feat: upgrade langchain (#430)

Co-authored-by: jyong <718720800@qq.com>

api/core/data_loader/file_extractor.py

@@ -0,0 +1,43 @@
+import tempfile
+from pathlib import Path
+from typing import List, Union
+
+from langchain.document_loaders import TextLoader, Docx2txtLoader
+from langchain.schema import Document
+
+from core.data_loader.loader.csv import CSVLoader
+from core.data_loader.loader.excel import ExcelLoader
+from core.data_loader.loader.html import HTMLLoader
+from core.data_loader.loader.markdown import MarkdownLoader
+from core.data_loader.loader.pdf import PdfLoader
+from extensions.ext_storage import storage
+from models.model import UploadFile
+
+
+class FileExtractor:
+    @classmethod
+    def load(cls, upload_file: UploadFile, return_text: bool = False) -> Union[List[Document] | str]:
+        with tempfile.TemporaryDirectory() as temp_dir:
+            suffix = Path(upload_file.key).suffix
+            file_path = f"{temp_dir}/{next(tempfile._get_candidate_names())}{suffix}"
+            storage.download(upload_file.key, file_path)
+
+            input_file = Path(file_path)
+            delimiter = '\n'
+            if input_file.suffix == '.xlsx':
+                loader = ExcelLoader(file_path)
+            elif input_file.suffix == '.pdf':
+                loader = PdfLoader(file_path, upload_file=upload_file)
+            elif input_file.suffix in ['.md', '.markdown']:
+                loader = MarkdownLoader(file_path, autodetect_encoding=True)
+            elif input_file.suffix in ['.htm', '.html']:
+                loader = HTMLLoader(file_path)
+            elif input_file.suffix == '.docx':
+                loader = Docx2txtLoader(file_path)
+            elif input_file.suffix == '.csv':
+                loader = CSVLoader(file_path, autodetect_encoding=True)
+            else:
+                # txt
+                loader = TextLoader(file_path, autodetect_encoding=True)
+
+            return delimiter.join([document.page_content for document in loader.load()]) if return_text else loader.load()

api/core/data_loader/loader/csv.py

@@ -0,0 +1,67 @@
+import logging
+from typing import Optional, Dict, List
+
+from langchain.document_loaders import CSVLoader as LCCSVLoader
+from langchain.document_loaders.helpers import detect_file_encodings
+
+from models.dataset import Document
+
+logger = logging.getLogger(__name__)
+
+
+class CSVLoader(LCCSVLoader):
+    def __init__(
+            self,
+            file_path: str,
+            source_column: Optional[str] = None,
+            csv_args: Optional[Dict] = None,
+            encoding: Optional[str] = None,
+            autodetect_encoding: bool = True,
+    ):
+        self.file_path = file_path
+        self.source_column = source_column
+        self.encoding = encoding
+        self.csv_args = csv_args or {}
+        self.autodetect_encoding = autodetect_encoding
+
+    def load(self) -> List[Document]:
+        """Load data into document objects."""
+        try:
+            with open(self.file_path, newline="", encoding=self.encoding) as csvfile:
+                docs = self._read_from_file(csvfile)
+        except UnicodeDecodeError as e:
+            if self.autodetect_encoding:
+                detected_encodings = detect_file_encodings(self.file_path)
+                for encoding in detected_encodings:
+                    logger.debug("Trying encoding: ", encoding.encoding)
+                    try:
+                        with open(self.file_path, newline="", encoding=encoding.encoding) as csvfile:
+                            docs = self._read_from_file(csvfile)
+                        break
+                    except UnicodeDecodeError:
+                        continue
+            else:
+                raise RuntimeError(f"Error loading {self.file_path}") from e
+
+        return docs
+
+    def _read_from_file(self, csvfile):
+        docs = []
+        csv_reader = csv.DictReader(csvfile, **self.csv_args)  # type: ignore
+        for i, row in enumerate(csv_reader):
+            content = "\n".join(f"{k.strip()}: {v.strip()}" for k, v in row.items())
+            try:
+                source = (
+                    row[self.source_column]
+                    if self.source_column is not None
+                    else ''
+                )
+            except KeyError:
+                raise ValueError(
+                    f"Source column '{self.source_column}' not found in CSV file."
+                )
+            metadata = {"source": source, "row": i}
+            doc = Document(page_content=content, metadata=metadata)
+            docs.append(doc)
+
+        return docs

