Grug-27b community fine-tune claims 90% token reduction over Qwen 3.6 27B

A community fine-tune of Qwen 3.6 27B named grug-27b claims significant improvements, including a 90% reduction in required tokens and better benchmark performance.