Datasets:
File size: 6,244 Bytes
ca6f7c5 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 | import json
import re
image_special_token = "<special_image_token_12345>"
user_template = "You are given a document with text and images, and a question. Answer the question as concisely as you can, using a single phrase or sentence if possible. If the question cannot be answered based on the information in the article, write 'Not answerable.' Write your answer in the following format:\nAnswer: [answer]\n\n{context}\n\nQuestion: {question}"
item_template = "Document {doc_id:.15} (page {page_id}): " + image_special_token
system_template = "Answer:"
def parse_qa_pair_output(model_output):
result = {
"qa_evidence_description": None,
"qa_json": None,
}
# 匹配: [JSON]:, ### JSON, ### Part 2: JSON Output, **JSON**, JSON: (忽略大小写)
json_header_pattern = (
r"(?:^|\n)\s*(?:[#*]+\s*)?(?:\[\s*)?(?:Part\s*\d+[:.]?\s*)?JSON(?:\s*Output)?(?:\]\s*)?\**[:]?"
)
# 匹配: [Evidence Description]:, ### Part 1: Evidence Description, ### Evidence Description
evidence_header_pattern = (
r"(?:^|\n)\s*(?:[#*]+\s*)?(?:\[\s*)?(?:Part\s*\d+[:.]?\s*)?Evidence(?:\s*Description)?(?:\]\s*)?\**[:]?"
)
# 2. locate the start of JSON part
json_match = re.search(json_header_pattern, model_output, re.IGNORECASE)
json_text_segment = ""
evidence_text_segment = ""
if json_match:
start_of_json_header = json_match.start()
end_of_json_header = json_match.end()
evidence_text_segment = model_output[:start_of_json_header].strip()
json_text_segment = model_output[end_of_json_header:].strip()
else:
print(f"警告: 未找到明确的 JSON 分隔符，尝试全文解析, 输入文本: {model_output}")
evidence_text_segment = model_output
json_text_segment = model_output
# 3. cut Evidence Description
if evidence_text_segment:
# remove header
ev_header_match = re.search(evidence_header_pattern, evidence_text_segment, re.IGNORECASE)
if ev_header_match:
evidence_text_segment = evidence_text_segment[ev_header_match.end() :].strip()
# remove markdown separator (like ---, ***)
evidence_text_segment = re.sub(r"\n\s*[-*_]{3,}\s*$", "", evidence_text_segment).strip()
result["qa_evidence_description"] = evidence_text_segment
# 4. parse JSON
# remove markdown code: ```json ... ```
json_text_segment = re.sub(r"^```[a-zA-Z]*\s*", "", json_text_segment) # remove the starting ```json
json_text_segment = re.sub(r"\s*```$", "", json_text_segment) # remove the ending ```
result["qa_json"] = robust_json_extraction(
json_text_segment, ["question", "answer", "answer_format", "evidence_pages", "evidence_sources"]
)
# must have keys: question, answer
if result["qa_evidence_description"] is None or result["qa_json"] is None:
raise ValueError(f"必须包含 Evidence Description 和 JSON 字段, 输入文本: {model_output}")
if not isinstance(result["qa_evidence_description"], str):
raise ValueError(f"Evidence Description 必须是字符串类型, 输入文本: {model_output}")
if result["qa_json"]["question"] is None or result["qa_json"]["answer"] is None:
raise ValueError(f"API未提供有效答案, 输入文本: {model_output}")
if (
result["qa_json"]["answer_format"] is None
or result["qa_json"]["evidence_pages"] is None
or result["qa_json"]["evidence_sources"] is None
):
raise ValueError(f"API未提供有效答案, 输入文本: {model_output}")
if not isinstance(result["qa_json"]["question"], str):
raise ValueError(f"JSON 中的 question 必须是非空字符串类型, 输入文本: {model_output}")
result["qa_json"]["answer"] = str(result["qa_json"]["answer"])
return result
def clean_latex_escape(text):
text = re.sub(r"(?<!\\)\\([a-zA-Z]{2,})", r"\\\\\1", text)
return text
def robust_json_extraction(text, key_list=None):
text = clean_latex_escape(text)
start_index = text.find("{")
if start_index != -1:
try:
decoder = json.JSONDecoder()
obj, _ = decoder.raw_decode(text, idx=start_index)
return obj
except json.JSONDecodeError as e:
print(f"标准JSON解析失败: {e}, \n 输入文本: {text} 尝试正则提取字段...")
extracted_data = {}
print(f"正在尝试正则提取以下字段: {key_list}")
for key in key_list:
# 1. "key" \s* : \s* -> 匹配键和冒号
# 2. 捕获组 (...):
# a. "(?:[^"\\]|\\.)*" -> 匹配双引号字符串 (修复了你的笔误，支持转义字符)
# b. -?\d+(?:\.\d+)? -> 匹配数字
# c. true|false|null -> 匹配字面量
# d. \[.*?\]|\{.*?\} -> 简单匹配数组/对象 (非递归，不支持嵌套太深)
pattern = rf'"{re.escape(key)}"\s*:\s*("(?:[^"\\]|\\.)*"|-?\d+(?:\.\d+)?|true|false|null|\[.*?\]|\{{.*?\}})'
match = re.search(pattern, text, re.DOTALL)
if match:
value_str = match.group(1)
try:
extracted_data[key] = json.loads(value_str)
except json.JSONDecodeError:
extracted_data[key] = None
# try:
# import ast
# extracted_data[key] = ast.literal_eval(value_str)
# except:
# extracted_data[key] = None
else:
extracted_data[key] = None
return extracted_data
if __name__ == "__main__":
test_text = """{"question": "What method did the Utah Symphony use to complete Step 2 of the shared value finding process, and what are the components of this method?", "answer": "['Structured \\'Values\\' Interviews using the Zaltman Metaphor Elicitation Technique (ZMET)', 'Storytelling', 'Missing Pictures', 'Triad Task', 'Sensory Metaphors', 'Additional Probing Techniques']", "answer_format": "List", "evidence_pages": [22, 23, 25, 27], "evidence_sources": ["Layout", "Text"]}"""
result = robust_json_extraction(
test_text, ["question", "answer", "answer_format", "evidence_pages", "evidence_sources"]
)
print(result)
|