Benchmark performance
In the standardized SimpleQA evaluation system, Jan-nano obtains an accuracy performance of 87.3% in the Chinese domain, outperforming models of the same parameter class by 15 percentage points. Its specially strengthened entity recognition capability achieves an F1 value of 91.2% when dealing with complex questions in specialized fields such as medicine, law, etc. The non-thinking chat template adopted by the model effectively reduces the redundant output of 17%, making the answers more direct and effective.
Practical application tests show that when faced with professional questions such as "Explaining the relationship between the application of Article 52 of the Contract Law and Article 153 of the Civil Code", Jan-nano can automatically refer to the database of legal texts and generate detailed analyses that include judicial interpretations and typical cases, with a level of accuracy that reaches the level of practicing attorneys, which is 821 TP3T. This performance makes Jan-nano a powerful assistant in specialized fields such as legal technology and medical consulting.
This answer comes from the articleJan-nano: a lightweight and efficient model for text generationThe
