Inkling-2-Small for Single DGX-Spark

Inkling 2 Small (594B total, 17B active, 384 experts × top-8, text + image in) with its 587B routed-expert weights vector-quantized to 12 bits per 8 weights by hVQ: 110.1 GiB of parts, sized to serve on one 128 GB DGX Spark (GB10) through ds4-dfm-rs with the vision tower and the MTP drafter.