From e22a78a2f2857ff98ec624b7c4f5c15b2c8362dd Mon Sep 17 00:00:00 2001
From: YM <479443481@qq.com>
Date: 星期一, 12 五月 2025 16:19:22 +0800
Subject: [PATCH] Merge branch 'master' of http://182.92.203.7:2001/r/KnowledgeBase
---
knowledgebase/doc/docx_split.py | 201 +++++++++++++++++++++++++
knowledgebase/gen_base_db/__init__.py | 6
.gitignore | 1
vision_test.py | 139 +---------------
knowledgebase/doc/image_to_text.py | 44 +++++
knowledgebase/gen_base_db/db_generate.py | 10 +
knowledgebase/gen_base_db/json_generate.py | 11 +
knowledgebase/llm.py | 17 ++
8 files changed, 302 insertions(+), 127 deletions(-)
diff --git a/.gitignore b/.gitignore
index 7bc84b4..9256fae 100644
--- a/.gitignore
+++ b/.gitignore
@@ -5,4 +5,5 @@
/.conda
/docs
/out*
+/packages
__pycache__
\ No newline at end of file
diff --git a/knowledgebase/doc/docx_split.py b/knowledgebase/doc/docx_split.py
new file mode 100644
index 0000000..52df48f
--- /dev/null
+++ b/knowledgebase/doc/docx_split.py
@@ -0,0 +1,201 @@
+# -*- coding: utf-8 -*-
+#
+# @author: lyg, ym
+# @date: 2025-5-8
+# @version: 1
+# @description: docx鏂囨。鎷嗗垎鍣紝鏍规嵁娈佃惤鎷嗗垎锛屽皢鍥剧墖鍜岃〃鏍艰浆鎹负json鏁版嵁銆�
+import docx
+import docx.table
+import json
+from dataclasses import dataclass
+from PIL import Image
+import io
+import re
+
+from knowledgebase.doc.image_to_text import ImageToText
+
+
+@dataclass
+class ParagraphInfo:
+ """
+ 娈佃惤淇℃伅
+ :param text: str - 娈佃惤鏂囨湰
+ :param level: int - 娈佃惤绾у埆锛�1-9绾ф爣棰橈紝0琛ㄧず姝ｆ枃
+ :param title_no: str - 鏍囬缂栧彿锛屽1.1銆�1.1.1绛�
+ """
+ text: str
+ level: int
+ title_no: str
+
+ @property
+ def full_text(self):
+ """
+ 鑾峰彇娈佃惤瀹屾暣鏂囨湰锛屽寘鍚爣棰樼紪鍙�
+ :return: str - 娈佃惤瀹屾暣鏂囨湰
+ """
+ return f"{self.title_no} {self.text}"
+
+ def __init__(self, text: str, level: int):
+ """
+ 娈佃惤淇℃伅
+ :param text: str - 娈佃惤鏂囨湰
+ :param level: int - 娈佃惤绾у埆锛�1-9绾ф爣棰橈紝0琛ㄧず姝ｆ枃
+ """
+ self.text = text
+ self.level = level
+ self.title_no = ''
+
+
+class DocSplit:
+ """
+ docx鏂囨。鎷嗗垎鍣紝鏍规嵁娈佃惤鎷嗗垎锛屽皢鍥剧墖鍜岃〃鏍艰浆鎹负json鏁版嵁銆�
+ 1.灏佽娈佃惤淇℃伅
+ 2.灏嗗浘鐗囧拰琛ㄦ牸杞崲涓簀son
+ 3.灏嗘钀芥寜鐓ф枃妗ｆ爣棰樼骇鍒粍鍚堟垚鏍戝舰缁撴瀯
+
+ """
+
+ def __init__(self, doc_file):
+ self.doc_file = doc_file
+ self.image_to_text = ImageToText()
+ self.paragraphs:list[ParagraphInfo] = []
+
+ def table_to_json(self, table: docx.table.Table):
+ """
+ 灏嗚〃鏍艰浆鎹负 JSON 鏍煎紡
+
+ :param table: docx.table.Table - 瑕佽浆鎹㈢殑琛ㄦ牸瀵硅薄
+ :return list - 琛ㄦ牸鏁版嵁锛屼互 JSON 鏍煎紡琛ㄧず
+ """
+ table_data = []
+ headers = []
+ first_row = True
+ row: docx.table._Row
+ for row in table.rows:
+ if first_row:
+ for cell in row.cells:
+ headers.append(cell.text)
+ first_row = False
+ continue
+ row_data = {}
+ row_idx = 0
+ for cell in row.cells:
+ if cell.tables:
+ # 宓屽琛ㄦ牸澶勭悊
+ if len(cell.tables) == 1:
+ text = self.table_to_json(cell.tables[0])
+ else:
+ text = []
+ for tbl in cell.tables:
+ tbl_json = self.table_to_json(tbl)
