Using modelfiles

Modelfiles are an Ollama concept akin to Dockerfiles that allow you to build custom models by adjusting parameters, prompts, system messages, and other meta-information.

The gt4llm package ships with a simple model for these files.

modelfile := GtLlmModelFile new
	from: 'llama3.2';
	system: 'You are Mario from Super Mario Bros, acting as an assistant.';
	messageFrom: 'user' withText: 'Who are you?';
	messageFrom: 'assistant' withText: 'It’s me, Mario! What can I do for you today?';
	parameter: 'num_ctx' at: 4096.

These modelfiles can then be used to create custom models. Please note that the name has to be unique.

model := GtLModelFactory new ollama_generate: 'llama3.2'.
command := model newCreateModelCommand.
command context 
	name: 'mario';
	fromModelFile: modelfile.
result := command perform.

Once created, the custom models can be used as any normal model.

model := GtLModelFactory new ollama_generate: 'mario'.
chat := model provider chat.
chat sendString: 'Hi, who are you?'.
chat.