Langchain-Chatchat

1
from langchain.text_splitter import CharacterTextSplitter
2
import re
3
from typing import List
4

5

6
class AliTextSplitter(CharacterTextSplitter):
7
    def __init__(self, pdf: bool = False, **kwargs):
8
        super().__init__(**kwargs)
9
        self.pdf = pdf
10

11
    def split_text(self, text: str) -> List[str]:
12
        # use_document_segmentation参数指定是否用语义切分文档，此处采取的文档语义分割模型为达摩院开源的nlp_bert_document-segmentation_chinese-base，论文见https://arxiv.org/abs/2107.09278
13
        # 如果使用模型进行文档语义切分，那么需要安装modelscope[nlp]：pip install "modelscope[nlp]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html
14
        # 考虑到使用了三个模型，可能对于低配置gpu不太友好，因此这里将模型load进cpu计算，有需要的话可以替换device为自己的显卡id
15
        if self.pdf:
16
            text = re.sub(r"\n{3,}", r"\n", text)
17
            text = re.sub('\s', " ", text)
18
            text = re.sub("\n\n", "", text)
19
        try:
20
            from modelscope.pipelines import pipeline
21
        except ImportError:
22
            raise ImportError(
23
                "Could not import modelscope python package. "
24
                "Please install modelscope with `pip install modelscope`. "
25
            )
26

27

28
        p = pipeline(
29
            task="document-segmentation",
30
            model='damo/nlp_bert_document-segmentation_chinese-base',
31
            device="cpu")
32
        result = p(documents=text)
33
        sent_list = [i for i in result["text"].split("\n\t") if i]
34
        return sent_list
35