Researchers propose UltraX, a function-calling refinement framework designed to address the limitations of existing large-scale data processing methods by combining insertion with deletion and modification for fine-grained instance-level editing.
- The framework utilizes dataset-adaptive prompt optimization to guide an expert LLM in generating high-quality end-to-end refined texts.
- Line Alignment Mapping and Dynamic Context Replacement convert original-refined text pairs into structured program supervision.
- Low-confidence example filtering and ratio-controlled sampling stabilize the training distribution and improve supervision quality.
- Sliding-window prediction, global operation aggregation, and systematic post-processing ensure stability during inference and execution.
Experiments demonstrate that UltraX achieves the highest average performance across all corpora while matching or surpassing baselines with fewer training tokens, indicating stronger data efficiency and refinement reliability.