metadata
license: apache-2.0
base_model:
- deepseek-ai/DeepSeek-R1-Distill-Qwen-14B
Kaggle AI Mathematical Olympiad - Progress Prize 2 - 9th Place Solution (Fast-Math-R1-14B)
Team
- Hiroshi Yoshihara @ Aillis Inc., The Univ. of Tokyo
- Yuichi Inoue @ Sakana AI
- Taiki Yamaguchi @ Rist Inc.
Summary
By applying SFT and GRPO on difficult math problems, we enhanced the performance of DeepSeek-R1-Distill-Qwen-14B and developed Fast-Math-R1-14B,
which achieves up to 60% faster inference while maintaining accuracy.
Technical details can be found in Kaggle Discussion and Github.
| AIME 2024 | AIME 2025 | ||||
|---|---|---|---|---|---|
| Model | Token budget | Pass@1 (avg. 64) | Output tokens | Pass@1 (avg. 64) | Output tokens |
| DeepSeek-R1-Distill-Qwen-14B | 16384 | 63.3 | 9590 | 46.7 | 10602 |
| 12800 | 58 | 6444 | 41.9 | 6684 | |
| 8192 | 45.6 | 4920 | 30.6 | 4611 | |
| Light-R1-14B-DS | 16384 | 66.8 | 10146 | 51.3 | 11308 |
| 12800 | 59.2 | 6974 | 43.8 | 6869 | |
| 8192 | 42.4 | 5500 | 30.4 | 4908 | |
| Fast-Math-R1-14B | 16384 | 66 | 7932 | 49.2 | 9066 |
| 12800 | 63 | 5996 | 46.1 | 6127 | |
| 8192 | 51.4 | 4269 | 37.2 | 3905 |
Dataset
Inference
vLLM
from vllm import LLM, SamplingParams
vllm_engine = LLM(
model='RabotniKuma/Fast-Math-R1-14B',
max_model_len=8192,
gpu_memory_utilization=0.9,
trust_remote_code=True,
)
sampling_params = SamplingParams(
temperature=1.0,
top_p=0.90,
min_p=0.05,
max_tokens=8192,
stop='</think>', # Important: early stop at </think> to save output tokens
)
vllm_engine.generate('1+1=', sampling_params=sampling_params)