A Google hackathon on Kaggle is now open for entries showing how to use Tunix, Google's JAX-native library for LLM post-training, to train a model to show its work by laying out a reasoning trace before landing on an answer.

Kaggle is a platform that hosts data science competitions. Users can compete to build the best models for a given problem, share code and data, and access a web-based environment for data exploration and model building. Kaggle has more than 4 million members from 194 countries, is the world’s largest data science competition community, and Kaggle has been part of Google since 2017.

Tunix is an open-source, JAX-native library built specifically for LLM post-training. Tunix provides a comprehensive and developer-friendly toolkit for aligning models at scale. It is built for performance on TPUs, and includes algorithms for Supervised Fine-Tuning (SFT), preference tuning, knowledge distillation, and advanced Reinforcement Learning methods like PPO, GRPO and GSPO in a single, unified library. It was released by Google at the end of September 2025.

The Kaggle competition requires entrants to use Tunix to train an open-source or open-weight language model to show its work in terms of its reasoning in arriving at the final answer created by the language model.

The competition description says that reasoning traces help make language models more capable, trustworthy, transparent, and valuable.

Entrants start with Google's open-weight Gemma model (Gemma2 2B or Gemma3 1B), and fine-tune it with Tunix on TPU, then teach it how to reason through complex questions. The entries should create a model that not only gets the right answer, but also explains how it got there.

The solution should include a working training pipeline using Tunix and Gemma. Entries opened on November 11, and the final submission deadline is January 12, 2026. Judging will take place from January 13-23, 2026. The competition rules say that given limited compute on Kaggle TPUs (9 hours per session, 20 hours per week), a maximum output token of less that 1K is fine.

The first prize winner will receive $30,000, with prizes of decreasing values down to a 6th Place prize of $10,000.

Entries are open now.

More Information

Google Tunix Hack On Kaggle

Tunix On GitHub

Related Articles

Kaggle Contest To Detect Chatbot Essays

Kaggle Enveloped By Google Cloud

To be informed about new articles on I Programmer, sign up for our weekly newsletter, subscribe to the RSS feed and follow us on Facebook or Linkedin.

Comments

Make a Comment or View Existing Comments Using Disqus

or email your comment to: comments@i-programmer.info