Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Data sources include urban bus operation service standards released by the national transportation authority, bus fare adjustment rules issued by municipal-level transportation authorities, and internal safety management regulations and assessment methods of bus operation enterprises. Data updates happen synchronously when new policies launch, and routine documents are reviewed quarterly. Most documents follow a chapter-based structure, with fixed fields including document number, issuing authority, effective date, and clause details. Covered metrics include operating mileage, passenger capacity, fare standards, and assessment scores.
Constraints on document parsing and chunking
Chapter-based document structures require chunking to avoid breaking the contextual relevance of clauses. Do not split a single policy clause across multiple chunks. Metadata fields such as document number and effective date must be bound to their corresponding main text. Extracting these fields separately causes semantic loss. Multi-dimensional operation-related metrics require retaining the association between adjacent fields. Prevent ambiguity where values and their descriptions are separated. Some local policies are available only as scanned documents. Adapt the OCR parsing process to fully extract text content. Non-fixed frequency updates require the parsing link to support incremental synchronization. Avoid re-parsing already processed documents.
How to configure settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Aligns with the chapter length of bus policy documents, avoids splitting complete clauses |
chunkSplitter | By heading and paragraph boundaries | Preserves the relevance of policy clauses, matches the structural characteristics of chapter-based documents |
customPdfParse | Enable enhanced parsing mode | Adapts to scanned document formats of some local policies, fully extracts metadata and main text |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Addresses parsing time for long documents, prevents parsing tasks from timing out |
UPLOAD_FILE_MAX_SIZE | 500 MB | Supports batch uploads of local policy compilation documents |
chunkSettingMode | Manual configuration | Allows flexible adjustment of parameters for format differences across local policies |
Three common configuration errors
- When uploading policy documents that include external image links, parsing results are empty or only return text fragments. The
customPdfParseenhanced parsing mode is not enabled. The default parsing cannot capture embedded text associated with external images. - After locally deploying MinerU and enabling the enhanced parsing option, knowledge base parsing still uses the default engine. The FastGPT backend is not configured with the MinerU API address and port. The system cannot call the locally deployed parsing service.
- Clause truncation appears after parsing long policy compilation documents. The
chunkSizevalue is too small, not adapted to the length of chapter-based documents, leading to complete clauses being split.
How to verify correct configuration
- Upload a local scanned bus policy document, check if the parsing result includes metadata such as document number and effective date.
- Upload a policy compilation document larger than 100 MB, confirm that the parsing task does not trigger a timeout error.
- View the knowledge base parsing logs, confirm that the parsing service called is the MinerU enhanced mode.
- Manually adjust
chunkSizeto different values, upload the same policy document, compare whether the chunking results retain clause integrity.
The values listed above are common starting points, and should be evaluated 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.