llama.cpp adds support for Nemotron 3.6 Academia looped hybrid model

maternion · reddit · 2026-11-22

A Reddit post highlights two llama.cpp PRs (#32766 and #32765): the first adds support for Nemotron 3.6 Academia, a looped hybrid model combining Gated Delta Product (GDP), dense attention, and MLP layers, with shared recurrent weights, loop conditioning, and both perloop and lastloop attention KV caches; the second enables rectangular GDN states.

Original post →

More from coding & agent

coding & agent channel →