+ text.append(tbl_json)
+ else:
+ # 鍗曞厓鏍兼枃鏈幏鍙�
+ text = cell.text
+ row_data[headers[row_idx]] = text
+ row_idx += 1
+
+ table_data.append(row_data)
+ return table_data
+
+ def split(self):
+ """
+ 灏嗘枃妗ｆ媶鍒嗘垚娈佃惤锛屽苟杩斿洖娈佃惤鍒楄〃
+
+ :return: list[ParagraphInfo] - 娈佃惤鍒楄〃
+ """
+ document = docx.Document(self.doc_file)
+ table_cnt = 0
+ paragraph_cnt = 0
+
+ for element in document.element.body:
+ if element.tag.endswith('p'): # 娈佃惤
+ # 鑾峰彇鏍囬澶氱骇缂栧彿
+ paragraph = document.paragraphs[paragraph_cnt]
+ paragraph_text = paragraph.text
+ if paragraph_text:
+ self.paragraphs.append(ParagraphInfo(paragraph_text, self.get_header_level(paragraph)))
+ # 妫�鏌ユ槸鍚︽槸鍥剧墖锛屽鏋滄槸鍥剧墖鍒欒浆鎹负鏂囨湰
+ img_data = self.get_image_blob(paragraph)
+ if img_data:
+ text = self.gen_text_from_img(img_data)
+ self.paragraphs.append(ParagraphInfo(text, 0))
+ paragraph_cnt += 1
+ elif element.tag.endswith('tbl'): # 琛ㄦ牸
+ table = document.tables[table_cnt] # 鑾峰彇褰撳墠琛ㄦ牸瀵硅薄
+ table_cnt += 1
+ table_data = self.table_to_json(table)
+ self.paragraphs.append(ParagraphInfo(json.dumps(table_data, indent=4, ensure_ascii=False), 0))
+ else:
+ continue
+ # 鐢熸垚鏍囬缂栧彿
+ self.gen_title_no(self.paragraphs)
+
+ @staticmethod
+ def get_image_blob(paragraph):
+ # 閬嶅巻娈佃惤涓殑鎵�鏈塕un瀵硅薄锛堝浘鐗囬�氬父鍦ㄥ崟鐙殑Run涓級
+ for run in paragraph.runs:
+ xml = run._element.xml
+ if xml.find('v:imagedata') != -1:
+ # 浣跨敤姝ｅ垯琛ㄨ揪寮忔煡鎵緍:id灞炴��
+ match = re.search(r'r:id="([^"]+)"', xml)
+ if match:
+ r_id = match.group(1)
+ if r_id:
+ # 鑾峰彇鍥剧墖淇℃伅
+ image_part = paragraph.part.rels[r_id].target_part
+ return DocSplit.image_convert(image_part.blob, "png")
+ if xml.find('wp:inline') != -1 or xml.find('wp:anchor') != -1:
+ # 浣跨敤姝ｅ垯琛ㄨ揪寮忔煡鎵緍:embed灞炴��
+ match = re.search(r'r:embed="([^"]+)"', xml)
+ if match:
+ r_id = match.group(1)
+ if r_id:
+ # 鑾峰彇鍥剧墖淇℃伅
+ image_part = paragraph.part.rels[r_id].target_part
+ return DocSplit.image_convert(image_part.blob, "png")
+ return None
+
+ @staticmethod
+ def gen_title_no(paragraphs: list[ParagraphInfo]):
+ title_levels = [1, 1, 1, 1, 1, 1, 1, 1, 1]
+ for i in range(len(paragraphs)):
+ if paragraphs[i].level > 0:
+ for j in range(paragraphs[i].level - 1):
+ title_levels[j] = 1
+ paragraphs[i].title_no = '.'.join([str(x) for x in title_levels[0:paragraphs[i].level]])
+ title_levels[paragraphs[i].level - 1] += 1
+ else:
+ title_levels = [1, 1, 1, 1, 1, 1, 1, 1, 1]
+
+ @staticmethod
+ def get_header_level(paragraph) -> int:
+ if paragraph.style.base_style:
+ style = paragraph.style.base_style
+ else:
+ style = paragraph.style
+ if style and style.name.startswith('Heading'):
+ # 鑾峰彇鏍囬绾у埆
+ level = int(style.name.split(' ')[1])
+ return level
+ else:
+ return 0
+
+ @staticmethod
+ def image_convert(_in: bytes, _out_format: str) -> bytes:
+ in_io = io.BytesIO()
+ in_io.write(_in)
