[Question]: [ERROR][Exception]: 'text'

[panxiaohuan9]

Self Checks

• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report (Language Policy).
• Non-english title submitions will be closed directly ( 非英文标题的提交将会被直接关闭 ) (Language Policy).
• Please do not modify this template :) and fill in all the required fields.

Describe your problem

I encountered a 'text' error while using the pipeline to parse documents.
dataset——> files——>parse(ingestion pipeline)

18:04:33 Task has been received.

[File]:
18:04:34: File fetched.
18:04:34: Done
Start the pipeline...

[Parser_0]:
18:04:34: Start to work on a PDF.
18:04:44:
18:04:46: OCR finished (12.11s)
18:04:52: Layout analysis (6.75s)
18:04:53: Table analysis (0.62s)
18:04:53: Text merged (0.00s)
18:04:53: Structured (0.25s)
18:04:54: Done

[按标题分割_0]:
18:04:54: Start to merge hierarchically.
18:04:54: Done.
18:04:54: Done

[提取器_摘要]:
18:04:54: Start to generate.
18:04:55: Done

[提取器_关键词]:
18:04:55: Start to generate.
18:04:56: Done

[提取器_问题]:
18:04:57: Start to generate.
18:04:59: Done

[提取器_元数据]:
18:04:59: Start to generate.
18:05:01: Done

[分词器_0]:
18:05:01: Start to tokenize.
18:05:01: Finish tokenizing.
18:05:01: Start embedding inference.
18:05:01: Finish embedding.
18:05:01: Done
18:05:01 [ERROR][Exception]: 'text'

[panxiaohuan9]

[dosubot]

The error [ERROR][Exception]: 'text' during embedding inference means the pipeline expected a text field in the data passed to the embedding step, but it was missing. This usually happens if an upstream component (like the parser or hierarchical merger) produces chunks without the required text field, or if the input format is mismatched—often seen with certain parsers or document types.

Recent fixes in the codebase address cases where the tokenizer mishandled the fields parameter, ensuring it always treats it as a list and correctly extracts text from each chunk for embedding inference. If you’re still seeing this error, check that each chunk passed to the embedding step contains a non-empty text field. Switching from the deepdoc parser to the naive parser has also helped some users avoid this issue, as the deepdoc parser can sometimes produce incomplete data for certain PDFs. For more details, see the relevant code fix and discussion.

If your pipeline uses custom chunkers or extractors, make sure they always output chunks with a valid text field before embedding. If the error persists, reviewing the upstream steps for missing or malformed data is the next step.

To reply, just mention @dosu.

---

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

[Magicbook1108]

@panxiaohuan9
Can you share your document and the ingestion pipeline with me via xiaoyang.bao@infiniflow.org

[panxiaohuan9]

Thank you,The issue has been resolved. It was caused by not configuring the "Select and Reference Variables Output by Upstream Nodes" option.

[panxiaohuan9]

I have another issue: I cannot find where the extracted summaries are stored, and they are not visible in the chunk preview details.