Tag
A comparison of Gemma 4 12B and 31B models shows that the smaller model retains reasoning capabilities nearly intact but suffers significant knowledge loss, making it ideal for reasoning tasks while the larger model is better for broad knowledge Q&A.
A full educational series on local LLMs, covering inference, tokens, weights, and system-level understanding for beginners and reference.