llama.cpp

server: stop generation at `n_ctx_train` if `n_predict` is not set

#6638

Merged