NTLoss - a regression-like loss for LLMs

ntloss is a PyPI package of the "Number Token Loss" for language models. A regression-like loss that improves LLM performance on math tasks. Follows Regress, Don't Guess, ICML 2025

---

📖 Overview

This repo maintains the code for the ntloss PyPI package

• 🧑🏽‍💻 Paper source code: Regress, Don't Guess – ICML 2025
• 📄 Paper: Regress, Don't Guess – A Regression-like Loss on Number Tokens for Language Models
• 🌐 Project Page: Landing Page
• 🎮 Demo: HuggingFace Spaces Demo (Streamlit)
• 📖 Docs: Documentation for the PyPI package

🏃‍♂️ Quick Start

Simply install ntloss into your existing project

uv add ntloss
pip install ntloss # if you are oldschool

Use like this:

from ntloss import NTLoss
ntl_fn = NTLoss(tokenizer=tokenizer)
ntl = ntl_fn(logits, labels)

# We recommend
loss = cross_entropy(logits, labels) + 0.3 * ntl

NOTE: ntloss is currently in alpha phase and pre-release. Feedback & PRs are very welcome.

📝 Citation

If you use ntloss, please cite our paper:

@inproceedings{zausinger2025regress,
  title   = {Regress, Don't Guess – A Regression-like Loss on Number Tokens for Language Models},
  author  = {Jonas Zausinger and Lars Pennig and Anamarija Kozina and Sean Sdahl
             and Julian Sikora and Adrian Dendorfer and Timofey Kuznetsov
             and Mohamad Hagog and Nina Wiedemann and Kacper Chlodny
             and Vincent Limbach and Anna Ketteler and Thorben Prein
             and Vishwa Mohan Singh and Michael Danziger and Jannis Born},
  booktitle = {Proc. of the 42nd International Conference on Machine Learning (ICML)},
  year    = {2025},
  url     = {https://ibm.biz/ntl-main}
}

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.