Chinese-Jev: Bringing System One Model to Chinese-Language Tasks
Chinese-Jev is pitched as a faster Chinese-language decision model, not a chat model, with reported gains over closed-source Jev on Chinese tasks.
The paper introduces a unified pipeline that turns varied Chinese annotations into probability targets over candidate answers. The model uses a lightweight encoder-only backbone, then trains for decision scoring across general, medical, legal, and financial settings. Its authors report a 1.24% general-domain accuracy edge over Jev after pre-training, with a 20.3x speedup. Domain fine-tuning improves medicine accuracy by 4.0% over Jev and reaches 92% of Jev’s specialized-domain average accuracy at 17x speedup. HF Daily Papers' note
The paper introduces a unified pipeline that turns varied Chinese annotations into probability targets over candidate answers. The model uses a lightweight encoder-only backbone, then trains for decision scoring across general, medical, legal, and financial settings. Its authors report a 1.24% general-domain accuracy edge over Jev after pre-training, with a 20.3x speedup. Domain fine-tuning improves medicine accuracy by 4.0% over Jev and reaches 92% of Jev’s specialized-domain average accuracy at 17x speedup. HF Daily Papers' note
score 4