Tag
This paper shows that the repetition penalty decoding parameter enables inference-time control of the precision–recall trade-off in LLM-based grammatical error correction, without any changes to model architecture, training, or prompts.
FORK提出了一种联合优化召回率与精确率的奖励函数，通过rubric驱动的必要性信用和事实正确性评分器来缓解基于rubric的强化学习中的奖励破解问题，在多个基准上显著提升胜率并减少冗余输出。
A practical guide on using confidence scores in document extraction to balance automation and accuracy by setting thresholds for precision and recall.