llama.cpp

Try to fix Baichuan2 models by using vocab size in config.json

#3299

Merged