llama.cpp
CUDA: Improve performance via less synchronizations between token
#17795
Merged
Merged