Skip to content

winnow:12b

3 TagsUpdated 12B params8192 context100+ languagesApache-2.0by EldanRing

Winnow-12B, Q8_0 GGUF: 0.702 on typed decisions, 85.7 % on the author's JevBench set; a 16 GB GPU holds it.

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

Details

  • weightsb710efc4c0d0 · 12.7 GBgguf · Q8_0 · huggingface.co/EldanRing/Winnow-12B/resolve/b6ac22b…/gguf/Winnow-12B-Q8_0.gguf
  • decisiond8c4d73b8e66 · 2 KB{"engine": "llama", "family": "winnow", "layout": "winnow-v1", "gguf": {…}, …}
  • calibration0088393cc0e6 · 189 B{"temperature": [1.0, 1.0, 1.0]}
  • licensecaa0c5ceea85 · 12 KBWinnow-12B

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.

Winnow readme and all models