Reference only; verify against the target FastGPT release and deployment documentation.
What This Category of Data Looks Like
Data in this category primarily comes from local public data open platforms, integrated big data platforms, departmental business databases, and electronic certificate repositories. The update frequency is once per month. Commonly processed documents include public data resource catalogs, data open lists, data item descriptions, desensitization rules, interface documents, and update frequency descriptions. These documents contain fields such as data item name, open scope, desensitization method, interface address, and update cycle. Some interface documents also include call parameters and format descriptions.
Constraints for Document Parsing and Chunking
Batch documents updated monthly require parsing and chunking workflows to support batch processing, to avoid single-run timeouts. Mixed multi-type documents (structured catalogs, semi-structured interface documents, rule texts) require adapted parsing logic, to prevent incorrect splitting of structured fields. Parameters and addresses in interface documents are critical business information, which must be kept complete and not split during chunking. Associated information of data items, such as open scope and desensitization rules, must retain context to avoid breaking business logic during chunking. Single documents may contain multi-page structured tables; chunking must balance table integrity and length limits.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Aligns with the field length of public data catalog documents to avoid splitting critical business fields |
chunkSplitter | Split by headings and delimiters | Matches the structure of public data documents organized by chapter and data item, preserving business context |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses parsing time for single large interface documents or batch catalog files, preventing timeout errors |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Fits the size of batch catalog and interface document packages exported from local public data open platforms |
chunkSettingMode | Manual configuration | Adjusts chunking parameters for different types of public data documents (catalogs, interface documents, desensitization rules) to match business needs |
Three Common Mistakes
- Uploading a 10+ MB public data interface document triggers a
timeout of 360000ms exceededprompt in the interface. The cause is failure to adjust thePARSE_FILE_TIMEOUT_SECONDSparameter. The default timeout duration is insufficient for parsing large-volume interface documents. - Some docx-format public data desensitization rule documents cannot be parsed normally. The cause is that the documents contain complex nested tables or custom styles, and the default parsing rules do not adapt to this type of government document format.
- After configuring chunking parameters, the knowledge base question and answer response does not fully match the original text of the preset question and answer pair. The cause is that chunking parameters were not adjusted according to business requirements, leading to critical response text being split or not fully recalled.
How to Verify Proper Configuration
- Upload a single public data catalog document of around 100 MB, and check whether the parsing task duration is lower than the value set for
PARSE_FILE_TIMEOUT_SECONDS. - Import a public data open list containing multiple data items, and verify that the chunking results retain complete fields and associated descriptions for each data item.
- Upload a docx-format desensitization rule document, and confirm that the parsed text does not lose desensitization rule entries within tables.
- After setting
chunkSizeto 800–1200 characters, check that the chunking results do not split the description of a single data item into multiple independent chunks.
The values provided 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.