[Core] Optimizing cross-attention `QKVParallelLinear` computation#12325

Open

NickLucche wants to merge 9 commits intovllm-project:mainfrom NickLucche:encdec-separate-crossattn

+125-50

Commits on Feb 19, 2025

first draft
NickLucche
committed
cleanup
NickLucche
committed
submodules in dict to avoid param registration
NickLucche
committed
mllama test
NickLucche
committed
format
NickLucche
committed
clean up comments
NickLucche
committed
fix distributed use

NickLucche
authored and
NickLucche
committed
format
NickLucche
committed
address review
NickLucche
committed