What the compiler reports for the design on your machine: size and compute block by block, the memory a training step needs and whether it fits, where the step spends its time, and the same design at every precision. Every figure says where it comes from — exact, measured, estimated, or unknown with the reason — and what a run actually does is in Training.
The one workspace where something is running. Start a run on your own machine and watch it: loss, throughput, memory and GPU state as the process reports them, beside what NEURAX predicted — so you can see where the analysis held and where it did not.
The workspace for the person answerable rather than the person building. What the design commits, what could go wrong, and the footprint that has to be reported. It computes nothing new: every figure traces to the compiler, and where a figure does not exist it shows a gap rather than a zero.
Page 28 of 33. This guide is in English for now. It is the same guide that is built into the NEURAX desktop app (the ? button, or F1). Found a mistake? Tell us.
