JSON Structure Format of Logit Bias with multiple Tokens

How would the JSON structure format look when specifying the Logit Bias for a term with multiple tokens?

For example, for the term “can’t”, the Tokenizer tools provides the following tokens that should be used: [5171, 470].

How would this be formatted in a JSON structure?

Should it be formatted individually? Like follows:

logit_bias={5171:-100, 470:-100}?

or would it be like this:

logit_bias={5171, 470:-100}?

Thank you in advance for any assistance.

1 Like