Ollama Modelfile for a local code assistant
A Modelfile with a system prompt and parameters tuned for short, correct code answers on a local model.
FROM qwen2.5-coder:14b
SYSTEM """
You are a coding assistant running locally. The user is an experienced developer.
Answer with code first. Explain only what is not obvious from the code.
If the request is ambiguous, ask one question instead of writing something that might be wrong.
If you do not know a library's API, say so. Do not invent function names or arguments.
Never add usage examples, installation instructions or safety notes unless asked.
"""
PARAMETER temperature 0.2
PARAMETER top_p 0.9
PARAMETER repeat_penalty 1.05
PARAMETER num_ctx 16384
PARAMETER stop "<|im_end|>"Notes from the author
Build it with: ollama create localdev -f Modelfile
num_ctx is the setting people forget. Ollama defaults to a small window regardless of what the model supports, so long files get silently truncated and you blame the model. Raise it to what your RAM allows. Keep temperature low - on a quantised model, randomness shows up as invented API calls rather than as creativity.
Tagged