aminroudaki/ decisio-gemma:q4_k_m

25 3 days ago

Decision Model: Answers yes/no, multiple-choice and scoring questions about any text, returning a probability for every option.

tools decision thinking 
curl http://localhost:11434/v1/systemone \
  -H "Content-Type: application/json" \
  -d '{
    "model": "aminroudaki/decisio-gemma:q4_k_m",
    "state": "Hello World",
    "questions": {
      "says_hello": {
        "type": "noul",
        "instructions": "Does the state text contain a greeting?",
        "criteria": {
          "true": "The state text contains a greeting.",
          "false": "The state text does not contain a greeting."
        }
      }
    }
  }'

Details

3 days ago

f7a2d8611829 · 7.7GB

model
gemma4
·
11.9B
·
Q4_K_M
system
Answer with the letter only.
params
{ "num_ctx": 8192 }

Readme

Decisio-gemma

decisio-gemma answers closed questions about any text with Gemma 4 12B, with a probability for every option. Give it a piece of text and one or more questions (yes/no, a choice among up to 26 options, or a score on a scale you define); it returns, for each question, the chosen answer and how likely each option is, from one forward pass per question.

How to use it

Requires Ollama 0.35.1 or later: the decision route (POST /v1/systemone) serves this model through its declared decision capability.

ollama pull aminroudaki/decisio-gemma
curl http://localhost:11434/v1/systemone -d '{
  "model": "aminroudaki/decisio-gemma",
  "state": "Order 4471 shipped on Monday. The customer writes: My parcel arrived with the box crushed and the lamp inside broken.",
  "questions": {
    "intent": {"type": "choice", "instructions": "What does the customer want?",
               "criteria": {"track_order": null, "report_damage": null, "cancel_order": null, "change_address": null}},
    "refund": {"type": "noul", "instructions": "Is the customer likely to ask for a refund or replacement?"}
  }
}'

What is in it

  • Weights: Gemma 4 12B (google/gemma-4-12B-it at revision 707f0a3b, Apache-2.0), unmodified, in two community quantisations from bartowski/gemma-4-12B-it-GGUF (llama.cpp b9496): q4_k_m (also latest) and q8_0.
  • Prompt: Ollama builds its own prompt for each question; this model adds one system line (“Answer with the letter only.”) and the chat template’s rendering, which on this route matches Google’s template for the pinned revision byte for byte.
  • No calibration: the probabilities are the model’s own softmax over the option letters; Ollama applies no temperature and no calibration.

How it compares with the decisio server

On 50 questions from decisio’s test suite, this listing chose the same answer as a decisio server on the Gemma base for 43 (q4_k_m) and 44 (q8_0). The decisio server builds its own prompt for Gemma and calibrates every answer; for that, run decisio itself with --base gemma-4-12b: github.com/aminry/decisio. For the Qwen base, see aminroudaki/decisio.