compass.web.file_loader.AsyncDoclingWebFileLoader#

class AsyncDoclingWebFileLoader(header_template=None, verify_ssl=True, aget_kwargs=None, pw_launch_kwargs=None, html_read_kwargs=None, html_read_coroutine=None, file_cache_coroutine=None, browser_semaphore=None, use_scrapling_stealth=False, num_pw_html_retries=3, to_md_kwargs=None, pytesseract_exe_fp=None, pdf_pipeline_options=None, re_fetch_failed_with_elm=False, **extra)[source]#

Bases: BaseAsyncFileLoader

Async web file loader using Docling

Parameters:

  • header_template (dict, optional) – Optional GET header template. If not specified, uses DEFAULT_HEADERS. By default, None.

  • verify_ssl (bool, optional) – Option to use aiohttp’s default SSL check. If False, SSL certificate validation is skipped. By default, True.

  • aget_kwargs (dict, optional) – Other kwargs to pass to aiohttp.ClientSession.get(). By default, None.

  • pw_launch_kwargs (dict, optional) – Keyword-value argument pairs to pass to async_playwright.chromium.launch (only used when reading HTML). By default, None.

  • html_read_kwargs (dict, optional) – Keyword-value argument pairs to pass to the html_read_coroutine. By default, None.

  • html_read_coroutine (callable(), optional) – HTML file read coroutine. Must by an async function. Should accept HTML text as the first argument and kwargs as the rest. Must return a elm.web.document.HTMLDocument. If None, a default function that runs in the main thread is used. By default, None.

  • file_cache_coroutine (callable(), optional) – File caching coroutine. Can be used to cache files downloaded by this class. Must accept an BaseDocument instance as the first argument and the file content to be written as the second argument. If this method is not provided, no document caching is performed. By default, None.

  • browser_semaphore (asyncio.Semaphore, optional) – Semaphore instance that can be used to limit the number of playwright browsers open concurrently. If None, no limits are applied. By default, None.

  • use_scrapling_stealth (bool, default False) – Option to use scrapling stealth scripts instead of playwright-stealth. By default, False.

  • num_pw_html_retries (int, default 3) – Number of attempts to load HTML content. This is useful because the playwright parameters are stochastic, and sometimes a combination of them can fail to load HTML. The default value is likely a good balance between processing attempts and retrieval success. Note that the minimum number of attempts will always be 2, even if the user provides a value smaller than this. By default, 3.

  • to_md_kwargs (dict, optional) – Keyword-value argument pairs to pass to to Docling’s export_to_markdown() method for converting the raw content to a markdown document. Can be useful to specify image placeholders (i.e. "image_placeholder"="") or page break placeholders (i.e. "page_break_placeholder"="<!-- page break -->"). By default, ``None.

  • pytesseract_exe_fp (path-like, optional) – Path to the pytesseract executable. If specified, OCR will be used to extract text from scanned PDFs using Google’s Tesseract. By default None.

  • pdf_pipeline_options (dict, optional) – Dictionary of keyword-value arguments to pass to docling.datamodel.pipeline_options.PdfPipelineOptions initializer. Table structure defaults to enabled when not explicitly specified. OCR defaults to enabled only when a Tesseract executable path is provided. If None, the default options are used.

  • re_fetch_failed_with_elm (bool, default False) –

    Option to re-fetch failed sources using ELM’s default fetcher. This can be useful if Docling fails to parse a document, but ELM’s fetcher can still retrieve it. To make sure this functions properly, be sure to specify pdf_read_kwargs and pdf_read_coroutine, in the extra kwargs as you would for the elm-based AsyncWebFileLoader.

    Note

    This is meant to be a _fast_ fallback option for the longer Docling parse, so OCR PDF parsing is completely disabled for the ELM fallback.

    By default, False.

Methods

fetch(source)

Fetch a document for the given source.

fetch_all(*sources)

Fetch documents for all requested sources.
async fetch_all(*sources)[source]#

Fetch documents for all requested sources.

Parameters:

*sources – Iterable of sources (as strings) used to fetch the documents.

Returns:

list – List of parsed documents.

async fetch(source)#

Fetch a document for the given source.

Parameters:

source (str) – Source used to load the document.

Returns:

elm.web.document.Document – Document instance containing text, if the load was successful.