alex-treebeard/

gemma-3-270m-it-tldr

This is a fine-tune of gemma, trained using GRPO with a reward function to incentivise ~40 char outputs beginning with "I" such that it outputs TL;DR summaries for reddit comments.