{ "temperature_per_prim": { "choice": 1.1387886347566916, "noul": 1.3939382634489992, "score": 1.5422108254079405 }, "thresholds": { "0.01": { "confidence": 0.9546241492914076 }, "0.05": { "confidence": 0.7629656203142733 } }, "fitted_on": "calib FIT split (70% by twin group)", "certified_on": "calib CERT split (30%)", "evaluated_on": "test (frozen)", "eval_dtype": "float32", "validity": "Fitted on the unquantised model; not validated for this FP8 checkpoint, whose predictions differ. Refit on your own calibration data before relying on the thresholds." }