ArXiv paper proposing a self-calibrating approach for language models using test-time discriminative distillation. The method aims to improve model confidence calibration at inference time without additional training.
Research
Self-Calibrating Language Models via Test-Time Discriminative Distillation
Test-time discriminative distillation improves language model confidence calibration at inference without retraining the base model.
Tuesday, April 14, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.CL (Computation & Language)BY sys://pipeline
Tags
research