feat(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): give GLM-5.3-Flash's k-pool indexer a device it can run on - #2432

Open

localai-org-maint-bot wants to merge 7 commits into

mainfrom

row/MODEL-MM-GLM53-FLASH-KPOOL-CUDA

Commits on Aug 31, 2026

spec(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): take #2415's decision and write the kernel down before writing it
mudler
committed
795cb7c
feat(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): give the k-pool indexer a device it can run on
mudler
committed
71e213e
test(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): bound the score comparison by the margin the top-k decides on, not by a number
mudler
committed
83fa7c8
merge: origin/main into the k-pool device-op branch, keeping both tests/CMakeLists.txt additions
mudler
committed
b6df9a7
chore(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): allowlist the k-pool probe's two kCUDA lookups, with the reason beside them
mudler
committed
4034c36
record(MODEL-MM-GLM53-FLASH-KPOOL-CUDA): the k-pool device gate is taken on GB10, and the two mutations that are not assertion kills say so
mudler
committed
fe2f1db
record(MODEL-MM-GLM53-FLASH): --device cuda reaches the KV binding, not the refusal this row keeps citing
mudler
committed
b20267f