top
Optional. Specifies the top-k sampling threshold. The model considers only the top k most probable tokens for the next token. This can be useful for generating more coherent and less random text. For example, a top_k of 40 means the model will choose the next word from the 40 most likely words.