data_juicer.ops.mapper.dialog_memory_consistency_mapper module

LLM: final reply vs prior constraints/facts (1–5).

Inspired by OpenJudge-style context memory; DJ: single call, truncated turns.

Reference (similar public rubric intent): https://agentscope-ai.github.io/OpenJudge/built_in_graders/multi_turn/ #contextmemorygrader

class data_juicer.ops.mapper.dialog_memory_consistency_mapper.DialogMemoryConsistencyMapper(*args, **kwargs)

基类：_DialogTurnQualityMapper

Whether the final assistant turn respects prior user constraints and facts.

OP_NAME = 'dialog_memory_consistency_mapper'

META_KEY = 'dialog_memory_consistency'