Skip to content

cygnet:12b

2 TagsUpdated 12B params16384 context100+ languagesApache-2.0by blockbrain-ai

Gemma 4 12B IT (Q8_0) with Cygnet's prompt and temperature 3.4: 0.683 on typed decisions, up to 20 options.

multilingual12b
ollaya run cygnet:12b --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • weightsabfc3044b937 · 12.7 GBgguf · Q8_0 · huggingface.co/ggml-org/gemma-4-12B-it-GGUF/resolve/e3e6817…/gemma-4-12B-it-Q8_0.gguf
  • decision9f8101a4a32d · 2 KB{"engine": "llama", "family": "cygnet", "layout": "cygnet-v1", "gguf": {…}, …}
  • calibrationc474a28d0466 · 175 B{"temperature": [3.4, 3.4, 3.4]}
  • license0e3d9fc27a00 · 10 KBCygnet by blockbrain-ai (https://github.com/blockbrain-ai/cygnet-recipe), a prompt and calibration for Gemma 4 12B IT by Google DeepMind, MIT. The weights are google/gemma-4-12B-it (Apache-2.0), as ggml-org's GGUF conversion of the revision Cygnet pins.

Every layer is checked against its sha256 when it is pulled. Weights and tokenizers download from the model author's Hugging Face repository at a pinned commit; Ollaya never re-hosts them.

Cygnet readme and all models