Autonomous LLM post-training with Tunix on TPUs- Google Developers Blog

Introduces 'autofinetune', a project applying autonomous research loops to LLM post-training (SFT and RL via GRPO) using Google's AI stack (Tunix, Gemma, Cloud TPUs, Antigravity CLI, Gemini Flash). Details the paradigm shift from manual tuning to AI-driven loops, outlining the 'Design the Arena', 'Provide the Execution Code', and 'Let the Agent Iterate' process. Presents two case studies: SFT on FunctionGemma using Tunix on Cloud TPU v5e, and RL (GRPO) on Gemma 3 1B for math reasoning using Tunix on Cloud TPU v6e. Highlights the use of `program.md` for defining agent boundaries and `results.tsv` for logging outcomes.