"""Standalone inference for the decision checkpoints exported by scripts/export_release.py (pplx-decider's native decision-checkpoint layout, format_version 1). model = DecisionModel.from_pretrained("", device="cuda") model.decide("My integration keeps failing.", {"team": {"type": "choice", "instructions": "Which team should handle this?", "criteria": {"billing": "Charges and refunds", "support": "Technical errors"}}}) -> {"model": ..., "answers": {"team": {"type": "choice", "choice": "support", "probabilities": {...}}}} One decision per row in the model's chat template, as in training: system: Classify the supplied state using the question and option descriptions. ... user: [images] State:\\n\\n\\nQuestion:\\n\\n\\nOptions:\\nA: