llama.cpp

Avoid calls to tokenizer.added_tokens_decoder

#12473

Merged