File size: 4,029 Bytes
188f4d8 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 | """Prompt / Extract 共用文本工具。
- ``strip_thinking``: 剥离模型输出里的 ``<think>...</think>`` / ``<answer>...</answer>``
- ``extract_by_prefix``: 按多前缀（"字体分类:" / "字体:" / ...）从纯文本中抽取冒号后的内容
- ``clean_value``: 清理首尾空白、代码块围栏、常见标点引号
"""
from __future__ import annotations
import re
_OTHER_PREFIX_RE = re.compile(r"^(?:字体分类|字体|分类|提取文本|提取结果|识别结果|识别文本|文本)\s*[:：]")
def strip_thinking(text: str) -> str:
"""剥离 thinking 段，只保留最终答案。
支持 ``<think>...</think><answer>...</answer>`` / 仅 ``</think>`` / 仅 ``<think>`` 等多种形态。
"""
if not text:
return ""
s = text
ans_m = re.search(r"<\s*answer\s*>([\s\S]*?)<\s*/\s*answer\s*>", s, flags=re.IGNORECASE)
if ans_m:
return ans_m.group(1).strip()
close_matches = list(re.finditer(r"<\s*/\s*(?:think|thinking|reasoning)\s*>", s, flags=re.IGNORECASE))
if close_matches:
last = close_matches[-1]
tail = s[last.end() :].strip()
tail = re.sub(r"^\s*<\s*answer\s*>\s*", "", tail, flags=re.IGNORECASE)
tail = re.sub(r"\s*<\s*/?\s*answer\s*>\s*$", "", tail, flags=re.IGNORECASE)
return tail.strip()
open_m = re.search(r"<\s*(?:think|thinking|reasoning)\s*>", s, flags=re.IGNORECASE)
if open_m:
head = s[: open_m.start()].strip()
return head if head else ""
return s.strip()
def strip_code_fence(s: str) -> str:
if s is None:
return ""
s = s.strip()
m = re.match(r"^`{3,}[^\n`]*\n?(.*?)\n?`{3,}\s*$", s, re.DOTALL)
if m:
return m.group(1).strip()
return s.strip("`").strip()
def clean_value(s: str) -> str:
if not s:
return ""
s = strip_code_fence(s)
s = s.strip().strip("。.；;，,\"'“”‘’")
return s.strip()
def extract_by_prefix(text: str, prefixes: list[str], merge_trailing_lines: bool = False) -> str:
"""按 ``前缀:`` 抽取冒号之后的答案，命中多次取最后一次。
``merge_trailing_lines=True`` 时，"同行答案 + 后续多行非空"一并合并，
用于"字符提取"任务（模型常把多行文本写在前缀之后）。
"""
if not text:
return ""
prefix_pattern = "|".join(re.escape(p) for p in prefixes)
head_pattern = re.compile(rf"(?:{prefix_pattern})\s*[:：]")
matches = list(head_pattern.finditer(text))
if not matches:
return ""
last = matches[-1]
tail = text[last.end() :]
first_line, _nl, rest = tail.partition("\n")
first_line_stripped = first_line.strip()
starts_with_fence = bool(re.match(r"^`{3,}", first_line_stripped))
first_line_no_fence = re.sub(r"^`{3,}[^\n`]*", "", first_line_stripped).strip("` ").strip()
if not starts_with_fence and first_line_no_fence:
if not merge_trailing_lines:
return clean_value(first_line_no_fence)
follow_lines: list[str] = []
for ln in rest.splitlines():
ln_s = ln.strip().strip("`").strip()
if not ln_s:
continue
if _OTHER_PREFIX_RE.search(ln_s):
break
follow_lines.append(ln_s)
head_clean = clean_value(first_line_no_fence)
if not follow_lines:
return head_clean
cleaned_follow = [(clean_value(ln) or ln) for ln in follow_lines]
return "\n".join([head_clean, *cleaned_follow]).strip()
multiline_src = rest
if starts_with_fence:
close_m = re.search(r"\n?`{3,}\s*(\n|$)", rest)
if close_m:
multiline_src = rest[: close_m.start()]
lines = [ln.strip().strip("`").strip() for ln in multiline_src.splitlines()]
lines = [ln for ln in lines if ln]
if not lines:
return ""
if len(lines) == 1:
return clean_value(lines[0])
cleaned = [(clean_value(ln) or ln) for ln in lines]
return "\n".join(cleaned).strip()
|