run-llama/llama_cloud_services
1
0
Fork 0
mirror of https://github.com/run-llama/llama_cloud_services.git synced 2026-07-18 15:54:38 -04:00
Code Issues 352 Packages Projects Releases 10 Wiki Activity

Same PDF was parsed correctly on September 5, but not longer on September 9 #544

New Issue
Open
opened 2026-02-16 00:18:12 -05:00 by yindo · 7 comments
yindo commented 2026-02-16 00:18:12 -05:00
Owner
Copy Link
Copy Source

Originally created by @raphaelh on GitHub (Sep 26, 2025).

Describe the bug
On September 5, 2025 at 11:31PM a PDF file was parsed correctly (Job ID : 5727f30f-a6dd-44b6-8a46-dc71c0ecc47a)

On September 9, 2025 at 9:31AM the same PDF file was no longer parsed correctly (Job ID : 2d6eb138-35d3-4095-a2a6-0aee1d7bcfe7)

The problem is in the first table :

  • some cell values are missing
  • the following text is showing up in the middle of the table :

Files
Cannot share it here

Client:
Please remove untested options:

  • API (also tested with Frontend (cloud.llamaindex.ai))

Additional context
My configuration:

    parseMode: 'parse_page_without_llm',
    disableOcr: true,
    disableImageExtraction: true,
    adaptiveLongTable: true,
    outlinedTableExtraction: true,
    outputTablesAsHTML: false,
    skipDiagonalText: true,

I also tried with "parse_page_with_llm" but same problem

Originally created by @raphaelh on GitHub (Sep 26, 2025). **Describe the bug** On September 5, 2025 at 11:31PM a PDF file was parsed correctly (Job ID : 5727f30f-a6dd-44b6-8a46-dc71c0ecc47a) On September 9, 2025 at 9:31AM the **same PDF file** was no longer parsed correctly (Job ID : 2d6eb138-35d3-4095-a2a6-0aee1d7bcfe7) The problem is in the first table : - some cell values are missing - the following text is showing up in the middle of the table : > | | | | | | | | | | > | - | - | - | - | - | - | - | - | - | > | | | | | | | | | | > | | | | | | | | | | > | | | | | | | | | | > | | | | | | | | | | > | | | | | | | | | | **Files** Cannot share it here **Client:** Please remove untested options: - API (also tested with Frontend (cloud.llamaindex.ai)) **Additional context** My configuration: parseMode: 'parse_page_without_llm', disableOcr: true, disableImageExtraction: true, adaptiveLongTable: true, outlinedTableExtraction: true, outputTablesAsHTML: false, skipDiagonalText: true, I also tried with "parse_page_with_llm" but same problem
yindo added the bug label 2026-02-16 00:18:12 -05:00
yindo commented 2026-02-16 00:18:12 -05:00
Author
Owner
Copy Link
Copy Source

@raphaelh commented on GitHub (Sep 28, 2025):

Hi LlamaTeam :)

Did you have chance to have a look?

Perfectly happy about LlamaParse otherwise, but this regression is problematic

Regards

@raphaelh commented on GitHub (Sep 28, 2025): Hi LlamaTeam :) Did you have chance to have a look? Perfectly happy about LlamaParse otherwise, but this regression is problematic Regards
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@Jpzinn654 commented on GitHub (Sep 29, 2025):

Hi @raphaelh

I took a look at your issue. It’s a shame that the function stopped working. I went through the documentation and noticed there might be other solutions such as "Parse with Layout Agent", "Parse with LLM", or for more complex tables, "Parse Document with LLM" or "Parse Document with Agent". It might be worth exploring these options, as they could help with your process.

Also, take a look at the documentation here it might give you some more insights:

https://developers.llamaindex.ai/typescript/cloud/llamaparse/presets_and_modes/advance_parsing_modes/#_top

@Jpzinn654 commented on GitHub (Sep 29, 2025): Hi @raphaelh I took a look at your issue. It’s a shame that the function stopped working. I went through the documentation and noticed there might be other solutions such as **"Parse with Layout Agent", "Parse with LLM"**, or for more complex tables, **"Parse Document with LLM" or "Parse Document with Agent"**. It might be worth exploring these options, as they could help with your process. Also, take a look at the documentation here it might give you some more insights: https://developers.llamaindex.ai/typescript/cloud/llamaparse/presets_and_modes/advance_parsing_modes/#_top
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@raphaelh commented on GitHub (Sep 29, 2025):

Hi @Jpzinn654

thanks for your reply

I also tried in the Parse Playgroud :

  • "parse_page_with_agent" setting : Job ID 6b632397-22e9-4fde-ba90-dfa20f65a63c -> same result
  • "parse_page_with_layout_agent" setting : Job ID efd792fc-772b-49ea-898f-1ace62f38813 -> same result

I use the text output

The main problem for me is that I've developed a parsing pipeline base on LlamaParse output which stopped working

