feat(A2-Q2b): lm_head reaches the device on NVFP4, and the seam stops handing out an address-keyed cache - #1415

Merged

localai-bot merged 25 commits into

mainfrom

Aug 22, 2026

feat(A2-Q2b): lm_head reaches the device on NVFP4, and the seam stops handing out an address-keyed cache
mudler
committed
1c62d99
test(A2-Q2b): the dgx:gpu0 gate runner, with the four void conditions checked before the expensive part
mudler
committed
806b263
docs(A2-Q2b): lm_head is a device arm now, and the host arms carry a measured per-token cost
mudler
committed
8fa900a
fix(A2-Q2b): the gate runner provisions its own CUDA toolkit, which is why the first submission VOIDed in 60 seconds
mudler
committed
29b1128
fix(A2-Q2b): the new test had never compiled, so the red it claimed could not have existed
mudler
committed
fedf78d
merge: origin/main into row/A2-Q2b-lmhead-nvfp4 for the review repair
mudler
committed
49564cd
fix(A2-Q2b): guard the lm_head fallback with the predicate, not with a process-wide counter
mudler
committed
bff2b7b

Navigation Menu