+ img = Image.open(in_io, "r")
+ out_io = io.BytesIO()
+ img.save(out_io, "png")
+ out_io.seek(0)
+ return out_io.read()
+
+ def gen_text_from_img(self, img_data:bytes):
+ return self.image_to_text.gen_text_from_img(img_data)
+
+if __name__ == '__main__':
+ doc_file = r'D:\workspace\PythonProjects\KnowledgeBase\doc\ZL鏍煎紡(鍏紑).docx'
+ doc_split = DocSplit(doc_file)
+ doc_split.split()
+ print("\n".join([x.full_text for x in doc_split.paragraphs]))
diff --git a/knowledgebase/doc/image_to_text.py b/knowledgebase/doc/image_to_text.py
new file mode 100644
index 0000000..b98c0b9
--- /dev/null
+++ b/knowledgebase/doc/image_to_text.py
@@ -0,0 +1,44 @@
+# -*- coding: utf-8 -*-
+#
+# @author: lyg
+# @date: 2025-5-8
+# @version: 1
+# @description: 鍒╃敤LLM灏嗗浘鐗囪浆涓烘枃鏈��
+
+from langchain_core.prompts import HumanMessagePromptTemplate, ChatPromptTemplate
+from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_core.output_parsers import JsonOutputParser
+import json
+import base64
+
+from knowledgebase.llm import vision_llm
+
+
+class ImageToText:
+ def __init__(self):
+ self.llm = vision_llm
+ self.prompt = ChatPromptTemplate.from_messages([
+ ("system", "浣犳槸涓�涓祫娣辫蒋浠跺伐绋嬪笀锛岃鍒嗘瀽鍥剧墖涓殑鍐呭銆�"),
+ (
+ "user",
+ [
+ {"type": "text", "text": "{msg}"},
+ {
+ "type": "image_url",
+ "image_url": {"url": "data:image/jpeg;base64,{image}"},
+ }
+ ],
+ )
+ ])
+
+ def gen_text_from_img(self, image: bytes) -> str:
+ """
+ 浠庡浘鐗囩敓鎴愭枃鏈��
+
+ :param image: 鍥剧墖鏁版嵁
+ :return: 鏂囨湰
+ """
+ image = base64.b64encode(image).decode()
+ chain = self.prompt | self.llm
+ resp = chain.invoke({"msg": "浣跨敤鑷劧璇█杈撳嚭鍥剧墖涓殑鍐呭锛屼笉瑕佸仛杩囧鐨勮В閲娿�傝緭鍑烘牸寮忎负绾枃鏈��", "image": image})
+ return resp.content
diff --git a/knowledgebase/gen_base_db/__init__.py b/knowledgebase/gen_base_db/__init__.py
new file mode 100644
index 0000000..2bb45e4
--- /dev/null
+++ b/knowledgebase/gen_base_db/__init__.py
@@ -0,0 +1,6 @@
+# -*- coding: utf-8 -*-
+#
+# @author:
+# @date:
+# @version:
+# @description:
diff --git a/knowledgebase/gen_base_db/db_generate.py b/knowledgebase/gen_base_db/db_generate.py
new file mode 100644
index 0000000..cbfa28a
--- /dev/null
+++ b/knowledgebase/gen_base_db/db_generate.py
@@ -0,0 +1,10 @@
+# -*- coding: utf-8 -*-
+#
+# @author:
+# @date:
+# @version:
+# @description:
+
+class DbGenerate:
+ def __init__(self):
+ pass
\ No newline at end of file
diff --git a/knowledgebase/gen_base_db/json_generate.py b/knowledgebase/gen_base_db/json_generate.py
new file mode 100644
index 0000000..c19de82
--- /dev/null
+++ b/knowledgebase/gen_base_db/json_generate.py
@@ -0,0 +1,11 @@
+# -*- coding: utf-8 -*-
+#
+# @author:
+# @date:
+# @version:
+# @description:
+
+from knowledgebase.llm import llm
+class JsonGenerate:
+ def __init__(self):
+ self.llm = llm
diff --git a/knowledgebase/llm.py b/knowledgebase/llm.py
