df6eafeae2
- Add strip_parenthetical_asides_for_chat in reply pipeline before [SPLIT] - Expand output_rules bans (performance parens) and voice as warm host - Refocus opening/guided prompts on pulling conversation toward memoir oral history - Align interview opening fallbacks with memoir-first tone - Add unit tests for parenthetical stripping
126 lines
4.2 KiB
Python
126 lines
4.2 KiB
Python
"""访谈/资料追问：回复条数与单条字数硬限制（不靠长 prompt）。"""
|
||
|
||
from __future__ import annotations
|
||
|
||
import re
|
||
|
||
|
||
def strip_markdown_for_chat(text: str) -> str:
|
||
"""
|
||
将模型偶然输出的常见 Markdown 剥成纯文本，供 App 聊天气泡展示。
|
||
保留换行与字面量 [SPLIT]；不做完整 MD 解析，以简单可预测为主。
|
||
"""
|
||
if not text:
|
||
return text
|
||
s = text
|
||
# 围栏代码块（含首行语言标记）：整段替换为块内正文，去掉栅栏
|
||
s = re.sub(
|
||
r"```(?:[^\n`]*)\n([\s\S]*?)```",
|
||
r"\1",
|
||
s,
|
||
flags=re.MULTILINE,
|
||
)
|
||
s = s.replace("```", "")
|
||
# 图片  → alt；链接 [label](url) → label
|
||
s = re.sub(r"!\[([^\]]*)\]\([^)]*\)", r"\1", s)
|
||
s = re.sub(r"\[([^\]]*)\]\([^)]*\)", r"\1", s)
|
||
# ATX 标题
|
||
s = re.sub(r"(?m)^#{1,6}\s+", "", s)
|
||
# 无序列表行首（仅限行首减号/星号/+ 后接空格，避免误判「—」）
|
||
s = re.sub(r"(?m)^\s*[-*+]\s+", "", s)
|
||
# 有序列表「数字. 」仅行首
|
||
s = re.sub(r"(?m)^\s*\d+\.\s+", "", s)
|
||
# 粗体/删除线常见标记
|
||
s = s.replace("**", "").replace("__", "")
|
||
s = s.replace("~~", "")
|
||
# 行内反引号
|
||
s = s.replace("`", "")
|
||
# 孤立 emphasis：*词* 或 _词_（不含跨行）
|
||
s = re.sub(r"(?<![*])\*([^*\n]+)\*(?![*])", r"\1", s)
|
||
s = re.sub(r"(?<![_])_([^_\n]+)_(?![_])", r"\1", s)
|
||
# 分割线
|
||
s = re.sub(r"(?m)^\s*---+\s*$", "", s)
|
||
return s
|
||
|
||
|
||
def strip_parenthetical_asides_for_chat(text: str) -> str:
|
||
"""
|
||
去掉模型输出的表演性括注（全角「（…）」与半角「(...)」），迭代至不再有可删对。
|
||
|
||
口述回忆录场景下助理回复几乎不需要夹注；若写成「（约1993年）」等说明也会被删，属产品上有意识取舍，
|
||
与禁止「（轻轻笑）」类舞台说明一致。须在 strip_markdown_for_chat 之后调用（链接里的 () 已先处理）。
|
||
"""
|
||
if not text:
|
||
return text
|
||
s = text
|
||
prev: str | None = None
|
||
while prev != s:
|
||
prev = s
|
||
s = re.sub(r"（[^）]*）", "", s)
|
||
s = re.sub(r"\([^)]*\)", "", s)
|
||
s = re.sub(r"[ \t]{2,}", " ", s)
|
||
return s.strip()
|
||
|
||
|
||
def segments_from_llm_response(
|
||
response_text: str,
|
||
*,
|
||
max_segments: int = 3,
|
||
min_paragraph_chars: int = 12,
|
||
) -> list[str]:
|
||
"""
|
||
优先按字面 [SPLIT] 拆段；若模型只输出一段、但用空行写了多段，再按段落拆。
|
||
解决「两段话 + 换行」却未写 [SPLIT] 时仍要拆气泡 / 多段 TTS 的情况。
|
||
"""
|
||
text = strip_markdown_for_chat((response_text or "").strip())
|
||
text = strip_parenthetical_asides_for_chat(text)
|
||
if not text:
|
||
return []
|
||
primary = [p.strip() for p in text.split("[SPLIT]") if p.strip()]
|
||
if len(primary) > 1:
|
||
return primary[:max_segments]
|
||
blob = primary[0] if primary else text
|
||
if "\n" not in blob:
|
||
return [blob]
|
||
paras = [p.strip() for p in re.split(r"\n\s*\n+", blob) if p.strip()]
|
||
if len(paras) < 2:
|
||
return [blob]
|
||
paras = [p for p in paras if len(p) >= min_paragraph_chars]
|
||
if len(paras) < 2:
|
||
return [blob]
|
||
return paras[:max_segments]
|
||
|
||
|
||
def nonempty_segments_or_fallback(
|
||
segments: list[str],
|
||
*,
|
||
fallback: str,
|
||
) -> list[str]:
|
||
"""去掉空段；若全部为空白/空串则返回单条 fallback，避免 WS 下发空 text。"""
|
||
cleaned = [s for s in segments if (s or "").strip()]
|
||
if cleaned:
|
||
return cleaned
|
||
fb = (fallback or "").strip()
|
||
return [fb] if fb else ["…"]
|
||
|
||
|
||
def truncate_chat_segments(
|
||
segments: list[str],
|
||
*,
|
||
max_segments: int,
|
||
max_chars_per_segment: int,
|
||
) -> list[str]:
|
||
"""保留前 max_segments 条，每条截断至 max_chars_per_segment（按字符数，中文友好）。"""
|
||
if not segments:
|
||
return []
|
||
out: list[str] = []
|
||
for raw in segments[:max_segments]:
|
||
s = (raw or "").strip()
|
||
if not s:
|
||
continue
|
||
if len(s) > max_chars_per_segment:
|
||
# 保留 1 个字符给省略号，使总长度不超过上限
|
||
s = s[: max_chars_per_segment - 1].rstrip() + "…"
|
||
out.append(s)
|
||
return out
|