Learn Mistral

3

Managing the Model

Give a man a fish, and you feed him for a day;

Teach a man to fish, and you feed him for a lifetime.

— Maimonides

Imagine a seasoned fisher standing at the shore, casting a net into a carefully chosen lake where the right kind of catch awaits. The decision about where and how to cast is much like setting the conditions for a language model—choices that shape the results we can expect. As the net sinks, its weave decides whether only prized fish are caught or a broader variety slips through, echoing how parameters such as top_k and top_p control the model’s selectivity.

In the same way that a fisher depends on the right waters, net, and bait, we depend on parameters to steer generation. Carrying that image forward, we’ll ...