PatMT2, introduced in a new arXiv preprint, is a two-stage LoRA pipeline that adapts the Hy-MT2-7B translation model for patent translation. The system covers four languages: Chinese, English, Japanese, and Korean.

The abstract describes the approach as a two-stage pipeline, but does not provide details on the stages or the training data. It also mentions evaluation on six translation tasks, though no results are included in the abstract.

The paper is available on arXiv with identifier 2610.16466. Readers interested in the full methodology and findings will need to consult the complete manuscript.