How to set temperature parameters for creative output

What this does

Temperature controls the randomness of token selection. Higher values (0.7-1.0) produce more diverse and creative outputs, making them ideal for storytelling, brainstorming, and creative writing.

Steps

1. Set temperature via Ollama command line.

   ollama run llama3.2
   /set parameter temperature 0.9
   

   Then prompt: "Write a fantasy story opening."

2. Set temperature via the API.

   curl -s http://localhost:11434/api/generate \
     -d '{"model": "llama3.2", "prompt": "Write a poem about AI",
          "options": {"temperature": 0.8}, "stream": false}' \
     | jq -r '.response'
   

3. Create a Modelfile for permanently creative defaults.

   FROM llama3.2
   PARAMETER temperature 0.85
   PARAMETER top_p 0.95
   

   ollama create creative-llama -f Modelfile
   

4. Compare outputs at different temperatures.

   echo "=== Temperature 0.2 ==="
   curl -s ... -d '{"options":{"temperature":0.2}}' | jq -r '.response'
   echo "=== Temperature 0.9 ==="
   curl -s ... -d '{"options":{"temperature":0.9}}' | jq -r '.response'
   

   Expected: Low temperature produces predictable text; high temperature produces more varied word choices.

Verification

# Generate the same prompt 3 times at temperature 0.9
for i in 1 2 3; do
  curl -s http://localhost:11434/api/generate \
    -d '{"model":"llama3.2","prompt":"Name a fantasy kingdom","options":{"temperature":0.9},"stream":false}' \
    | jq -r '.response'
done
# Expected: Three different kingdom names (e.g., Eldoria, Drakemoor, Silvervale)

Common failures

• Temperature too high causes gibberish: Values above 1.2 can produce incoherent output. Stay within 0.0-1.5.
• Temperature too low for creativity: Values below 0.5 restrict word choice. For creative tasks, use 0.8-1.0.
• Perplexity increase with high temperature: Higher temperature = more surprising words. For plausible creativity, combine with top_p: 0.9.

Operator checkpoint

Before treating this as solved, write down the local runtime, model or package version, hardware/backend if relevant, and the verification output. This keeps the guide useful as a Will-It-Run style decision instead of a one-off command transcript.

Related guides

• How to set temperature to zero for deterministic responses
• How to tune temperature based on task type