Data

Tools

Indexes

Applications

Experiments

Open Source Science

multiverse

https://github.com/infini-ai-lab/multiverse

Science Score: 54.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • ✓
    CITATION.cff file
    Found CITATION.cff file
  • ✓
    codemeta.json file
    Found codemeta.json file
  • ✓
    .zenodo.json file
    Found .zenodo.json file
  • ○
    DOI references
  • ✓
    Academic publication links
    Links to: arxiv.org
  • ○
    Academic email domains
  • ○
    Institutional organization owner
  • ○
    JOSS paper metadata
  • ○
    Scientific vocabulary similarity
    Low similarity (14.4%) to scientific vocabulary
Last synced: 6 months ago · JSON representation ·

Repository

Basic Info
  • Host: GitHub
  • Owner: Infini-AI-Lab
  • License: apache-2.0
  • Language: Python
  • Default Branch: main
  • Size: 25.4 MB
Statistics
  • Stars: 79
  • Watchers: 0
  • Forks: 9
  • Open Issues: 4
  • Releases: 0
Created 8 months ago · Last pushed 7 months ago
Metadata Files
Readme License Citation

README.md

Multiverse

[📄 Paper] | [🌐 Website] | [🤗 Huggingface] | [🐦 Twitter]


⚡ TL;DR

Multiverse is a generative modeling framework that natively supports parallel generation for efficient test-time scaling. We provide an end-to-end ecosystem for building and deploying Multiverse models in real-world applications.

🎬 Demo

We showcase a Multiverse model solving a math reasoning problem, demonstrating its parallel generation capabilities.

🏛️ Repository Structure

This repository provides a complete ecosystem for building and deploying Multiverse models. Our structure is organized as follows:

🗂️ data → 📈 train → 🚀 inference

Multiverse ├── data/ │ └── src | └── run | └── README.md │ ├── train │ └── README.md │ ├── inference/ │ └── engine | └── README.md │ └── README.md

  • data/: Contains the Multiverse Curator toolkit for dataset preparation. Use it to generate your own Multiverse-1K dataset for training.

  • training/: Implements the Multiverse Attention algorithm for the efficient training of Multiverse models. We also includes the code for AR baselines

  • inference/: Features the Multiverse Engine implementation, a high-performance inference server optimized for Multiverse models.

For detailed documentation and usage instructions, please refer to the README.md files in each directory.

📝 Todo List

  • [ ] Add evaluation code based on lighteval
  • [ ] Support context parallelism ## 📚 References

Thank you for your interest in Multiverse Engine! We hope this tool will be helpful for your research and development. If you find it useful, please consider citing our work. Happy coding! 🚀

bibtex @misc{yang2025multiverselanguagemodelssecretly, title={Multiverse: Your Language Models Secretly Decide How to Parallelize and Merge Generation}, author={Xinyu Yang and Yuwei An and Hongyi Liu and Tianqi Chen and Beidi Chen}, year={2025}, eprint={2506.09991}, archivePrefix={arXiv}, primaryClass={cs.LG}, url={https://arxiv.org/abs/2506.09991}, }

Owner

  • Name: Infini-AI-Lab
  • Login: Infini-AI-Lab
  • Kind: organization

Citation (CITATION.cff) 

cff-version: 1.2.0
message: "If you use this software, please cite it as below."
abstract: "Minimal recipe for test-time scaling and strong reasoning performance matching o1-preview with just 1,000 examples & budget forcing"
repository-code: "https://github.com/simplescaling/s1"
authors:
  - family-names: "Muennighoff"
    given-names: "Niklas"
  - family-names: "Yang"
    given-names: "Zitong"
  - family-names: "Shi"
    given-names: "Weijia"
  - family-names: "Li"
    given-names: "Xiang Lisa"
  - family-names: "Fei-Fei"
    given-names: "Li"
  - family-names: "Hajishirzi"
    given-names: "Hannaneh"
  - family-names: "Zettlemoyer"
    given-names: "Luke"
  - family-names: "Liang"
    given-names: "Percy"
  - family-names: "Candès"
    given-names: "Emmanuel"
  - family-names: "Hashimoto"
    given-names: "Tatsunori"
title: "s1: Simple test-time scaling"
type: software
keywords:
  - "test-time scaling"
  - "language models"
  - "reasoning"
  - "budget forcing"
references:
  - type: software
    title: "s1-32B"
    url: "https://hf.co/simplescaling/s1-32B"
  - type: dataset
    title: "s1K"
    url: "https://hf.co/datasets/simplescaling/s1K"
  - type: dataset
    title: "s1-prob"
    url: "https://hf.co/datasets/simplescaling/s1-prob"
  - type: dataset
    title: "s1-teasers"
    url: "https://hf.co/datasets/simplescaling/s1-teasers"
  - type: dataset
    title: "data_ablation_full59K"
    url: "https://hf.co/datasets/simplescaling/data_ablation_full59K"
date-released: "2025"
url: "https://arxiv.org/abs/2501.19393"
identifiers:
  - type: arxiv
    value: 2501.19393
    description: The ArXiv preprint
preferred-citation:
  type: misc
  authors:
    - family-names: "Muennighoff"
      given-names: "Niklas"
    - family-names: "Yang"
      given-names: "Zitong"
    - family-names: "Shi"
      given-names: "Weijia"
    - family-names: "Li"
      given-names: "Xiang Lisa"
    - family-names: "Fei-Fei"
      given-names: "Li"
    - family-names: "Hajishirzi"
      given-names: "Hannaneh"
    - family-names: "Zettlemoyer"
      given-names: "Luke"
    - family-names: "Liang"
      given-names: "Percy"
    - family-names: "Candès"
      given-names: "Emmanuel"
    - family-names: "Hashimoto"
      given-names: "Tatsunori"
  title: "s1: Simple test-time scaling"
  year: 2025
  url: "https://arxiv.org/abs/2501.19393"
  arxiv: "2501.19393"

GitHub Events

Total
  • Issues event: 6
  • Watch event: 52
  • Issue comment event: 6
  • Push event: 2
  • Fork event: 8
Last Year
  • Issues event: 6
  • Watch event: 52
  • Issue comment event: 6
  • Push event: 2
  • Fork event: 8

Dependencies

train/requirements.txt pypi
  • accelerate ==1.0.1
  • datasets ==3.1.0
  • gradio ==4.44.0
  • ipykernel ==6.28.0
  • ipython ==8.20.0
  • openai ==1.56.1
  • torch ==2.5.1
  • torchaudio ==2.5.1
  • torchvision ==0.20.1
  • transformers ==4.46.1
  • trl ==0.12.0
  • wandb ==0.17.3