nip.utils.types.DpoDatasetItem

Contents

nip.utils.types.DpoDatasetItem#

class nip.utils.types.DpoDatasetItem[source]#

A single item in a DPO dataset.

This is used for training language models with Direct Preference Optimization (DPO).

Attributes

input

The input chat history for the DPO dataset item.

preferred_output

The preferred part of the preference pair.

non_preferred_output

The non-preferred part of the preference pair.

Methods