Finetuning Gemma 3 270M model

  • 2025/08/26
  • Gemma 3 270M finetuning

task

I had never fine-tuned an LLM before, so when the Gemma 3 270M model was released with claims that fine-tuning unlocks its true potential, I decided to follow a tutorial to see what the process actually involves. I wanted to understand the effort required to fine-tune a model and build both a frontend and backend around it. I used the example from Google's business-email-assistant tutorial and had Claude Opus 4.1 in Zed editor handle the rest of the implementation.

This is the outcome.

./images/gemma-3-270m-finetune-1.png

Links