TQLite Framework for Distilled MQM Translation Quality Evaluation

TQLite uses a multi-LRM jury to generate synthetic data for distilling high-level translation quality evaluation capabilities into small language models. The framework allows SLMs to approximate the MQM-based reasoning performance of much larger, more expensive models.