Research map / Architectures and efficiency / RNNs and normalization
UniScale: Adaptive Unified Inference Scaling via Online Joint Optimization of Model Routing and Test-Time Scaling
Kaiyu Huang, Xingyu Wang, Mingze Kong, Zhubo Shi, Yuqian Hou, Hong Xu, Zhongxiang Dai, Minchen Yu, Qingjiang Shi
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 semantic action representations (ablation) | UniScale · UIS Q-Pri | cumulative regret @210: 38.24 -> 33.98 | Improved |
| add semantic action representations (ablation) | UniScale · UIS C-Sen | cumulative regret @210: 41.56 -> 34.43 | Improved |
| add semantic action representations (ablation) | UniScale · UIS Q-Pri | accuracy: 50.62 -> 57.50 | Improved |
| add semantic action representations (ablation) | UniScale · UIS C-Sen | accuracy: 46.88 -> 46.25 | No clear effect |
| add semantic action representations (ablation) | UniScale · UIS C-Sen | cost: 629.7 -> 48.2 | Improved |
| add dense verification feedback (ablation) | UniScale · UIS Q-Pri | accuracy: 51.88 -> 57.50 | Improved |
| add dense verification feedback (ablation) | UniScale · UIS Q-Pri | cost: 2674.9 -> 1303.3 | Improved |
| add dense verification feedback (ablation) | UniScale · UIS C-Sen | accuracy: 43.75 -> 46.25 | Improved |
| add path-aware early exiting (ablation) | Standard TTS · all UIS configs | accuracy: 65.43 -> 64.52 | No clear effect |
| add path-aware early exiting (ablation) | Standard TTS · all UIS configs | cost: 4851.4 -> 1002.9 | Improved |
| use UniScale instead of k-NN router under model removal (ablation) | UniScale · non-stationary Rem. env | accuracy: 50.62 -> 53.12 | Improved |
| use joint model routing + TTS space instead of routing alone | UniScale · Cost-Sensitive | reward: 0.6196 -> 0.7079 | Improved |
Closest papers by meaning
- Inference Scaling Laws: An Empirical Analysis of Compute-Optimal Inference for LLM Problem-Solving (ICLR 2025 · similarity 0.91)
- Scaling Inference-Efficient Language Models (ICML 2025 · similarity 0.91)
- Scaling Laws Meet Model Architecture: Toward Inference-Efficient LLMs (ICLR 2026 · similarity 0.91)
- OptScale: Probabilistic Optimality for Inference-time Scaling (AAAI 2026 · similarity 0.89)
- Universal Model Routing for Efficient LLM Inference (ICLR 2026 · similarity 0.89)
- Inference-Aware Fine-Tuning for Best-of-N Sampling in Large Language Models (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.
