What the Data for This Category Looks Like
The data for this category mainly comes from the national land spatial planning one-map, project approval and supervision platform, land use status database, and ecological protection redline data. Update frequency is irregular. Daily processed documents include land pre-examination and site selection opinions, construction land planning permits, redline drawings, and planning conditions. Document structures contain fields such as approval document numbers, planning control indicators, redline coordinates, and prior approval requirements. Some documents include vector-format planning redline drawings and structured tables.
Constraints Imposed by These Characteristics on the "Document Parsing and Chunking" Link
Documents in this category have cross-system sources, leading to diverse formats. These include plain-text approval documents, planning condition documents with tables, and vector redline drawings. Parsing processes must support multi-format content. Approval clauses, planning indicators, and redline coordinates in the documents are strongly correlated. Chunking must retain contextual logic, and avoid splitting policy basis across chapters. Irregularly updated data requires parsing processes to be flexible. They must adapt to different versions of approval document formats. Parsing errors directly affect project approval and construction. It is necessary to ensure no key information is missed.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Land pre-examination documents contain continuous planning indicators and approval clauses. This value range balances contextual association and chunk granularity, and avoids losing key information |
chunkSplitter | By paragraph + title anchor | Documents in this category have clear approval chapters and clause numbers. Splitting by title retains the complete logic of policy basis, and avoids chunking across chapters |
customPdfParse | Enable table and vector image parsing | Documents in this category include redline drawings and planning condition tables. Default parsing cannot fully extract vector coordinates and table data. Enabling this option retains key information |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large redline drawing PDFs or multi-page planning permit documents take a long time to parse. The default timeout is insufficient. Extending to 600 seconds covers the complete parsing process |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Documents exported from the national land spatial planning one-map for some projects have large file sizes. Adjusting the upload limit prevents files from being blocked |
chunkSettingMode | Manual configuration | Documents in this category have large format differences. Manual configuration allows targeted adjustment of chunking rules to adapt to the structure of different approval documents |
Three Common Mistakes
- Phenomenon: The model cannot read uploaded files, and the interface displays parsing failure status code 413. Cause: The
UPLOAD_FILE_MAX_SIZEconfiguration is not adjusted, and documents exceeding the default limit are blocked. - Phenomenon: Tables and redline drawings in Word documents are identified as empty or garbled. Cause: The vector image and table parsing module of
customPdfParseis not enabled. Default parsing only extracts plain text content. - Phenomenon: Chunked policy clauses are forcibly truncated at non-logical nodes. Cause: The default character chunking rule is used, and the
chunkSplitterconfiguration is not set to title anchor.
How to Confirm the Configuration Is Correctly Set
Reference only; verify against the target FastGPT release and deployment documentation.
- Upload a single planning redline drawing PDF document with a size exceeding 500MB, confirm that the parsing task has no timeout errors and completes normally.
- Import a Word document containing structured tables and vector redline drawings, check whether the knowledge base parsing results fully retain table data and redline coordinate annotations.
- Manually adjust
chunkSizeto 1000 characters, verify that the chunked document fragments contain complete single approval clauses and associated prior requirements. - View the parsing logs in the FastGPT backend, confirm that the
PARSE_FILE_TIMEOUT_SECONDSconfiguration value is 600 seconds, and there are no prematurely terminated parsing records.
The values listed above 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.