# Google Tunix Hack - Train a model to show its work — Data

来源：https://www.kaggle.com/competitions/google-tunix-hackathon/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

Welcome to the Google Tunix Hackathon! Here are some resources you can use to help get you started:

- Tunix repository on GitHub: https://github.com/google/tunix/ 
- Tunix documentation website: https://tunix.readthedocs.io/en/latest/index.html 
- GRPO Starter notebook on Kaggle (Gemma2 2B): https://www.kaggle.com/code/windmaple/grpo-demo-gemma2-2b
- GRPO Starter notebook on Kaggle (Gemma3 1B): https://www.kaggle.com/code/windmaple/grpo-demo-gemma3-1b
- JAX: https://jax.dev
- Flax documentation website: https://flax.readthedocs.io/

Paper references:
- [DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning](https://arxiv.org/abs/2501.12948)
- [Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains](https://arxiv.org/abs/2507.17746)
- [Extrapolating RLVR to General Domains without Verifiers](https://arxiv.org/pdf/2506.18254)
- [Bridging Offline and Online Reinforcement Learning for LLMs](https://arxiv.org/pdf/2506.21495)