An SLM trained on $8 ESP32-S3

A researcher successfully trained a transformer model from scratch on an $8 ESP32-S3 microcontroller, demonstrating that model training—not just inference—is possible on budget hardware without external GPUs or datacenters. Unlike existing edge AI projects that run pre-trained models, this "Qapla'" project enables on-device learning for scenarios where training data only becomes available after deployment, such as a sensor learning the unique vibration patterns of a specific machine in the field.