Google uses Tunix to train multi-turn tool-calling agents on TPUs
Understand this story
Google has released a tutorial on how to use Tunix's agent reinforcement learning pipeline to train Gemma 4 E2B on Cloud TPU v6e-8 to perform multi-turn tool calling. The tutorial uses the Plane Strike board game as the environment, and claims that after training the win rate rose from 9.4% to 95%, with the average number of steps dropping from 22.44 to about 16. The tutorial breaks down four modules: the environment, Gemma 4 native tool-call parsing, the conversational agent, and GRPO training, and provides a complete code repository and a 100-step training command. The above changes in win rate and step count are the results Google gives in the tutorial.
AI overview based on source reports · Updated 18 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Google Developers BlogMulti-Turn Agentic Reinforcement Learning with Tunix and Gemma 4 on Google TPUs
This post shows how to train a multi-turn, tool-using LLM agent with Tunix's agentic reinforcement learning pipeline on a single Cloud TPU v6e-8, using Gemma 4 E2B in thinking mode to play the 6x6 Plane Strike board game. The pipeline is built from four components: a BaseTaskEnv subclass, a Gemma4ChatTemplateParser subclass for native tool declarations and in-turn tool responses, a ConversationAgentBase subclass, and a GRPOLearner with asynchronous vllm-tpu rollouts. After 100 steps of agentic GRPO, the agent's win rate rises from 9.4% on held-out boards to about 95%, with average game length falling from 22.44 to roughly 16 strikes.
Activity for this story
- Current comparable activity
- 4
- Peak comparable activity
- 6Nov 20 02:00
- Change in the past 24 hours
- –
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.