AMD BC-250

Vision projector placement; read the whole first page; capped refresh

Shorter How it works; don't retry refused header reads

One entry per model; explain the score

Keep the whole model in VRAM; laptop GPUs

Memory estimates: embeddings per quant, SWA slots, vision projector; plainer descriptions

Older GPUs and mixed cards

Lowest quant to consider

Local Model Explorer v1 (private)