OpenTrain AI
Maintained implementation availablepytorchPretrained Models Available

PonderLM: Pretraining Language Models to Ponder in Continuous Space

Boyi Zeng, Shixiang Song, Siyuan Huang, Yixuan Wang, He Li +4 more

May 27, 2025arXiv: 2505.20674
1 repo26 stars~a few hours to reproduce
arXiv PDF

Abstract

Humans ponder before articulating complex sentence elements, enabling deeper cognitive processing through focused effort. In this work, we introduce this pondering process into language models by repeatedly invoking the forward process within a single token generation step. During pondering, instead of generating an actual token sampled from the prediction distribution, the model ponders by yielding a weighted sum of...

Best Implementation

Official implementation of the paper "Pretraining Language Models to Ponder in Continuous Space"

26 0 Jul 2025 Apache-2.0
License
CI
Deps
Docker
  • Selected lumia-group/ponderinglm as the strongest maintained implementation for new work.
  • Includes CI workflow signals.
  • Includes dependency/environment manifest signals.
  • Repository activity is within the last 24 months.

Reproduction Path

  1. 1

    Start with lumia-group/ponderinglm and validate setup instructions in README.

  2. 2

    Reproduce the baseline result with the provided defaults before modifying hyperparameters.

  3. 3

    Log exact dependency versions and runtime environment for reproducibility.

Time to first repro: a few hoursNo repository-level red flags were detected, but paper-specific preprocessing and hyperparameter details may still be under-specified.

Additional Implementations

No additional verified repositories beyond the primary recommendation.

Hugging Face Artifacts

No direct paper-linked artifacts were found. Showing strongest curated related artifacts.

Research Context