topK

For each token selection step, the `top_k` tokens with the highest probabilities are sampled. Then tokens are further filtered based on `top_p` with the final token selected using temperature sampling. Use a lower number for less random responses and a higher number for more random responses.