Google DeepMind(@GoogleDeepMind)

This progress allow us to rethink global compute: 🔘 We successfully trained a 12B @GoogleGemma mode...

5.0内容质量
This progress allow us to rethink global compute: 🔘 We successfully trained a 12B @GoogleGemma mode...

TL;DR · AI 摘要

核心要点

打开原文

🔘 We successfully trained a 12B @GoogleGemma model across four US regions using low-bandwidth networks 🔘 We showed we can mix different hardware generations, such as TPU6e and TPUv5p, without slowing down performance during https://t.co/g6dwehJguM" / X

Image 1: Square profile picture
Image 1: Square profile picture

Google DeepMind

@GoogleDeepMind

This progress allow us to rethink global compute: Image 2: 🔘 We successfully trained a 12B

@GoogleGemma

model across four US regions using low-bandwidth networks Image 3: 🔘 We showed we can mix different hardware generations, such as TPU6e and TPUv5p, without slowing down performance during training

Image 4: Image
Image 4: Image

3:05 PM · Apr 23, 2026

10.1K Views