Things I’m learning about machine learning, and what happens when I try them for myself.
September 2026 · 8 min read
I wanted to learn Hugging Face Spaces, so I built a pet breed classifier. I distilled a SigLIP 2 teacher into MobileNetV4 and tried to understand what the smaller model could learn from the teacher’s predictions.
Read article ↗
August 2026 · 16 min read
I wanted to understand what happens when a model chooses only some of its experts. I built small PyTorch examples, then looked at 271,104 routes from IBM Granite to see how the choices changed between prompt groups.
Read article ↗