Google shows how to turn DiffusionGemma into a scorer via vLLM templates

bodonoghue85 · x · 2026-09-30

Google's Gemma team shared a trick to turn DiffusionGemma into a Jev-like model: seed the canvas with a response template in vLLM, and you can extract confidence scores and probability distributions for yes/no, multiple-choice, and scored questions in a single denoising step — fast, efficient, and ready to scale for structured tasks.

Original post →

More from coding & agent

coding & agent channel →