Yuichi Inoue, Hiroshi Yoshihara, Taiki Yamaguchi
We lifted 6 functions out of this paper's own repositories and ran 0 of them in a sandbox. "Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results.
| Repository | Role | Ran |
|---|---|---|
| analokmaus/kaggle-aimo2-fast-math-r1 | canonical | 0 of 6 |
| Function | Status | Where it lives |
|---|---|---|
| create_prompt | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/experiments/train_fast_nemotron_14b.py pointer only (licence: NONE) · get_code("522a7f189e5ea3b4") |
| create_prompt | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/experiments/train_first_stage.py pointer only (licence: NONE) · get_code("e14b5e417be024bf") |
| exact_div | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/finetune/openr1_mod/faster_grpo_trainer.py pointer only (licence: NONE) · get_code("aed7a3866e065bb8") |
| extract_boxed_text | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/finetune/grpo_rewards.py pointer only (licence: NONE) · get_code("d1630e3f92e89d5d") |
| extract_contents | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/finetune/grpo_rewards.py pointer only (licence: NONE) · get_code("2f5695b9a0fb82de") |
| preprocess_dataset | Not yet run | analokmaus/kaggle-aimo2-fast-math-r1/experiments/train_token_scheduler.py pointer only (licence: NONE) · get_code("37c8141d07e7590d") |
Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.
Enhancing the mathematical reasoning of Large Language Models (LLMs) is a pivotal challenge in advancing AI capabilities. While Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) are the dominant training paradigms, a systematic methodology for combining them to maximize both accuracy and efficiency remains largely unexplored. This paper introduces a practical and effective training recipe that strategically integrates extended SFT with RL from online inference (GRPO). We posit that these methods play complementary, not competing, roles: a prolonged SFT phase first pushes the model's accuracy to its limits, after which a GRPO phase dramatically improves token efficiency while preserving this peak performance. Our experiments reveal that extending SFT for as many as 10 epochs is crucial for performance breakthroughs, and that the primary role of GRPO in this framework is to optimize solution length. The efficacy of our recipe is rigorously validated through top-tier performance on challenging benchmarks, including a high rank among over 2, 200 teams in the strictly leak-free AI Mathematical Olympiad (AIMO). This work provides the community with a battle-tested blueprint for developing state-of-the-art mathematical reasoners that are both exceptionally accurate and practically efficient. To ensure full reproducibility and empower future research, we will opensource our entire framework, including all code, model checkpoints, and training configurations at https://github.com/analokmaus/kaggle-aimo2fast-math-r1.
The same record, over MCP at https://syntology.ai/mcp:
get_harvested_code_for_paper("2507.08267")
get_code_for_paper("2507.08267")
have("2507.08267")
Connect an agent — have() is free.