GeneratorArgs.is_torchtune_model is a misnomer

[Jack-Khuu]

🚀 The feature, motivation and pitch

is_torchtune_model is a misnomer and can result in buggy code. It gates logic for models that have tune suffix, but not all torchtune models end with this suffix. For example Flamingo (Llama3.2 11B) is also a torchtune model

This results in code like this:

torchchat/torchchat/generate.py

Lines 611 to 614 in d0993b3

	if (
	self.is_torchtune_model
	or self.model.config.model_type == ModelType.Flamingo
	):

Alternatives

No response

Additional context

No response

RFC (Optional)

While the logic around torchtune models itself needs some refactor, the short term solution is some combination of:

• Rename the field to be more accurate
• Refactor the logic such that it does account for torchtune models that do not end in tune

[krammnic]

@Jack-Khuu Hey! If it is still relevant I can take it. (We want to have better integration with torchtune)