The percentage of most-likely candidates that are considered for the next token.
Must be 0 to 1.
For example, if you choose a value of 0.8 for topP, the model selects from the top 80% of the probability distribution of tokens that could be next in the sequence.
