NVIDIA / NVIDIA/NeMo-Retriever
[BUG]: Default ingest pipeline via CLI skips OCR extract_text
Nobody has claimed this yet.
- Dominant language
- Python
- Stars
- 3k
- Forks
- 349
- Avg merge
- 1d 23h
- Merged PRs (30d)
- 116
Description
Version
2026.8.7.dev20260807180251 commit: b3c8a37363f5
Which installation method(s) does this occur on?
Source
Describe the bug.
When retriever ingest default pipeline runs with an image that contains just text it does not go into the lancedb. ocr_kwargs["extract_text"] never is set to True in the default pipeline, so the ingestion pipeline is like "since there are no tables, charts there's nothing in this image to ingest." But there is something to ingest.
On
2026.8.7.dev20260807180251 commit: b3c8a37363f5
Minimum reproducible example
Sample files to test are attached.
retriever ingest scanned_doc.png api_changelog.txt --lancedb-uri /tmp/b --table-name t
output is
Ingested 2 file(s) → 1 row(s) in LanceDB /tmp/b/t.
On the other hand, if you run
retriever ingest --method ocr scanned_doc.png api_changelog.txt --lancedb-uri /tmp/b --table-name t2
output is
Ingested 2 file(s) → 2 row(s) in LanceDB /tmp/b/t2.
Other/Misc.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Reproduce the issue with the provided retriever ingest commands, then inspect the CLI's default pipeline and compare its OCR configuration with the --method ocr path. Done means an image containing only text is ingested alongside the text file and produces 2 rows in LanceDB without explicitly selecting --method ocr.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- python
- Domain
- cli, databases
- Issue type
- Bug
- Difficulty
- 3/5
- Estimated time
- 1-2 days
- Activity status
- Quiet
- Clarity
- Mostly clear
- Newbie friendliness
- 64/100