This document applies to FastGPT v4.17.0
What This Category of Data Consists Of
Core data for this category comes from tender notices, tender documents, enterprise qualification databases, and supplier quotations. Commonly processed documents include tender documents, qualification certificates, business deviation tables, and technical deviation tables. These documents typically have fixed-format sections such as qualification response areas, business clause areas, and technical solution areas. Fields include tender number, project name, qualification number, quotation amount, original deviation clauses, and more. Data update frequency is per project; all processed documents relate to new tender projects each time, with no fixed reusable historical datasets.
Constraints Imposed on Document Parsing and Chunking
Documents in this category come from diverse sources and have inconsistent formats. These include styled DOCX tender documents, fixed-layout PDF tender documents, and some qualification materials may be scanned image formats. This requires the parsing link to support multiple file formats. At the same time, document content must strictly align with tender requirements; splitting or losing any key clause may lead to compliance issues. Therefore, chunking must preserve the integrity of individual response content, and avoid splitting a single qualification or business deviation content into multiple chunks. Additionally, single projects have a large number of attachments with high total volume; the parsing link must support batch processing and avoid timeout interruptions, otherwise tender preparation progress may be delayed.
Configuration Recommendations
The following configurations are formulated for the document parsing and chunking needs of the tender document preparation scenario:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Matches the length of individual response clauses in tender documents, to avoid splitting key qualification or business content |
chunkSplitter | Split by title + paragraph | Adapts to the fixed section structure of tender documents, preserving the integrity of each response unit |
customPdfParse | Enabled | Preserves the original layout and section hierarchy of tender document PDFs, avoiding loss of formatting in text flow |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Adapts to parsing time for multi-attachment uploads, avoiding timeout errors for large file parsing |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates multiple attachments such as qualifications and quotations attached to tender projects |
chunkSettingMode | Manual mode | Allows flexible adjustment of chunking rules to accommodate format differences across different tender projects |
Three Common Mistakes
- Issue: A single deviation entry in the business deviation table is split into multiple independent chunks. Cause: The
chunkSplitteris not configured to follow the section titles of the tender document, and the default character-based splitting strategy is used, resulting in truncation of a single clause. - Issue: Some files fail to complete parsing after uploading multiple attachments. Cause:
UPLOAD_FILE_MAX_SIZEorPARSE_FILE_TIMEOUT_SECONDSare not adjusted, causing files exceeding limits to be automatically truncated or parsing to time out. - Issue: The parsed result does not retain hierarchical association between parent and child chunks. Cause: The corresponding configuration is not enabled in
chunkSettingMode, or the parent-child chunking function supported by FastGPT v4.17.0 is not used.
How to Verify Correct Configuration
- Upload a test document containing a complete business deviation table, and check whether each deviation entry is an independent chunk in the chunking result.
- Upload multiple test attachments in different formats at the same time, and confirm that all files show a completed parsing status.
- View the parsed text flow, and confirm that the title hierarchy and paragraph separations of the original document are retained.
- Upload a tender document containing multi-level titles, and check whether the parsed chunks retain parent-child hierarchical associations.
- Test uploading a single large-volume attachment, and confirm that no timeout errors occur during the parsing process.
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 2026-10-19.