Xingchen Lab has formally announced the release and open-sourcing of its intent understanding model, Xing4.0-29B-Intent. Accompanying this release is a comprehensive package that includes 2.75 million training corpora, model weights, and inference images. This innovative model leverages a unified, pattern-driven semantic parsing framework, which seamlessly integrates intent classification and parameter extraction capabilities. By doing so, it effectively transforms natural language instructions into a structured JSON format, enhancing the model's interpretability and usability.

According to rigorous evaluations, the Xing4.0-29B-Intent model has demonstrated exceptional performance, achieving a comprehensive score of 58. This score not only sets a new benchmark but also surpasses the previous highest score of 49 points, which was recorded among 18 mainstream general large models. This achievement underscores Xingchen Lab's commitment to advancing the field of intent understanding and natural language processing.