llama.cpp
llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4
#25370
Merged
Merged
llama : make all KQ masks f16 if FA is used, remove zero attention bias, remove raw_k repeats in DeepSeek V4 #25370
fairydreaming
merged
3 commits into
ggml-org:master
from
fairydreaming:dsv4-remove-kq-bias
llama : make all KQ masks (except the lightning indexer one) f16 if F…
17252233
am17an
approved these changes
on 2026-07-07
llama : remove dead code that repeats unified raw_k cache for each st…
a0e9766f
fairydreaming
changed the title
Merge remote-tracking branch 'upstream/master' into dsv4-remove-kq-bias
2b0d9804
am17an
approved these changes
on 2026-07-08
ggerganov
approved these changes
on 2026-07-10
Login to write a write a comment.
Login via GitHub