new file mode 100644
index 0000000..ac6e035
--- /dev/null
+++ b/knowledgebase/llm.py
@@ -0,0 +1,17 @@
+# -*- coding: utf-8 -*-
+#
+# @author: lyg
+# @date: 2025-5-8
+# @version: 1
+# @description: 鍏叡langchain LLM 瀹炰緥
+from langchain_openai.chat_models import ChatOpenAI
+
+llm = ChatOpenAI(temperature=0,
+ model="qwen2.5-72b-instruct",
+ base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+ api_key="sk-15ecf7e273ad4b729c7f7f42b542749e")
+
+vision_llm = ChatOpenAI(temperature=0,
+ model="qwen2.5-vl-32b-instruct",
+ base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+ api_key="sk-15ecf7e273ad4b729c7f7f42b542749e")
\ No newline at end of file
diff --git a/vision_test.py b/vision_test.py
index f9872d4..fcc7138 100644
--- a/vision_test.py
+++ b/vision_test.py
@@ -1,5 +1,5 @@
# -*- coding: utf-8 -*-
-#
+#
# @author: lyg
# @date: 2025-5-7
# @version: 1
@@ -7,25 +7,24 @@
from langchain_openai.chat_models import ChatOpenAI
from langchain_core.prompts import HumanMessagePromptTemplate, ChatPromptTemplate
-from langchain_core.messages import HumanMessage, SystemMessage
+from langchain_core.messages import HumanMessage,SystemMessage
from langchain_core.output_parsers import JsonOutputParser
-from docx import Document
-from PIL import Image
-from io import BytesIO
-import re
import json
import base64
class VisionTest:
- def __init__(self, file):
- self.llm = ChatOpenAI(temperature=0, model="qwen2.5-72b-instruct",
- base_url="https://dashscope.aliyuncs.com/compatible-mode/v1", api_key="sk-15ecf7e273ad4b729c7f7f42b542749e")
+ def __init__(self,file):
+ self.llm = ChatOpenAI(temperature=0,
+ model="qwen2.5-72b-instruct",
+ base_url="https://dashscope.aliyuncs.com/compatible-mode/v1",
+ api_key="sk-15ecf7e273ad4b729c7f7f42b542749e")
+
image = base64.b64encode(open(file, 'rb').read()).decode()
self.prompt = ChatPromptTemplate.from_messages([
SystemMessage("浣犳槸涓�涓祫娣辫蒋浠跺伐绋嬪笀锛岃鍒嗘瀽鍥剧墖鍥炵瓟闂銆�"),
HumanMessage(content=[
- {"type": "text", "text": "{msg}"},
+ {"type": "text", "text": "describe the weather in this image"},
{
"type": "image_url",
"image_url": {"url": f"data:image/jpeg;base64,{image}"},
@@ -33,125 +32,11 @@
])
])
- def run(self, msg):
+ def run(self,msg):
chain = self.prompt | self.llm
resp = chain.invoke({"msg": msg})
print(resp.content)
- # def get_document_chapters(doc_path):
- # doc = Document(doc_path)
- # chapters = []
- # current_chapter = None
-
- # for para in doc.paragraphs:
- # if para.style.name.startswith('Heading'): # 妫�鏌ユ槸鍚︿负鏍囬鏍峰紡
- # level = int(para.style.name.replace('Heading', '')) # 鑾峰彇鏍囬绾у埆
- # current_chapter = {'level': level, 'title': para.text, 'content': []}
- # chapters.append(current_chapter)
- # elif current_chapter is not None:
- # current_chapter['content'].append(para.text) # 娣诲姞鍐呭鍒板綋鍓嶇珷鑺�
-
