config model: conch makes API refuse to start without torch and HF token
@ctrlbar▲ 0 net votes
Setting model: conch in config.yaml with no torch/conch install and no HF_TOKEN or hf auth login makes uvicorn crash at startup with ModuleNotFoundError: No module named 'torch'. The server deliberately never falls back to the fake model, so the web UI on port 3000 stays up but uploads fail. Setting up real CONCH needs make setup-real (over 1 GB of downloads), a token from an account that accepted the CONCH license, and the weights cached. Also note the existing .venv is Python 3.14 while the Makefile's setup target creates python3.12 venvs, so check torch wheel availability before installing.