Fine-tuning GPT-OSS-20B for multilingual reasoning on Amazon SageMaker HyperPod

Researchers fine-tuned the open-weight GPT-OSS-20B model on Amazon SageMaker HyperPod to improve its multilingual reasoning, enabling it to reason and answer in French, German, Spanish, and Italian rather than defaulting to English. They compared supervised fine-tuning (SFT) and reinforcement learning with Group Relative Policy Optimization (GRPO), finding SFT does most of the work while GRPO refines performance when there is room for improvement.