Skill issue: stop deploying vision language models, use them with Skills — Merve Noyan, Hugging Face
The author argues that developers should stop deploying heavy vision language models directly and instead use smaller vision models combined with specialized toolkits and training pipelines. By leveraging vision language models as labelers and judges, developers can automate dataset annotation and train smaller, highly robust models like RFDETR at minimal cost. This approach addresses limitations in real-time performance, robustness, and licensing associated with direct vision language model deployments. Open article →