api/core/data_loader/loader/excel.py

@@ -0,0 +1,43 @@
+import json
+import logging
+from typing import List
+
+from langchain.document_loaders.base import BaseLoader
+from langchain.schema import Document
+from openpyxl.reader.excel import load_workbook
+
+logger = logging.getLogger(__name__)
+
+
+class ExcelLoader(BaseLoader):
+    """Load xlxs files.
+
+
+    Args:
+        file_path: Path to the file to load.
+    """
+
+    def __init__(
+        self,
+        file_path: str
+    ):
+        """Initialize with file path."""
+        self._file_path = file_path
+
+    def load(self) -> List[Document]:
+        data = []
+        keys = []
+        wb = load_workbook(filename=self._file_path, read_only=True)
+        # loop over all sheets
+        for sheet in wb:
+            for row in sheet.iter_rows(values_only=True):
+                if all(v is None for v in row):
+                    continue
+                if keys == []:
+                    keys = list(map(str, row))
+                else:
+                    row_dict = dict(zip(keys, row))
+                    row_dict = {k: v for k, v in row_dict.items() if v}
+                    data.append(json.dumps(row_dict, ensure_ascii=False))
+
+        return [Document(page_content='\n\n'.join(data))]

api/core/data_loader/loader/html.py

@@ -0,0 +1,35 @@
+import logging
+from typing import List
+
+from bs4 import BeautifulSoup
+from langchain.document_loaders.base import BaseLoader
+from langchain.schema import Document
+
+logger = logging.getLogger(__name__)
+
+
+class HTMLLoader(BaseLoader):
+    """Load html files.
+
+
+    Args:
+        file_path: Path to the file to load.
+    """
+
+    def __init__(
+        self,
+        file_path: str
+    ):
+        """Initialize with file path."""
+        self._file_path = file_path
+
+    def load(self) -> List[Document]:
+        return [Document(page_content=self._load_as_text())]
+
+    def _load_as_text(self) -> str:
+        with open(self._file_path, "rb") as fp:
+            soup = BeautifulSoup(fp, 'html.parser')
+            text = soup.get_text()
+            text = text.strip() if text else ''
+
+        return text

