llama.cpp

server : add `n_discard` parameter to specify the number of tokens to discard when context is shifted

#6300

Merged