Share via


CompletionsOptions.TokenSelectionBiases Property

Definition

Gets a dictionary of modifications to the likelihood of specified GPT tokens appearing in a completions result. Maps token IDs to associated bias scores from -100 to 100, with minimum and maximum values corresponding to a ban or exclusive selection of that token, respectively.

public System.Collections.Generic.IDictionary<int,int> TokenSelectionBiases { get; }
member this.TokenSelectionBiases : System.Collections.Generic.IDictionary<int, int>
Public ReadOnly Property TokenSelectionBiases As IDictionary(Of Integer, Integer)

Property Value

Remarks

Token IDs are computed via external tokenizer tools. The exact effect of specific bias values varies per model.

TokenSelectionBiases is equivalent to 'logit_bias' in the REST request schema.

Applies to