arXiv 2512.14856

T5Gemma 2: Seeing, Reading, and Understanding Longer

By Biao Zhang, Paul Suganthan, et al.

Published 2025-12-16

Wiki summary

Explore the paper's summary, context, and related research on Papiers.

We introduce T5Gemma 2, the next generation of the T5Gemma family of lightweight open encoder-decoder models, featuring strong multilingual, multimodal and long-context capabilities. T5Gemma 2 follows the adaptation recipe (via UL2) in T5Gemma -- adapting a pretrained decoder-only model into an encoder-decoder model, and extends it from text-only regime to multimodal based on the Gemma 3 models. We further propose t…

View the original paper on arXiv