llama.cpp
server : add `n_discard` parameter to specify the number of tokens to discard when context is shifted
#6300
Merged
Merged