What this category of data looks like
This category of data primarily comes from the national land use planning unified map, project approval and supervision platform, current land use database, and ecological protection redline data, with irregular update frequency. Commonly processed documents include land pre-examination and site selection opinions, construction land planning permits, construction engineering planning permits, redline maps, and planning conditions. Redline maps contain geographic coordinate fields; planning conditions mark clear fields and units such as land use nature, floor area ratio, and land area. Some documents are CAD format vector files or official-annotated PDF approval documents.
What Constraints Do These Characteristics Impose on the Document Parsing and Chunking Link
The cross-system sources, semi-structured formats, and irregular update characteristics of this category of data impose multiple constraints on document parsing and chunking. Mixed structured and unstructured files from different systems require parsing logic adapted to different formats. Irregularly updated approval documents contain new fields, so chunking configurations must support non-standard content extraction. Redline maps and CAD vector files cannot be parsed via general text parsing, so targeted parsing rules must be configured. Numeric control fields such as land area and floor area ratio in planning conditions require accurate extraction to avoid splitting. Approval documents with cross-department annotations must retain contextual associations to avoid breaking control requirements.
How to Configure the Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapt to the text length of planning conditions and approval opinions, avoid splitting key control indicators |
chunkSplitter | Split by title + paragraph | Adapt to structured documents with approval document numbers and project names, retain contextual associations |
customPdfParse | Enabled | Adapt to official-annotated PDF approval documents, retain text content in annotation areas |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Adapt to the parsing time of large redline map CAD files, avoid timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Support large vector files exported from the national land use planning unified map |
chunkSettingMode | Manual configuration | Adapt to irregularly updated non-standard fields, flexibly adjust chunking rules |
Three Common Configuration Mistakes
- After uploading a large CAD-format redline map, the parsing task times out and displays a timeout error. Cause: The
PARSE_FILE_TIMEOUT_SECONDSparameter was not adjusted to a value suitable for large vector files, and the default timeout duration is insufficient. - After upgrading FastGPT v4.17.0, the originally configured CSV-format project approval files cannot be chunked normally. Cause: The default rules for
chunkSplitterwere adjusted in the new version, and the custom chunking logic was not updated synchronously. - When attempting to upload a PDF approval file directly to the large model, the system's default text parsing process is still triggered. Cause: The automatic parsing switch was not turned off, or the
customPdfParseparameter was not set to the off state to preserve the original file workflow.
How to Verify Successful Configuration
- Upload a single typical document (such as a planning condition PDF), view the parsed text fragments, and confirm that key control fields are not split.
- Upload a large vector file, check whether the parsing task duration matches the preset
PARSE_FILE_TIMEOUT_SECONDSduration, and there are no timeout error reports. - Call the API to create a file collection, verify that the specified parsing configuration items are included in the request parameters, and the returned results meet expectations.
- Compare the chunking results before and after the upgrade, confirm that the custom
chunkSplitterrules take effect, and there are no abnormal errors.
The values provided are common starting points and should be measured against the reader'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.