By configuring llama.cpp with n_predict=1 and n_probs=10 and disabling reasoning, any GGUF model can be prompted to output only a binary label (0 or 1) for tasks such as spam detection. The model then returns confidence percentages for each class, e.g., 1 = 94.9% and 0 = 5.08%. An example command shows how to launch the server with a specific model file.