HuggingFace ml-intern: Open-Source ML Engineer Agent Automates Post-Training — Reads Papers, Trains Models, Ships Results

GitHub·high signalXBlueskyLinkedInCopy link

HuggingFace released ml-intern on April 21, an open-source AI agent built on smolagents that autonomously reads arXiv papers, discovers datasets, runs training jobs, debugs failures, and iterates. Early results show +22 points on GPQA and +60% on HealthBench in 10 hours of autonomous operation. Gained 530 stars today, reaching 2,058 total — the fastest-rising new repo on GitHub trending.