Skip to content

decider:4b

5 TagsUpdated 4.2B params32768 contextEnglishApache-2.0by Mapika

Qwen3.5-4B base (v2.1), 0.680 on typed decisions, one fitted temperature per answer type. Best on a GPU.

4.2b
ollaya run decider:4b --preset triage "I was charged twice for my subscription this month and want a refund."

Details

  • graph403657f8e9ff · 11 MBonnx · decider · 4.2B · fp32
  • weightsee8ce585b3ce · 8.4 GBhuggingface.co/Mapika/decider-4b/resolve/eb5fbdf…/model.safetensors
  • tokenizer06b9509352d2 · 20 MBhuggingface.co/Mapika/decider-4b/resolve/eb5fbdf…/tokenizer.json
  • decision80132837e286 · 9 KB{"engine": "onnx", "family": "decider", "encoder": "", "layout": "decider-slots-v1", …}
  • calibration7c9333f3da10 · 290 B{"temperature": [1.11, 1.0, 1.56]}
  • licenseb4d9d0d19903 · 10 KBdecider by Mapika (https://huggingface.co/Mapika/decider-2b)

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.

decider readme and all models