api/core/data_loader/loader/markdown.py

@@ -0,0 +1,134 @@
+import logging
+import re
+from typing import Optional, List, Tuple, cast
+
+from langchain.document_loaders.base import BaseLoader
+from langchain.document_loaders.helpers import detect_file_encodings
+from langchain.schema import Document
+
+logger = logging.getLogger(__name__)
+
+
+class MarkdownLoader(BaseLoader):
+    """Load md files.
+
+
+    Args:
+        file_path: Path to the file to load.
+
+        remove_hyperlinks: Whether to remove hyperlinks from the text.
+
+        remove_images: Whether to remove images from the text.
+
+        encoding: File encoding to use. If `None`, the file will be loaded
+        with the default system encoding.
+
+        autodetect_encoding: Whether to try to autodetect the file encoding
+            if the specified encoding fails.
+    """
+
+    def __init__(
+        self,
+        file_path: str,
+        remove_hyperlinks: bool = True,
+        remove_images: bool = True,
+        encoding: Optional[str] = None,
+        autodetect_encoding: bool = True,
+    ):
+        """Initialize with file path."""
+        self._file_path = file_path
+        self._remove_hyperlinks = remove_hyperlinks
+        self._remove_images = remove_images
+        self._encoding = encoding
+        self._autodetect_encoding = autodetect_encoding
+
+    def load(self) -> List[Document]:
+        tups = self.parse_tups(self._file_path)
+        documents = []
+        for header, value in tups:
+            value = value.strip()
+            if header is None:
+                documents.append(Document(page_content=value))
+            else:
+                documents.append(Document(page_content=f"\n\n{header}\n{value}"))
+
+        return documents
+
+    def markdown_to_tups(self, markdown_text: str) -> List[Tuple[Optional[str], str]]:
+        """Convert a markdown file to a dictionary.
+
+        The keys are the headers and the values are the text under each header.
+
+        """
+        markdown_tups: List[Tuple[Optional[str], str]] = []
+        lines = markdown_text.split("\n")
+
+        current_header = None
+        current_text = ""
+
+        for line in lines:
+            header_match = re.match(r"^#+\s", line)
+            if header_match:
+                if current_header is not None:
+                    markdown_tups.append((current_header, current_text))
+
+                current_header = line
+                current_text = ""
+            else:
+                current_text += line + "\n"
+        markdown_tups.append((current_header, current_text))
+
+        if current_header is not None:
+            # pass linting, assert keys are defined
+            markdown_tups = [
+                (re.sub(r"#", "", cast(str, key)).strip(), re.sub(r"<.*?>", "", value))
+                for key, value in markdown_tups
+            ]
+        else:
+            markdown_tups = [
+                (key, re.sub("\n", "", value)) for key, value in markdown_tups
+            ]
+
+        return markdown_tups
+
+    def remove_images(self, content: str) -> str:
+        """Get a dictionary of a markdown file from its path."""
+        pattern = r"!{1}\[\[(.*)\]\]"
+        content = re.sub(pattern, "", content)
+        return content
+
+    def remove_hyperlinks(self, content: str) -> str:
+        """Get a dictionary of a markdown file from its path."""
+        pattern = r"\[(.*?)\]\((.*?)\)"
+        content = re.sub(pattern, r"\1", content)
+        return content
+
+    def parse_tups(self, filepath: str) -> List[Tuple[Optional[str], str]]:
+        """Parse file into tuples."""
+        content = ""
+        try:
+            with open(filepath, "r", encoding=self._encoding) as f:
+                content = f.read()
+        except UnicodeDecodeError as e:
+            if self._autodetect_encoding:
+                detected_encodings = detect_file_encodings(filepath)
+                for encoding in detected_encodings:
+                    logger.debug("Trying encoding: ", encoding.encoding)
+                    try:
+                        with open(filepath, encoding=encoding.encoding) as f:
+                            content = f.read()
+                        break
+                    except UnicodeDecodeError:
+                        continue
+            else:
+                raise RuntimeError(f"Error loading {filepath}") from e
+        except Exception as e:
+            raise RuntimeError(f"Error loading {filepath}") from e
+
+        if self._remove_hyperlinks:
+            content = self.remove_hyperlinks(content)
+
+        if self._remove_images:
+            content = self.remove_images(content)
+
+        return self.markdown_to_tups(content)

