# only to run serve/merge_lora.py and its tests (the demo and launcher do not need them) torch safetensors