Hi all, I’m currently working on fine-tuning transformer models to process and predict patterns from IoT sensor data, specifically multivariate time-series streams. The IoT framework I’m using follows a typical setup. While researching general IoT architecture, I came across this helpful article https://www.theengineeringprojects.com/2023/06/top-iot-starter-kits-for-the-beginners-to-learn-programming.html which outlines the system-level design I’m implementing. Still, I need more guidance on applying transformer models for this specific use case.
Has anyone here worked with fine-tuning Hugging Face models on IoT sensor data before? Any tips for optimizing training with limited hardware would be appreciated.
1 Like
I’m not very familiar with it, but I think this is the class that comes closest to what you’re looking for in Transoformers…
When deploying to IoT, I think most people convert to ONNX format and then to TensorRT format if necessary.
Hi, thanks for the article. I read it, and in its way it helped me. At first, it seemed remote monitoring was only for large enterprises, but in Germany I connected sensors to an old packaging line and realized I could actually track energy and equipment status without constant supervision. Through Smartmakers a signal came in that a motor was consuming 18% more energy than usual, and we managed to shut it down before a failure. Have you ever felt the thrill of avoiding major costs? In a quarter, this saved $12,300 and reduced downtime by 27%, and now we use this system across all lines.
1 Like