- # return chapters
-
- def has_image(self, paragraph):
- # 閫氳繃妫�鏌ML涓殑宓屽叆寮忓璞℃潵鍒ゆ柇鏄惁鏈夊浘鐗�
- xml = paragraph._element.xml
- return 'w:object' in xml or 'w:drawing' in xml
-
- def convert_blob_to_png_base64(self, image_blob):
- try:
- # 鎵撳紑鍥剧墖
- image = Image.open(BytesIO(image_blob))
- # 鍒涘缓鍐呭瓨缂撳啿鍖�
- buffer = BytesIO()
- # 淇濆瓨涓篜NG鏍煎紡
- image.save(buffer, format="PNG")
- # 鑾峰彇PNG鏍煎紡鐨勪簩杩涘埗鏁版嵁
- png_data = buffer.getvalue()
- # 杞崲涓築ase64缂栫爜
- base64_data = base64.b64encode(png_data).decode('utf-8')
- return base64_data
- except Exception as e:
- print(f"Error: {e}")
- return None
-
- def get_image_blob(self, paragraph):
- # 閬嶅巻娈佃惤涓殑鎵�鏈塕un瀵硅薄锛堝浘鐗囬�氬父鍦ㄥ崟鐙殑Run涓級
- for run in paragraph.runs:
- xml = run._element.xml
- if xml.find('v:imagedata') != -1:
- # 浣跨敤姝ｅ垯琛ㄨ揪寮忔煡鎵緍:id灞炴��
- match = re.search(r'r:id="([^"]+)"', xml)
- if match:
- r_id = match.group(1)
- if r_id:
- # 鑾峰彇鍥剧墖淇℃伅
- image_part = paragraph.part.rels[r_id].target_part
- return image_part.blob
- if xml.find('wp:inline') != -1 or xml.find('wp:anchor') != -1:
- # 浣跨敤姝ｅ垯琛ㄨ揪寮忔煡鎵緍:embed灞炴��
- match = re.search(r'r:embed="([^"]+)"', xml)
- if match:
- r_id = match.group(1)
- if r_id:
- # 鑾峰彇鍥剧墖淇℃伅
- image_part = paragraph.part.rels[r_id].target_part
- return image_part.blob
- return None
-
- def loadDoc(self):
- doc = Document('./static/doc/ZL鏍煎紡(鍏紑).docx')
- # 鎸夌収鏍囬鑾峰彇娈佃惤鐨勫眰绾х粨鏋�
- titles = []
- for paragraph in doc.paragraphs:
- if paragraph.text != "":
- # 鏂囧瓧涓嶄负绌�
- if paragraph.style.base_style is not None:
- # 鏈塨ase_style
- if paragraph.style.base_style.name.startswith('Heading'):
- # 鏄爣棰�
- level = int(paragraph.style.base_style.name.split(' ')[-1])
- obj = {}
- obj["level"] = level
- obj["text"] = paragraph.text
- obj["child"] = []
- titles.append(obj)
- else:
- length = len(titles)
- if "child" in titles[length -1]:
- obj = {}
- obj["text"] = paragraph.text
- titles[length -1]['child'].append(obj)
- else:
- # 娌℃湁base_style
- length = len(titles)
- obj = {}
- obj["text"] = paragraph.text
- if length > 0 and "child" in titles[length -1]:
- # 濡傛灉鏄爣棰樺唴鐨刟ppend杩涙爣棰樼殑child
- titles[length -1]['child'].append(obj)
- else:
- # 闈炴爣棰樺唴鐨勭洿鎺ユ斁鍦ㄧ涓�灞�
- titles.append(obj)
- else:
- # 鏂囧瓧涓虹┖鏃讹紝鍙兘鏄浘鐗囨垨鑰呰〃鏍�
- if self.has_image(paragraph):
- # 褰撳墠娈佃惤涓哄浘鐗�
- obj = {}
- # 鑾峰彇鍥剧墖鐨刡lob
- img = self.get_image_blob(paragraph)
- if img is not None:
- imgBase64 = self.convert_blob_to_png_base64(img)
- if imgBase64 is not None:
- obj["imgBase64"] = imgBase64
- titles[length -1]['child'].append(obj)
- # 鍦ㄨ繖閲屾墿灞曞垽鏂〃鏍�
- print(titles)
- # for para in doc.paragraphs:
- # print(para.text)
- # print('------------------------')
-
if __name__ == '__main__':
- vision = VisionTest("./static/images/test.png")
- # vision.run("闂")
- vision.loadDoc()
+ vision = VisionTest("image_path")
+ vision.run("闂")
--
Gitblit v1.9.1