Use Yolox-table-structure to extract tables as markdown

[edknv]

Description

This PR introduces yolox-table-structure as an optional extraction mode. This new NIM allows us to extract tables in markdown format by detecting the bounding boxes of each table cell.

Usage

Since the new NIM is not yet public, users must manually specify the image name and tag in the .env file. Additionally, the service must be started using docker compose with a profile, e.g.,

docker compose --profile yolox-table-structure up --build

To enable this extraction mode in the client, users must explicitly set paddle_output_format="markdown", e.g.,

ingestor = (
    Ingestor(
        message_client_hostname="nv-ingest-ms-runtime",
        message_client_kwargs={"max_retries": 10},
    )
    .files(files)
    .extract(
        extract_text=True,
        extract_tables=True,
        extract_charts=True,
        extract_images=False,
        text_depth="page",
        paddle_output_format="markdown",  #default is "pseudo_markdown" until table structure NIM release 
    )

Notes

• The default paddle_output_format remains "pseudo_markdown" until the table structure NIM is officially released.
• Ensure the .env file contains the correct image name and tag before running the extraction.

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[edknv]

With a443502, users must manually specify the image name and tag in the .env file. Additionally, the service must be started using docker compose with a profile, e.g.,

docker compose --profile yolox-table-structure up --build

To enable this extraction mode in the client, users must explicitly set paddle_output_format="markdown", e.g.,

ingestor = (
    Ingestor(
        message_client_hostname="nv-ingest-ms-runtime",
        message_client_kwargs={"max_retries": 10},
    )
    .files(files)
    .extract(
        extract_text=True,
        extract_tables=True,
        extract_charts=True,
        extract_images=False,
        text_depth="page",
        paddle_output_format="markdown",  #default is "pseudo_markdown" until table structure NIM release 
    )

[drobison00]

Do we want this to be a full failure of the job? It indicates a lack of agreement between one yolox model and another, but its not clear that it should cause document processing to fail.

[edknv]

How about something like 34d7ce1?