AI Paper Atlas

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

$p\textrm{-less}$ Sampling: A Robust Hyperparameter-Free Approach for LLM Decoding

Runyan Tan, Shuang Wu, Phillip Howard

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Parallel and early-exit decoding

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 p-less instead of top-pLlama2-7b · GSM8KAUC: 0.210 -> 0.267Improved
use p-less instead of min-pMistral-7b · GSM8KAUC: 0.523 -> 0.562Improved
use p-less instead of min-pLlama3-70b · CSQA, QASCAUC: CSQA 0.820 vs 0.819No clear effect
use p-less instead of min-pLlama2-7b · GPQAAUC: 0.243 vs 0.242No clear effect
use p-less instead of other truncation samplers at temperature 2.0Llama-2-7b · Writing Promptslength-controlled win rate: 48.94 (min-p) -> 65.64Improved
use p-less instead of top-p, epsilon, eta at temperature 1.5-2.0Llama-2-7b · Writing Promptslength-controlled win rate: top-p 0.00 -> 65.64 at temp 2.0Improved
use p-less at temperature 1.0 instead of epsilon-samplingLlama-2-7b · Writing Promptslength-controlled win rate: 62.18 -> 55.08Hurt
use p-less instead of min-p for sampling timeMistral-7b · GSM8K and GPQAsampling time per token: 0.02497 -> 0.01942 sImproved
use p-less instead of other samplers for generation lengthLlama2-7b · CSQA, QASC, GSM8Kgeneration lengthImproved
use p-less instead of top-p at temperature 2.0Mistral-7b · QASCdiversity: 1.00 -> 0.64Improved
use p-less_norm variant instead of p-less (ablation)Llama2-7b, Mistral-7b · math and reasoningAUC: 0.242 -> 0.248 (GPQA Llama2)No clear effect
p-less vs human annotators default sampling at temperature 2.0Llama2-7b · Writing Prompts (human eval)win rate: 58.8% majority voteImproved

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