@raphaelh commented on GitHub (Sep 29, 2025): Hi @Jpzinn654 thanks for your reply I also tried in the Parse Playgroud : - "parse_page_with_agent" setting : Job ID 6b632397-22e9-4fde-ba90-dfa20f65a63c -> same result - "parse_page_with_layout_agent" setting : Job ID efd792fc-772b-49ea-898f-1ace62f38813 -> same result I use the text output The main problem for me is that I've developed a parsing pipeline base on LlamaParse output which stopped working
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@Jpzinn654 commented on GitHub (Sep 29, 2025):

@raphaelh

Thanks for the update and for sharing the Job IDs. I understand the issue better now. It looks like this is indeed a regression affecting table parsing, specifically for the first table in your PDF. Since you’ve already tried multiple parsing modes (parse_page_with_agent, parse_page_with_layout_agent, and parse_page_with_llm) and all returned the same result, it’s likely related to how the internal table extraction logic handles certain table structures when disableOcr and disableImageExtraction are true.

A few suggestions for next steps:

  1. Try enabling OCR temporarily
    Even if your PDF is text-based, some internal table parsing relies on OCR to detect precise cell boundaries. Set disableOcr: false and see if the missing cells appear.
  2. Test without adaptive table extraction
    The adaptiveLongTable and outlinedTableExtraction settings might interfere with certain table formats. Try adaptiveLongTable: false and outlinedTableExtraction: false to check if the table is parsed correctly.
  3. Use “Parse Document with LLM” for complex tables
    If the table has merged cells, diagonal text, or complex formatting, the page-level parsing modes might fail. Document-level parsing with LLM or Agent often handles these structures more reliably.

I know this might require some testing, but these steps usually reveal whether the issue is configuration-related or a true regression in the core parsing engine.

Well, those were my last thoughts. I hope I helped in some way.

@Jpzinn654 commented on GitHub (Sep 29, 2025): @raphaelh Thanks for the update and for sharing the Job IDs. I understand the issue better now. It looks like this is indeed a regression affecting table parsing, specifically for the first table in your PDF. Since you’ve already tried multiple parsing modes (parse_page_with_agent, parse_page_with_layout_agent, and parse_page_with_llm) and all returned the same result, it’s likely related to how the internal table extraction logic handles certain table structures when disableOcr and disableImageExtraction are true. A few suggestions for next steps: 1. Try enabling OCR temporarily Even if your PDF is text-based, some internal table parsing relies on OCR to detect precise cell boundaries. Set disableOcr: false and see if the missing cells appear. 2. Test without adaptive table extraction The adaptiveLongTable and outlinedTableExtraction settings might interfere with certain table formats. Try adaptiveLongTable: false and outlinedTableExtraction: false to check if the table is parsed correctly. 3. Use “Parse Document with LLM” for complex tables If the table has merged cells, diagonal text, or complex formatting, the page-level parsing modes might fail. Document-level parsing with LLM or Agent often handles these structures more reliably. I know this might require some testing, but these steps usually reveal whether the issue is configuration-related or a true regression in the core parsing engine. Well, those were my last thoughts. I hope I helped in some way.
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@raphaelh commented on GitHub (Sep 29, 2025):

@Jpzinn654 Thanks

still on parse_mode="parse_page_without_llm"

  • same problem with outlined_table_extraction=true (default ?) -> same problem, Job ID c173eaa4-30bb-4051-8a73-eae56a97f657

  • outlined_table_extraction=false -> WORKS !!, Job ID efe82862-3d33-4f50-8a5c-ec252651a858

So it seems the regression is on outlined table extraction :)

@raphaelh commented on GitHub (Sep 29, 2025): @Jpzinn654 Thanks still on `parse_mode="parse_page_without_llm"` - same problem with `outlined_table_extraction=true` (default ?) -> same problem, Job ID c173eaa4-30bb-4051-8a73-eae56a97f657 - `outlined_table_extraction=false` -> **WORKS !!**, Job ID efe82862-3d33-4f50-8a5c-ec252651a858 So it seems the regression is on outlined table extraction :)
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@raphaelh commented on GitHub (Oct 5, 2025):

Hi LLama team,

in case of a bug like the one I'm reporting, are we supposed to file a bug here ?

Or is there any other way to contact Llama support ?

@raphaelh commented on GitHub (Oct 5, 2025): Hi LLama team, in case of a bug like the one I'm reporting, are we supposed to file a bug here ? Or is there any other way to contact Llama support ?
yindo commented 2026-02-16 00:18:13 -05:00
Author
Owner
Copy Link
Copy Source

@raphaelh commented on GitHub (Dec 5, 2025):

?

@raphaelh commented on GitHub (Dec 5, 2025): ?
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
LlamaCloud
LlamaExtract
LlamaParse
bug
client
complex layout
documentation
duplicate
enhancement
figure
font
header footer
help wanted
ocr
parser
pdf_debug
pdf_debug
pull-request
Mirrored from GitHub Pull Request
 python-lib
question
server_issue
server_issue
ui
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: run-llama/llama_cloud_services#544
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?