What the data for this category looks like
Data for this category comes from China’s national government procurement website, local government procurement websites, the fiscal integration system, the purchaser’s budget system, and supplier credit records. Updates occur daily. Common documents include procurement requirements, procurement plans, bidding documents, tender documents, bid evaluation reports, winning bid announcements, procurement contracts, and question responses. Most documents use fixed format frameworks, with standardized fields such as procurement numbers, supplier names, and evaluation scores. Some documents include embedded tables and image resources.
What constraints these characteristics impose on the "document parsing and chunking" link
Multi-source documents updated daily have large format differences. Native documents and scanned documents coexist. Some documents contain complex embedded tables and images. Fixed-format official announcements and flexibly formatted supplier tender files are mixed together. General parsing rules cannot adapt to all scenarios. Key fields such as procurement numbers and evaluation scores must be fully extracted. Missing or misaligned parsing will directly disrupt the review process. Batch parsing of multi-source data requires stability. A single failed file parse must not block overall task progress.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Config Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Government procurement documents often include long paragraphs and tables. This range balances context completeness and retrieval accuracy |
chunkSplitter | Split by paragraph + heading hierarchy | Government procurement documents have clear chapter headings and paragraph logic. This splitting method preserves semantic integrity |
customPdfParse | Enabled | Some tender documents and bid evaluation reports contain complex embedded tables and images. Enabling custom parsing fully extracts text and visual elements, and supports replacing third-party document parsing tools |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Large tender documents or multi-page bid evaluation reports require long parse times. This range prevents timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 100 MB | Complete files for government procurement projects may include multiple attachments. This upper limit covers most batch upload scenarios |
chunkSettingMode | Manual configuration | Chunking rules need adjustment for different document types. Manual configuration adapts to scenarios such as procurement requirements and tender documents |
Three common mistakes
- Symptom: After uploading a PDF-format tender document, the parsed result lacks embedded images or has abnormally displayed images. Cause: The
customPdfParseconfiguration is not enabled. The default parsing engine does not fully extract image resources embedded in the PDF. - Symptom: Uploading a large bid evaluation report triggers a parsing timeout error and returns a timeout status code. Cause: The value of
PARSE_FILE_TIMEOUT_SECONDSis too low, and does not match the full parse time required for long documents. - Symptom: Search results only return text chunks, and do not include metadata such as procurement numbers and supplier names for the corresponding document. Cause: Metadata binding logic is not configured, and document basic identification fields are not associated during chunking.
How to confirm the configuration is correct
- Upload a tender document PDF that includes embedded tables and images. Verify that the parsed result fully extracts table content and image elements.
- Upload a single 80 MB bid evaluation report. Confirm that the parsing process does not trigger a timeout error.
- Search for keywords in the test document. Check that returned chunk results include metadata such as the procurement number of the corresponding document.
- Batch upload 5 government procurement documents of different types. Confirm that all files are successfully parsed without errors.
The values provided here 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.