Discursive Tech tools

AI, software, and the hardware you actually touch

Local Model Fit Sheet
Enter the cards you own once, pick a model, and see which quantisations fit and how much context each one leaves. The cache arithmetic is computed from each model's published architecture, so it accounts for grouped-query attention and sliding windows instead of assuming every model is shaped the same. → Work out which quantisation of a model fits the VRAM you actually have and what context length it leaves you, against the machines you own, before downloading forty gigabytes to find out.
AI Rig Builder
Drag graphics cards onto a real motherboard and a processor into its socket, and see which slots the coolers cover, which links drop to x8 or to the processor's PCIe generation, and what goes on a riser. The power budget follows every part to its supply and the wall circuit, and the models that fit the cards' memory, or run with layers in system memory, are listed from their published configs. → Lay out a local-model machine on its real motherboard before buying it, with the processor, memory, each card's cooler, shared PCIe lanes, power supplies and wall circuit checked as the parts are dragged into place, and see which models its memory runs.