A Practical Two-Stage Recipe for Mathematical LLMs: Maximizing Accuracy with SFT and Efficiency with Reinforcement Learning
Paper
•
2507.08267
•
Published
•
8
Fast-Math
Fast-Math is a model series designed to significantly improve inference efficiency while preserving accuracy on math reasoning tasks.
-
-
RabotniKuma/Fast-Math-R1-14B
Text Generation • 15B • Updated • 16 • 3 -
RabotniKuma/Fast-OpenMath-Nemotron-14B
Text Generation • 15B • Updated • 19 • 3 -
RabotniKuma/Fast-Math-Qwen3-14B
Text Generation • 15B • Updated • 12 -
RabotniKuma/Fast-Math-R1-SFT
Viewer • Updated • 7.9k • 42 • 2 -
RabotniKuma/Fast-Math-R1-GRPO
Viewer • Updated • 3.26k • 20 • 1 -
RabotniKuma/Fast-Math-R1-Token-Scheduler
Viewer • Updated • 93.7k • 16