data_juicer.ops.mapper.dialog_quality_llm_utils module#

Prompt builders and JSON helpers for turn-quality LLM mappers.

data_juicer.ops.mapper.dialog_quality_llm_utils.extract_json_object(text: str) → dict | None[源代码]#

data_juicer.ops.mapper.dialog_quality_llm_utils.build_dialog_turn_eval_user_content(sample: dict, *, history_key: str, query_key: str, response_key: str, max_round: int, max_query_chars: int, max_response_chars: int) → str[源代码]#: Earlier turns + last user message + assistant span to score.

data_juicer.ops.mapper.dialog_quality_llm_utils.build_agent_trace_eval_user_content(sample: dict, *, text_key: str, max_chars: int) → str[源代码]#: Flattened session text (e.g. after agent_dialog_normalize_mapper).

data_juicer.ops.mapper.dialog_quality_llm_utils.build_agent_tool_fit_user_content(sample: dict, *, query_key: str, response_key: str, tool_types_key: str, primary_tool_key: str, max_query_chars: int, max_response_chars: int) → str[源代码]#

data_juicer.ops.mapper.dialog_quality_llm_utils.normalize_score_1_5(obj: dict | None) → Dict[str, Any][源代码]#