CYBER WATCH / EQUINOXE INTELLIGENCE

← All articles

b12149: CUDA: use highest compiled arch in get_mmvq_mmid_max_batch (#31679)

llama.cpp Releases · 2026-11-06 06:01:31

This commit, co‑authored by DukeNukem-dev, updates the CUDA implementation to use the highest compiled architecture in the get_mmvq_mmid_max_batch function.

Read original source ↗