Continued table across pages is split into multiple tables

[bimalsagar-syook]

---

Bug

When processing a PDF that contains a table spanning multiple pages, Docling does not merge the continued table into a single logical table.

If a table starts on one page and continues on the next page, it is split into multiple separate tables instead of being recognized as one continuous table.

This issue occurs in both:

• VLM pipeline
• Standard pipeline

In some cases, a single logical table is split into 2 separate tables.

---

Steps to reproduce

1. Use a PDF that contains a table spanning across two pages.

   • Example: Page 1 contains table header + rows.
   • Page 2 contains the continuation of the same table (remaining rows).

2. Run Docling using either:

   • VLM pipeline, or
   • Standard pipeline.

3. Inspect the extracted tables in the output.

Observed result:
The table is split into multiple independent tables.

Expected result:
The full table (including rows from both pages) should be merged into a single structured table.

---

Docling version

2.68.0

---

Python version

3.12.11

[dosubot]

Docling currently processes tables on a per-page basis in both the Standard and VLM pipelines, so tables that span multiple pages are split into separate tables rather than merged into a single logical table. This is due to how the TableStructureModel and pipeline stages are designed—they detect and predict tables independently for each page, with no built-in logic to link or merge tables across pages. The VLM pipeline also concatenates page-level documents but does not merge tables that continue across pages, treating each page's tables as separate entities (source) (source).

There are no configuration flags or pipeline options in Docling 2.68.0 to enable automatic merging of tables across pages. Some users have worked around this by post-processing the output—manually merging tables that appear to be continuations based on headers, structure, or content similarity. There is also a recent option for merging two tables (see PR #2741), but it is not documented as a general solution for multi-page tables and may require manual intervention (discussion).

For now, if you need a single logical table, you’ll need to merge the split tables in post-processing, for example by checking for repeated headers or matching column structures across pages. This limitation is a known issue, and perfect merging of multi-page tables may require custom logic or external tools (discussion).

To reply, just mention @dosu.

---

^{How did I do? Good | Irrelevant | Incorrect | Verbose | Hallucination | Report 🐛 | Other}