AI Paper Atlas

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

PonderLM: Pretraining Language Models to Ponder in Continuous Space

Boyi Zeng, Shixiang Song, Siyuan Huang, Yixuan Wang, He Li, Ziwei He, Xinbing Wang, Zhiyu li, Zhouhan Lin

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Faithfulness of LLM explanations

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
add pondering (3 steps) during pretrainingPythia-2.8B · 9 downstream tasks 5-shotaverage accuracy: 57.6 -> 61.5Improved
add pondering (3 steps) during pretrainingPythia-1B · 9 downstream tasks 0-shotaverage accuracy: 50.6 -> 55.0Improved
add pondering to GPT-2 and LLaMA pretrainingGPT-2 and LLaMA 405M-1.4B · Pile subsetpretraining lossImproved
replace pondering embedding with last hidden state (Coconut-style) (ablation)PonderPythia-70M · Pile validationperplexity: 14.16 -> 15.64Hurt
replace pondering with looped transformer (3 loops) (ablation)PonderPythia-70M · Pile validationperplexity: 14.16 -> 15.33Hurt
replace pondering with pause tokens (3 pauses) (ablation)PonderPythia-70M · Pile validationperplexity: 14.16 -> 16.53Hurt
add linear projector on last hidden state (ablation)Last Hidden State baseline · Pile validationperplexity: 15.64 -> 15.64No clear effect
increase pondering steps from 3 to 6 (ablation)PonderPythia-70M · Pile validationperplexity: 14.16 -> 13.56Improved
increase top-K from 10 to 100 (ablation)PonderPythia-70M · Pile validationperplexity: 15.18 -> 14.21Improved
use full vocabulary instead of top-100 tokens (ablation)PonderPythia-70M · Pile validationperplexity: 14.21 -> 14.20No clear effect
train with randomized pondering steps in [1,10] (ablation)PonderPythia-70M · Pile validationlossImproved
continually pretrain Pythia-1B with pondering instead of vanillaPythia-1B · 9 downstream tasks 5-shotaverage accuracy: 50.7 -> 51.5Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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