Skip to content
OpenTrain AIFor AI Companies

Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach

Published Feb 1, 2025
arXiv PDF
Researcher verdict
Starting point
Use as implementation starting point
Benchmark evidence
Thin evidence
Verify before relying
Time to first repro
A few hours
Fast first run
Risk flags
1
Review before use

Results and benchmarks

Freshness tier: hot
Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach is the primary contribution described in this paper.
Task Dataset Metric Value Source
Recurrent Depth Approach GSM8K Accuracy 20.00 paper-derived
Recurrent Depth Approach MBPP pass@1 43.80 paper-derived

Audit each benchmark finding before selecting an implementation path. Evidence refs map to the disclosure below.

Implementation

Best maintained implementation now

Recommended
Confidence: High
Reproducibility: Moderate
seal-rg/recurrent-pretraining

Pretraining and inference code for a large-scale depth-recurrent language model

961 stars · 87 forks · Last push Dec 29, 2025 · Apache-2.0 license

  • License
  • CI
  • Dependencies
  • Docker

Official implementation from Papers with Code · Repository link is mentioned in the paper metadata · Matched via arXiv identifier search

Why this implementation
Confidence: high

seal-rg/recurrent-pretraining is the strongest maintained implementation based on ranking signals. License is declared (Apache-2.0). Dependency/environment manifests are present.

Open seal-rg/recurrent-pretraining
Reproduction risks
  • No CI workflows detected
  • Selected seal-rg/recurrent-pretraining as the strongest maintained implementation for new work.
  • Includes dependency/environment manifest signals.
  • Repository activity is within the last 24 months.

Compare implementation paths

Compare maintenance quality, reproducibility coverage, and evidence confidence before choosing a reproduction baseline.

seal-rg/recurrent-pretraining
best maintained
Maintenance
Stale risk
Confidence
High
Reproducibility
Moderate
Stars
961
Last push
Dec 29, 2025 (261d)

Official implementation from Papers with Code · Repository link is mentioned in the paper metadata

  • No CI pipeline detected
  • No tagged releases
  • No Docker setup
gair-nlp/prox
alternative
Maintenance
Stale
Confidence
Low
Reproducibility
Moderate
Stars
273
Last push
Jul 8, 2025 (435d)

Community adoption signal (273 stars)

  • No push in 12+ months
  • No CI pipeline detected
  • No tagged releases
diamond2nv/hfpapers-crawler
alternative
Maintenance
Active
Confidence
Low
Reproducibility
Moderate
Stars
10
Last push
Sep 11, 2026 (5d)

Matched via arXiv identifier search

  • No CI pipeline detected
  • No Docker setup
  • Low confidence match

Reproduction readiness

Time to first repro: hours
Last checked: Sep 16, 2026

Setup required

Dependencies pinned, manual setup needed

  • seal-rg/recurrent-pretraining has pyproject.toml but requires manual environment setup.
  • Last push was 261 days ago, so expect possible dependency version conflicts.
  • No Dockerfile, so you will set up the environment manually.
  • No CI pipeline, so test coverage is unknown.
Open seal-rg/recurrent-pretraining

Quick start

git clone https://github.com/seal-rg/recurrent-pretraining.git
pip install -e .

Repositories and ecosystem

No additional verified repositories beyond the primary recommendation.

Hugging Face artifacts

No trustworthy direct or curated related Hugging Face artifacts were found yet. Use targeted searches to quickly locate candidate models, datasets, and demos.

Search models Search datasets Search spaces

Tip: start with models, then check datasets and spaces if you need evaluation data or demos.

Research context

Tasks

Recurrent Depth Approach

Methods

None detected

Domains

None detected

Evaluation and human feedback data

Open this paper in HFEPX to review benchmark signals, evaluation modes, and human-feedback protocol context.

Open in HFEPX
Explore similar papers

Jump to Paper2Code search queries derived from this paper's research context.

Recurrent Depth Approach

Data includes links from Papers with Code ( CC-BY-SA-4.0 ).