🦦 MOVED — This repo has been replaced

The Gemma 4 31B study has moved to its own repository:

→ https://github.com/KikoCisBot/gemma4-31b-study

Interactive site: https://kikocisbot.github.io/gemma4-31b-study/

Featured model

KikoCis/gemma-4-31b-it-IQ2_M-GGUF — 10.17 GB custom IQ2_M quantization that matches f16 Char-F1 and beats f16 on BLEU-4 (22.39 vs 21.02) at 6× smaller.

This repository is kept read-only for historical links. Please update your bookmarks.