What the data for this use case looks like
Internal office data for state-owned asset management companies mainly comes from state-owned asset supervision reporting systems, internal asset ledgers, employee office documents, network segment operation logs, and similar sources. Update frequencies include real-time (such as approval forms, operation logs), daily (such as attendance data), and monthly/quarterly (such as state-owned asset reports). Document structures include multi-column structured tables (such as asset ledgers, network segment logs), long-text policy documents, and mixed-format approval process documents. Fields cover asset number, book value, IP segment, using department, and similar items. Units include ten thousand yuan, count, days, and similar units.
What constraints do these characteristics impose on document parsing and chunking
The presence of multi-column structured tables requires that parsed cell-associated content is not split arbitrarily. Otherwise, retrieval will fail to match complete field and data relationships. Long-text policy documents have large variations in length. Chunk sizes must adapt to the context window. A chunk size that is too large will exceed model processing limits, while a size that is too small will break contextual logic. Large batches of uploaded document packages and multi-sheet Excel files impose higher requirements on parsing duration and file size limits. Default configurations may not cover parsing needs for all data.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to long texts such as state-owned asset policies and approval forms, retains contextual connections, and avoids overly fine splitting that loses logical coherence |
chunkSplitter | Paragraph + table recognition | Adapts to multi-column structured tables and mixed-format documents, retains the binding relationship between fields and content |
customPdfParse | Enable mineru API | Resolves incomplete recognition of multi-column PDF tables (such as network segment logs) and restores complete column data |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Meets parsing duration requirements for large state-owned asset reports and multi-sheet Excel files, avoids timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Allows uploading batches of state-owned asset ledgers and historical document packages, adapts to batch upload scenarios for internal office work |
chunkSettingMode | Manual configuration | For multi-column structured data, customizes chunking rules to avoid default chunking breaking table associations |
Three common mistakes
- Phenomenon: When uploading multi-column Excel documents (such as network segment logs), only the first two columns of data are recognized, and the remaining column content is missing. Cause:
customPdfParseis not enabled, or parsing rules for multi-column tables are not configured; the default parsing logic only recognizes basic columns. - Phenomenon: After the conditional judgment component processes boolean data parsed by the HTTP component, the parameter becomes empty. Cause: The serialized format of the original data type is not retained during parsing, and the identifier of the boolean value is lost during the chunking process, causing the component to fail to recognize valid parameters.
- Phenomenon: A "Cannot redefine property: toString" error occurs in the production environment. Cause: When uploading ultra-large files in batches, the configurations of
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDSare not adjusted, leading to parsing process conflicts that trigger underlying property redefinition exceptions.
How to confirm configurations are correctly set
Reference only; verify against the target FastGPT release and deployment documentation.
- Upload a single multi-column structured document, check the parsed text content to confirm that all column fields and data are fully extracted.
- Upload a long-text state-owned asset policy document, confirm that the split chunk content retains complete paragraphs and contextual logic, with no forced splitting.
- Test uploading a document larger than the default size, confirm that the upload and parsing process does not trigger timeout or file size limit error messages.
- Submit boolean parsing test data, use the conditional judgment component to verify that the parameter can be recognized and processed normally.
The values provided here are common starting points and should be evaluated against the user's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.