AI Reports · Amazon · 2026-10-20 · Blog / commentary
Fine-tuning GPT-OSS-20B for multilingual reasoning on Amazon SageMaker HyperPod
GPT-OSS-20B is an open-weight reasoning model that thinks in English; on a 50-example multilingual reasoning test, the base model reasoned in the correct target language only 22% of the time. Fine-tuning on Amazon SageMaker HyperPod raised reasoning-language accuracy from 22% to 100% with supervised fine-tuning (SFT), while reinforcement learning with GRPO added measurable gains where headroom existed. The key is matching the technique to the capability gap: SFT does most of the work, and GRPO refines when there is headroom.