How to Automate AI Model Documentation with the NVIDIA MCG Toolkit | NVIDIA Technical Blog

This post introduces the NVIDIA Model Card Generator (MCG) toolkit, a containerized pipeline that automates the generation of AI model documentation in Model Card++ format. It details the toolkit's three-stage architecture (Ingestion, Extraction, Rendering) which utilizes NVIDIA Inference Microservices (NIM) and large language models (GPT-OSS-120B, Nemotron embeddings/reranking) to extract information from source code, documentation, and config files. The post highlights the toolkit's flexibility through configurable models, templates, and field-level guides, and its deployment options (on-premises, own cloud, Kubernetes). Performance results are presented, showing generation times, completion rates, and accuracy across various models. The toolkit's behavior with sparse documentation (flagging gaps rather than guessing) is also discussed. Oracle is mentioned as an early adopter.