Research map / Large language models / Language, safety and interpretability
Confident Adaptive Language Modeling
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| copy hidden states d^j (not projected K,V) to skipped layers (ablation) | Oracle early exit · CNN/DM val | ROUGE-L: 23.02 -> 38.24 | Helped |
| early exit with oracle confidence and hidden-state copying vs full model | T5 8-layer · CNN/DM val | ROUGE-L: 38.32 -> 38.24 | No clear effect |
| use decaying threshold instead of a single fixed threshold (ablation) | CALM softmax · CNN/DM val | textual similarity vs efficiency tradeoff | Helped |
| train early-exit classifier with independent per-layer loss instead of geometric-like objective (ablation) | early-exit classifier · CNN/DM val | F1 at layer 1: .59 -> .62 | Helped |
| independent vs geometric-like classifier training (ablation) | early-exit classifier · WMT EN-FR val | F1 at layer 1: .33 -> .51 | Helped |
| softmax confidence measure vs hidden-state saturation | CALM · CNN/DM test, textual delta=0.1 | avg decoder layers | Helped |
| early-exit classifier vs softmax confidence | CALM · CNN/DM test, risk delta=0.05 | speedup / FLOPs | Mixed |
| hidden-state saturation measure vs static layer baseline | CALM · CNN/DM, WMT, SQuAD | performance-efficiency tradeoff | Mixed |
| relax tolerance delta (0.1 to 0.25) | CALM softmax, textual consistency · CNN/DM test | avg decoder layers | Helped |
Closest papers by meaning
- Improving Open-Ended Text Generation via Adaptive Decoding (ICML 2024 · similarity 0.90)
- Confidence-Based Decoding is Provably Efficient for Diffusion Language Models (NeurIPS 2026 · similarity 0.89)
- Adaptive Draft-Verification for Efficient Large Language Model Decoding (AAAI 2025 · similarity 0.89)
- Free Draft-and-Verification: Toward Lossless Parallel Decoding for Diffusion Large Language Models (NeurIPS 2026 · similarity 0.89)
- Reference Trustable Decoding: A Training-Free Augmentation Paradigm for Large Language Models (NeurIPS 2024 · similarity 0.89)
- Learning How Hard to Think: Input-Adaptive Allocation of LM Computation (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.