api/core/data_loader/loader/notion.py

@@ -0,0 +1,369 @@
+import json
+import logging
+from typing import List, Dict, Any, Optional
+
+import requests
+from flask import current_app
+from langchain.document_loaders.base import BaseLoader
+from langchain.schema import Document
+
+from extensions.ext_database import db
+from models.dataset import Document as DocumentModel
+from models.source import DataSourceBinding
+
+logger = logging.getLogger(__name__)
+
+BLOCK_CHILD_URL_TMPL = "https://api.notion.com/v1/blocks/{block_id}/children"
+DATABASE_URL_TMPL = "https://api.notion.com/v1/databases/{database_id}/query"
+SEARCH_URL = "https://api.notion.com/v1/search"
+RETRIEVE_PAGE_URL_TMPL = "https://api.notion.com/v1/pages/{page_id}"
+RETRIEVE_DATABASE_URL_TMPL = "https://api.notion.com/v1/databases/{database_id}"
+HEADING_TYPE = ['heading_1', 'heading_2', 'heading_3']
+
+
+class NotionLoader(BaseLoader):
+    def __init__(
+            self,
+            notion_access_token: str,
+            notion_workspace_id: str,
+            notion_obj_id: str,
+            notion_page_type: str,
+            document_model: Optional[DocumentModel] = None
+    ):
+        self._document_model = document_model
+        self._notion_workspace_id = notion_workspace_id
+        self._notion_obj_id = notion_obj_id
+        self._notion_page_type = notion_page_type
+        self._notion_access_token = notion_access_token
+
+        if not self._notion_access_token:
+            integration_token = current_app.config.get('NOTION_INTEGRATION_TOKEN')
+            if integration_token is None:
+                raise ValueError(
+                    "Must specify `integration_token` or set environment "
+                    "variable `NOTION_INTEGRATION_TOKEN`."
+                )
+
+            self._notion_access_token = integration_token
+
+    @classmethod
+    def from_document(cls, document_model: DocumentModel):
+        data_source_info = document_model.data_source_info_dict
+        if not data_source_info or 'notion_page_id' not in data_source_info \
+                or 'notion_workspace_id' not in data_source_info:
+            raise ValueError("no notion page found")
+
+        notion_workspace_id = data_source_info['notion_workspace_id']
+        notion_obj_id = data_source_info['notion_page_id']
+        notion_page_type = data_source_info['type']
+        notion_access_token = cls._get_access_token(document_model.tenant_id, notion_workspace_id)
+
+        return cls(
+            notion_access_token=notion_access_token,
+            notion_workspace_id=notion_workspace_id,
+            notion_obj_id=notion_obj_id,
+            notion_page_type=notion_page_type,
+            document_model=document_model
+        )
+
+    def load(self) -> List[Document]:
+        self.update_last_edited_time(
+            self._document_model
+        )
+
+        text_docs = self._load_data_as_documents(self._notion_obj_id, self._notion_page_type)
+
+        return text_docs
+
+    def _load_data_as_documents(
+            self, notion_obj_id: str, notion_page_type: str
+    ) -> List[Document]:
+        docs = []
+        if notion_page_type == 'database':
+            # get all the pages in the database
+            page_text = self._get_notion_database_data(notion_obj_id)
+            docs.append(Document(page_content=page_text))
+        elif notion_page_type == 'page':
+            page_text_list = self._get_notion_block_data(notion_obj_id)
+            for page_text in page_text_list:
+                docs.append(Document(page_content=page_text))
+        else:
+            raise ValueError("notion page type not supported")
+
+        return docs
+
+    def _get_notion_database_data(
+            self, database_id: str, query_dict: Dict[str, Any] = {}
+    ) -> str:
+        """Get all the pages from a Notion database."""
+        res = requests.post(
+            DATABASE_URL_TMPL.format(database_id=database_id),
+            headers={
+                "Authorization": "Bearer " + self._notion_access_token,
+                "Content-Type": "application/json",
+                "Notion-Version": "2022-06-28",
+            },
+            json=query_dict,
+        )
+
+        data = res.json()
+
+        database_content_list = []
+        if 'results' not in data or data["results"] is None:
+            return ""
+        for result in data["results"]:
+            properties = result['properties']
+            data = {}
+            for property_name, property_value in properties.items():
+                type = property_value['type']
+                if type == 'multi_select':
+                    value = []
+                    multi_select_list = property_value[type]
+                    for multi_select in multi_select_list:
+                        value.append(multi_select['name'])
+                elif type == 'rich_text' or type == 'title':
+                    if len(property_value[type]) > 0:
+                        value = property_value[type][0]['plain_text']
+                    else:
+                        value = ''
+                elif type == 'select' or type == 'status':
+                    if property_value[type]:
+                        value = property_value[type]['name']
+                    else:
+                        value = ''
+                else:
+                    value = property_value[type]
+                data[property_name] = value
+            database_content_list.append(json.dumps(data, ensure_ascii=False))
+
+        return "\n\n".join(database_content_list)
+
+    def _get_notion_block_data(self, page_id: str) -> List[str]:
+        result_lines_arr = []
+        cur_block_id = page_id
+        while True:
+            block_url = BLOCK_CHILD_URL_TMPL.format(block_id=cur_block_id)
+            query_dict: Dict[str, Any] = {}
+
+            res = requests.request(
+                "GET",
+                block_url,
+                headers={
+                    "Authorization": "Bearer " + self._notion_access_token,
+                    "Content-Type": "application/json",
+                    "Notion-Version": "2022-06-28",
+                },
+                json=query_dict
+            )
+            data = res.json()
+            # current block's heading
+            heading = ''
+            for result in data["results"]:
+                result_type = result["type"]
+                result_obj = result[result_type]
+                cur_result_text_arr = []
+                if result_type == 'table':
+                    result_block_id = result["id"]
+                    text = self._read_table_rows(result_block_id)
+                    text += "\n\n"
+                    result_lines_arr.append(text)
+                else:
+                    if "rich_text" in result_obj:
+                        for rich_text in result_obj["rich_text"]:
+                            # skip if doesn't have text object
+                            if "text" in rich_text:
+                                text = rich_text["text"]["content"]
+                                cur_result_text_arr.append(text)
+                                if result_type in HEADING_TYPE:
+                                    heading = text
+
+                    result_block_id = result["id"]
+                    has_children = result["has_children"]
+                    block_type = result["type"]
+                    if has_children and block_type != 'child_page':
+                        children_text = self._read_block(
+                            result_block_id, num_tabs=1
+                        )
+                        cur_result_text_arr.append(children_text)
+
+                    cur_result_text = "\n".join(cur_result_text_arr)
+                    cur_result_text += "\n\n"
+                    if result_type in HEADING_TYPE:
+                        result_lines_arr.append(cur_result_text)
+                    else:
+                        result_lines_arr.append(f'{heading}\n{cur_result_text}')
+
+            if data["next_cursor"] is None:
+                break
+            else:
+                cur_block_id = data["next_cursor"]
+        return result_lines_arr
+
+    def _read_block(self, block_id: str, num_tabs: int = 0) -> str:
+        """Read a block."""
+        result_lines_arr = []
+        cur_block_id = block_id
+        while True:
+            block_url = BLOCK_CHILD_URL_TMPL.format(block_id=cur_block_id)
+            query_dict: Dict[str, Any] = {}
+
+            res = requests.request(
+                "GET",
+                block_url,
+                headers={
+                    "Authorization": "Bearer " + self._notion_access_token,
+                    "Content-Type": "application/json",
+                    "Notion-Version": "2022-06-28",
+                },
+                json=query_dict
+            )
+            data = res.json()
+            if 'results' not in data or data["results"] is None:
+                break
+            heading = ''
+            for result in data["results"]:
+                result_type = result["type"]
+                result_obj = result[result_type]
+                cur_result_text_arr = []
+                if result_type == 'table':
+                    result_block_id = result["id"]
+                    text = self._read_table_rows(result_block_id)
+                    result_lines_arr.append(text)
+                else:
+                    if "rich_text" in result_obj:
+                        for rich_text in result_obj["rich_text"]:
+                            # skip if doesn't have text object
+                            if "text" in rich_text:
+                                text = rich_text["text"]["content"]
+                                prefix = "\t" * num_tabs
+                                cur_result_text_arr.append(prefix + text)
+                                if result_type in HEADING_TYPE:
+                                    heading = text
+                    result_block_id = result["id"]
+                    has_children = result["has_children"]
+                    block_type = result["type"]
+                    if has_children and block_type != 'child_page':
+                        children_text = self._read_block(
+                            result_block_id, num_tabs=num_tabs + 1
+                        )
+                        cur_result_text_arr.append(children_text)
+
+                    cur_result_text = "\n".join(cur_result_text_arr)
+                    if result_type in HEADING_TYPE:
+                        result_lines_arr.append(cur_result_text)
+                    else:
+                        result_lines_arr.append(f'{heading}\n{cur_result_text}')
+
+            if data["next_cursor"] is None:
+                break
+            else:
+                cur_block_id = data["next_cursor"]
+
+        result_lines = "\n".join(result_lines_arr)
+        return result_lines
+
+    def _read_table_rows(self, block_id: str) -> str:
+        """Read table rows."""
+        done = False
+        result_lines_arr = []
+        cur_block_id = block_id
+        while not done:
+            block_url = BLOCK_CHILD_URL_TMPL.format(block_id=cur_block_id)
+            query_dict: Dict[str, Any] = {}
+
+            res = requests.request(
+                "GET",
+                block_url,
+                headers={
+                    "Authorization": "Bearer " + self._notion_access_token,
+                    "Content-Type": "application/json",
+                    "Notion-Version": "2022-06-28",
+                },
+                json=query_dict
+            )
+            data = res.json()
+            # get table headers text
+            table_header_cell_texts = []
+            tabel_header_cells = data["results"][0]['table_row']['cells']
+            for tabel_header_cell in tabel_header_cells:
+                if tabel_header_cell:
+                    for table_header_cell_text in tabel_header_cell:
+                        text = table_header_cell_text["text"]["content"]
+                        table_header_cell_texts.append(text)
+            # get table columns text and format
+            results = data["results"]
+            for i in range(len(results) - 1):
+                column_texts = []
+                tabel_column_cells = data["results"][i + 1]['table_row']['cells']
+                for j in range(len(tabel_column_cells)):
+                    if tabel_column_cells[j]:
+                        for table_column_cell_text in tabel_column_cells[j]:
+                            column_text = table_column_cell_text["text"]["content"]
+                            column_texts.append(f'{table_header_cell_texts[j]}:{column_text}')
+
+                cur_result_text = "\n".join(column_texts)
+                result_lines_arr.append(cur_result_text)
+
+            if data["next_cursor"] is None:
+                done = True
+                break
+            else:
+                cur_block_id = data["next_cursor"]
+
+        result_lines = "\n".join(result_lines_arr)
+        return result_lines
+
+    def update_last_edited_time(self, document_model: DocumentModel):
+        if not document_model:
+            return
+
+        last_edited_time = self.get_notion_last_edited_time()
+        data_source_info = document_model.data_source_info_dict
+        data_source_info['last_edited_time'] = last_edited_time
+        update_params = {
+            DocumentModel.data_source_info: json.dumps(data_source_info)
+        }
+
+        DocumentModel.query.filter_by(id=document_model.id).update(update_params)
+        db.session.commit()
+
+    def get_notion_last_edited_time(self) -> str:
+        obj_id = self._notion_obj_id
+        page_type = self._notion_page_type
+        if page_type == 'database':
+            retrieve_page_url = RETRIEVE_DATABASE_URL_TMPL.format(database_id=obj_id)
+        else:
+            retrieve_page_url = RETRIEVE_PAGE_URL_TMPL.format(page_id=obj_id)
+
+        query_dict: Dict[str, Any] = {}
+
+        res = requests.request(
+            "GET",
+            retrieve_page_url,
+            headers={
+                "Authorization": "Bearer " + self._notion_access_token,
+                "Content-Type": "application/json",
+                "Notion-Version": "2022-06-28",
+            },
+            json=query_dict
+        )
+
+        data = res.json()
+        return data["last_edited_time"]
+
+    @classmethod
+    def _get_access_token(cls, tenant_id: str, notion_workspace_id: str) -> str:
+        data_source_binding = DataSourceBinding.query.filter(
+            db.and_(
+                DataSourceBinding.tenant_id == tenant_id,
+                DataSourceBinding.provider == 'notion',
+                DataSourceBinding.disabled == False,
+                DataSourceBinding.source_info['workspace_id'] == f'"{notion_workspace_id}"'
+            )
+        ).first()
+
+        if not data_source_binding:
+            raise Exception(f'No notion data source binding found for tenant {tenant_id} '
+                            f'and notion workspace {notion_workspace_id}')
+
+        return data_source_binding.access_token

api/core/data_loader/loader/pdf.py

@@ -0,0 +1,55 @@
+import logging
+from typing import List, Optional
+
+from langchain.document_loaders import PyPDFium2Loader
+from langchain.document_loaders.base import BaseLoader
+from langchain.schema import Document
+
+from extensions.ext_storage import storage
+from models.model import UploadFile
+
+logger = logging.getLogger(__name__)
+
+
+class PdfLoader(BaseLoader):
+    """Load pdf files.
+
+
+    Args:
+        file_path: Path to the file to load.
+    """
+
+    def __init__(
+        self,
+        file_path: str,
+        upload_file: Optional[UploadFile] = None
+    ):
+        """Initialize with file path."""
+        self._file_path = file_path
+        self._upload_file = upload_file
+
+    def load(self) -> List[Document]:
+        plaintext_file_key = ''
+        plaintext_file_exists = False
+        if self._upload_file:
+            if self._upload_file.hash:
+                plaintext_file_key = 'upload_files/' + self._upload_file.tenant_id + '/' \
+                                     + self._upload_file.hash + '.0625.plaintext'
+                try:
+                    text = storage.load(plaintext_file_key).decode('utf-8')
+                    plaintext_file_exists = True
+                    return [Document(page_content=text)]
+                except FileNotFoundError:
+                    pass
+        documents = PyPDFium2Loader(file_path=self._file_path).load()
+        text_list = []
+        for document in documents:
+            text_list.append(document.page_content)
+        text = "\n\n".join(text_list)
+
+        # save plaintext file for caching
+        if not plaintext_file_exists and plaintext_file_key:
+            storage.save(plaintext_file_key, text.encode('utf-8'))
+
+        return documents
+