China Telecom releases Xing4.0-29B-Intent for intent recognition and parameter extraction

What happened

China Telecom’s Xingchen General AI Lab released and open-sourced Xing4.0-29B-Intent, which combines intent recognition and parameter extraction in schema-constrained JSON output. The release includes 2.753 million supervised fine-tuning examples, model weights, and an inference image. The team reports higher scores than the compared models in its extraction and category-selection evaluations, while limiting those findings to the tested datasets and settings.

Read next

• More changes in today’s briefingMicrosoft Research and Collaborator Resolve the KLS Conjecture With Human-Directed AI

The story so far

China Telecom’s Xingchen General AI Lab released and open-sourced Xing4.0-29B-Intent, a model that combines intent recognition and parameter extraction to produce schema-constrained JSON output. The release includes model weights, an inference image, and 2.753 million supervised fine-tuning examples. The team reported that the model scored higher than the models it compared against in extraction and category-selection evaluations. Those results apply to the tested datasets and settings.