{ "type": "quantile_regression_forest", "task": "llm_output_length_upper_bound_prediction", "training_trace": "lmsys-chat", "models": ["llama-3.1-8b", "qwen2.5-7b"], "quantile": 0.95, "framework": "scikit-learn", "serialization": "joblib/pickle", "jitserve_version": "nsdi26", "notes": "QRF predictor and vectorizer must be loaded together" }