
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 |
|---|---|---|---|
| replace Value proxy with singular proxy (rank 128) (ablation) | SPA-Cache · GSM8K, LLaDA-8B | TPS: 165 -> 179 | Improved |
| replace Value proxy with singular proxy (rank 128) (ablation) | SPA-Cache · GSM8K, LLaDA-8B | accuracy (%): 78.59 -> 78.23 | No clear effect |
| add adaptive layer-wise budget allocation (ablation) | SPA-Cache · GSM8K, LLaDA-8B | TPS: 179 -> 189 | Improved |
| use uniform 16% update ratio instead of adaptive allocation (ablation) | SPA-Cache · GSM8K, LLaDA-8B | accuracy (%): 78.23% to 75.56% | Hurt |
| lower proxy rank from 512 to 32 (ablation) | SPA-Cache · GSM8K, LLaDA-8B | accuracy (%): 78.57 -> 76.71 accuracy, 172.56 -> 182.66 TPS | Mixed |
| use Attention output as identifier instead of Value (ablation) | SPA-Cache · identification accuracy | identification accuracy | Hurt |
| add SPA-Cache to LLaDA-8B | LLaDA-8B-Instruct · MBPP | TPS: 5.75 -> 46.12 | Improved |
| add SPA-Cache to LLaDA-8B | LLaDA-8B-Instruct · GSM8K | TPS: 29.67 -> 190.73 | Improved |
| add SPA-Cache to LLaDA-8B | LLaDA-8B-Instruct · GSM8K | accuracy (%): 78.62 -> 78.24 | No clear effect |
| add Fast-dLLM instead of SPA-Cache to Dream on HumanEval accuracy | Fast-dLLM · HumanEval, Dream-7B | accuracy (%): 58.54 -> 53.05 | Hurt |
| combine SPA-Cache with parallel decoding | SPA-Cache · BBH, LLaDA-8B | TPS: 108.31 -> 693.96 | Improved |
| use SPA-Cache instead of Fast-dLLM with parallel decoding | SPA-Cache · BBH, LLaDA-8B | TPS: 301.33 -> 693.96 | Improved |
Closest papers by meaning
- dKV-Cache: The Cache for Diffusion Language Models (NeurIPS 2025 · similarity 0.93)
- d$^2$Cache: Accelerating Diffusion-Based LLMs via Dual Adaptive Caching (ICLR 2026 · similarity 0.93)
- EntropyCache: Decoded Token Entropy Guided KV Caching for Diffusion Language Models (NeurIPS 2026 · similarity 0.92)
- Polestar: Drift-Aware Cache Calibration and Token Commitment for Efficient Inference of Diffusion LLMs (NeurIPS 2026 · similarity 0.90)
- From Next-Token to Next-Block: A Principled Adaptation Path for Diffusion LLMs (NeurIPS 2026 · similarity 0.89)
- CAKE: Cascading and Adaptive KV Cache Eviction with Layer Preferences (ICLR 2025 · similarity 0.89)
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.
