Tag
Jev, a structured text classification model, demonstrated the ability to process 724 live ads from 37 brands in 40 seconds for just $0.09 in tokens, achieving 216ms median latency per record through parallel processing and integration with Gemini.
The article announces the launch of together/Tev1-4B-experimental, a Jev-like classification model fine-tuned from Qwen3.5 4B, and provides a guide on training your own version for approximately $17.
The dedicated evaluation model JEV demonstrated high efficiency and low-cost potential in processing interview transcripts, emphasizing the advantages of using large language models as logical judgment layers rather than content generators.
This paper presents a cross-lingual analysis of entrainment behavior in Mandarin-English, Hindi-English, and Spanish-English code-switched dialogues, finding that lexical entrainment generalizes but acoustic-prosodic and style entrainment vary. It also evaluates how classification models capture these behaviors, showing they prioritize different features than humans.