Model Selection

Not every task needs the absolute smartest model in the world.

If someone should explain something to you, do you always need the most competent person on the planet? If someone should find a hotel for you, does that person need a PhD in history, physics, and literature?

No. The same is true for LLMs.

Good models for the right task

elmtree uses models that are highly capable and can use our search agent reliably. The goal is not to use the largest model for every simple request, but to choose the right model class for the task.

For many questions, what matters is that the model evaluates current information cleanly, understands sources correctly, and presents the answer clearly. To do that, we combine strong models with a search agent that can research current information and integrate it into the answer.

Fine-tuned for elmtree

We have fine-tuned the models with PEFT so they respond better to what elmtree should do for you: evaluate current information, respect privacy boundaries, handle pseudonymization cleanly, and structure answers so they are useful in practice.