When the model chooses the most likely next token for output, it restricts the choices to the value of top_k. the default is 40, but can be higher or lower.

#ollama/parameters