07c6478742
- 新增 utterance_substance：短时/应答/元话语可跳过记忆检索、阶段 LLM 与资料抽取 LLM；可配置 - 输入归一化：LLM 模式默认仅语音/ASR；配置项写入 .env.example - Memoir Phase1：可选 batch LLM 一次性抽取+分类（失败回退逐段）；Extraction 空槽位时阶段与 current_stage 对齐，prompt 约束收紧 - 叙事与忠实度：narrative_safety、证据重叠/场合锚点、标题 slots 与履历短语 grounded；fidelity 解析失败 fail-open 可配置 - 章节管线：锁 TTL 上调、锁竞争 Celery 重试、Phase2 immediate singleflight 等；story_pipeline_sync / chapter_compose / memoir_tasks 联动 - Memory：compaction / repo / summarizer / evidence 小修；事实 FTS 未命中是否回退最近事实可配置 - 新增 memoir_pipeline_trace；补充 memoir_reliability 文档与多项回归/门控测试
65 lines
2.0 KiB
Python
65 lines
2.0 KiB
Python
"""
|
||
聊天输入归一：供访谈 Agent / 编排层对 ASR 与键盘输入做可控预处理（规则 / 可选 LLM）。
|
||
|
||
不改变 segment 落库原文；仅作为模型侧派生净稿。
|
||
与 memoir 共用同一套确定性规则，避免聊天与回忆录对同一句理解割裂。
|
||
"""
|
||
|
||
from __future__ import annotations
|
||
|
||
from typing import Any
|
||
|
||
from app.core.config import settings
|
||
from app.core.text_normalize import apply_oral_rules, llm_normalize_text
|
||
from app.core.logging import get_logger
|
||
|
||
logger = get_logger(__name__)
|
||
|
||
apply_conversation_input_rules = apply_oral_rules
|
||
|
||
|
||
def _llm_normalize_chat_input(text: str, llm: Any) -> str | None:
|
||
"""仅修正明显错字与同音字，不增事实；失败返回 None。"""
|
||
return llm_normalize_text(
|
||
text,
|
||
llm,
|
||
max_input_chars=int(settings.chat_input_normalize_llm_max_input_chars),
|
||
max_tokens=int(settings.chat_input_normalize_llm_max_tokens),
|
||
agent_name="chat_input_normalize.llm",
|
||
)
|
||
|
||
|
||
def normalize_chat_input_for_agent(
|
||
text: str,
|
||
*,
|
||
llm: Any | None = None,
|
||
is_from_voice: bool = False,
|
||
) -> str:
|
||
"""
|
||
聊天侧单一出口：编排层与 InterviewAgent 共用。
|
||
|
||
- 全局关闭：原文
|
||
- off：原文
|
||
- rules：仅规则
|
||
- llm：先规则，再（可选）LLM；无 llm 或失败则保留规则结果
|
||
- chat_input_normalize_llm_voice_only：mode=llm 时仅 is_from_voice 为真才调用 LLM
|
||
"""
|
||
if not settings.chat_input_normalize_enabled:
|
||
return text or ""
|
||
mode = (settings.chat_input_normalize_mode or "rules").strip().lower()
|
||
if mode == "off":
|
||
return text or ""
|
||
|
||
base = apply_conversation_input_rules(text or "")
|
||
if mode != "llm":
|
||
return base
|
||
|
||
effective_llm = llm
|
||
if settings.chat_input_normalize_llm_voice_only and not is_from_voice:
|
||
effective_llm = None
|
||
|
||
refined = _llm_normalize_chat_input(base, effective_llm)
|
||
if refined is not None:
|
||
return refined
|
||
return base
|