Developer distills Jev model into tiny TF-IDF and Qwen students that match the teacher

A developer distilled the open Open-Jev 27B model into a 3GB Qwen student and a 70MB TF-IDF model using 1000 labeled examples, and both statistically matched the teacher's accuracy, dramatically cutting costs. The official hosted Jev scored 79.6% on the same test set, and the ZenML-based pipeline is open-sourced.

2026-09-25 ~ 2026-09-25 · 2 related posts