AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

Sampling from Your Language Model One Byte at a Time

Jonathan Hayase, Alisa Liu, Noah A. Smith, Sewoong Oh

ICML 2026 · Large language models › Language, safety and interpretability · approach: Controlled text generation · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
use ByteSampler instead of no mitigation for next-character predictionOLMo2-1B · English next-character predictionnext character accuracy: 29.490 -> 81.560Improved
use ByteSampler instead of token healing/alignment backtrackingOLMo2-1B · English next-character predictionnext character accuracy: 76.281 -> 81.560Improved
increase backtracking from 2 to 4 tokens in Token Alignment (ablation)Token Alignment · English next-character predictionnext character accuracy: 76.281 -> 75.407Hurt
apply ByteSampler to convert the LM to character levelOLMo2-1B · English text (OLMo2 corpus)bits per character: 6.94 -> 0.87Improved
apply ByteSampler to convert the LM to character levelQwen3-1.7B-Base · Chinese text (MADLAD-400 zh)bits per character: 5.47 -> 3.43Improved
use ByteSampler instead of Token Alignment backtrackingQwen3-1.7B-Base · Chinese next-character predictionnext character accuracy: 49.6 -> 52.7Improved
use ByteSampler instead of prior exact method with prefix caching for solving PBPOLMo2 tokenizer · PBP inference costoverhead vs BPE (inference tokens): +2.09 -> +0.72Improved
ensemble Qwen3, OLMo2 and Llama3.2 at byte levelbyte-level ensemble · seven downstream tasksaccuracy: LAMBADA 0.727 -> 0.755; TriviaQA 0.535 -> 0.526Mixed
proxy-tune Llama3.1-8B with OLMo2-1B-Instruct expert and OLMo2-1B anti-expert at byte levelLlama3.1-8B · GSM8Kaccuracy: 55.3 -> 76.6Improved
proxy-tune Llama3.1-8B with OLMo2 expert pair at byte levelLlama3.1-8B · MMLUaccuracy: 27.8 -> 59.5Improved
proxy-tune Llama3.1-8B with OLMo2 expert pair at byte levelLlama3.1-8B · AlpacaEval 2LC winrate: 0.88 -> 33.5Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy