Spaces:
Running
Running
Download plugins/_document_query/helpers/parsers/unstructured.py from Leon4gr45/openoperator: direct link, hf CLI and curl.
- Browser
- Download file 769 Bytes
-
https://huggingface.co/spaces/Leon4gr45/openoperator/resolve/main/plugins/_document_query/helpers/parsers/unstructured.py
- Command line
-
hf download hf://spaces/Leon4gr45/openoperator/plugins/_document_query/helpers/parsers/unstructured.py
-
curl -L -o unstructured.py https://huggingface.co/spaces/Leon4gr45/openoperator/resolve/main/plugins/_document_query/helpers/parsers/unstructured.py
769 Bytes
| """Catch-all parser using UnstructuredLoader.""" | |
| import os | |
| from plugins._document_query.helpers.fetch import FetchedDocument | |
| from .base import BaseParser | |
| os.environ.setdefault("USER_AGENT", "@mixedbread-ai/unstructured") | |
| class UnstructuredParser(BaseParser): | |
| mimetypes = ["*"] | |
| def _parse_sync(self, document: FetchedDocument, config: dict) -> str: | |
| from langchain_unstructured import UnstructuredLoader | |
| with document.local_file() as file_path: | |
| loader = UnstructuredLoader( | |
| file_path=file_path, | |
| mode="single", | |
| partition_via_api=False, | |
| strategy="hi_res", | |
| ) | |
| elements = loader.load() | |
| return "\n".join(e.page_content for e in elements) | |