Skip to content

tinker_cookbook.preference

Class Description
Comparison A pair of completions (A and B) generated from the same prompt conversation.
LabeledComparison A Comparison annotated with a human preference label (A, B, or Tie).
ComparisonRenderer Abstract renderer for converting Comparisons to model inputs for preference trai
ComparisonRendererFromChatRenderer Wraps a chat Renderer to render Comparisons for preference training.
PreferenceModel Abstract base class for models that score a Comparison and return a preference f
PreferenceModelBuilder Abstract builder that creates PreferenceModel instances.
PreferenceModelFromChatRenderer A PreferenceModel that uses a chat renderer and Tinker sampling client.
PreferenceModelBuilderFromChatRenderer Builds a PreferenceModel that uses a chat renderer and a Tinker sampling client.
Config Configuration for Direct Preference Optimization (DPO) training.