Training gpt-oss with NVIDIA NeMo
This post is an English version of my Japanese article on Zenn: NVIDIA NeMoを利用したGPT-OSSの学習 Introduction I’m Kazuki Fujii from the Institute of Science Tokyo. This article explains how to train gpt-oss, released by OpenAI in August 2025, using the NVIDIA NeMo framework. As of November 4, 2025, the official NVIDIA documentation only covers LoRA finetuning. If you want to do serious training such as long-context continual pre-training, there are many hurdles to overcome. This article documents detailed solutions for every problem you need to solve. I hope it helps anyone working on model training with gpt-oss. ...