Evidence before a claim.
An activation plot isn’t an explanation by itself. A GPU chart isn’t proof of distributed inference. The measurements and their limits should stay visible.
A LAB BUILT FROM CURIOSITY
Better questions.
Tools to investigate them.
I’m Danilo Canivel, creator of Dyno Lab. I’m building an open workbench that brings model serving, interpretability experiments and hardware visibility into one place.
Why I’m building Dyno
Running a model is a starting point. The interesting questions come next: what changed its answer, what can we measure inside it, and can someone else reproduce the result?
Dyno brings those questions closer to the everyday experience of using local models. The aim is to make experimentation easier to start, easier to inspect and easier to share.
Use the hardware you have.
Start on your Mac. Explore a supported local GPU pool when a model needs more room. Keep the infrastructure understandable.
Build in the open.
Clear docs, reproducible examples and focused contributions make the lab more useful. Bring a question, a bug report or an experiment.