Gemma 2 posts

Every post filed under Gemma, newest first.

August 20, 20266 min read

DeepMind's recirculation trick lifts Gemma 3 without…

Google DeepMind's recirculation paper adds inference-time recurrence to frozen Gemma 3 checkpoints. Adaptive recirculation cuts perplexity about 23% and lifts GSM8K accuracy about 21%, with almost no extra cost during token generation but slower prefill.

Unsloth’s Dynamic GGUFs make Gemma 4 12B fit on a laptop

Models

August 3, 20265 min read

Unsloth’s Dynamic GGUFs make Gemma 4 12B…

Google’s Gemma 4 12B is multimodal with a 256K context window. Unsloth’s Dynamic GGUFs squeeze a usable 4-bit build into roughly 8GB of memory without throwing quality off a cliff.