Skip to content
Google Developers Blog·· 1 d agoAI score55

Google Trains Gemma 4 Multi-Turn Tool-Calling Agent on TPU with Tunix

Multi-Turn Agentic Reinforcement Learning with Tunix and Gemma 4 on Google TPUs

AI summary

Google has released a tutorial on how to train a multi-turn tool-calling agent on a single Cloud TPU v6e-8 using Tunix's agentic RL pipeline, with Gemma 4 E2B's thinking mode as the policy model.

Source:Google Developers Blog · developers.googleblog.com