llama.cpp

llama : make tensor-split regex patterns static to avoid per-token recompilation

#24710

Merged