What the data for this category looks like
The entities involved in this category include the performance evaluation department of the finance bureau, budget units, third-party evaluation institutions, and competent departments. Daily processed documents include performance target forms, self-assessment forms, performance evaluation reports, supporting material lists, rectification reports, and more. Data sources cover the integrated budget management system, project declaration materials, fund payment data, business system ledgers, and satisfaction survey results. The update frequency is annual. The most time-consuming steps in the business process are collecting project output and effect supporting materials, and verifying the correspondence between fund expenditures and performance targets.
What constraints do these characteristics impose on the "document parsing and chunking" link
Document types cover structured tables, unstructured reports, and image-based supporting materials. Sources include standardized files exported from systems and non-standardized documents filled out manually. Chunking rules must balance the integrity of structured content and the semantic relevance of unstructured content. The annual update business feature means the system must adapt to batch processing of historical documents, avoiding parsing timeouts for single tasks. The core of the business is verifying the correspondence between funds and performance targets. During chunking, the association logic between performance targets, fund data, and supporting materials must be retained, to avoid damaging the integrity of business information and preventing subsequent business risks.
How to set the configurations
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the typical paragraph length of documents such as performance evaluation reports and self-assessment forms, and retains the semantic association between performance targets and corresponding supporting materials |
chunkSplitter | Split by paragraph + heading hierarchy | Matches the hierarchical structure of performance targets, supporting material lists, and rectification content in documents, avoiding cross-topic chunking |
customPdfParse | Enable image OCR parsing | Adapts to scanned documents and screenshot images in supporting materials, extracting text content within images |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Adapts to the parsing time required for batch historical documents and large-volume performance evaluation reports, avoiding parsing interruptions |
UPLOAD_FILE_MAX_SIZE | 100 MB | Covers the upload requirements of single-batch project declaration materials and supporting material packages |
chunkSettingMode | Manual configuration | Adjust chunking rules for different document types (such as performance target forms and rectification reports) to adapt to the unique characteristics of business data |
Three common mistakes
- Issue: After uploading performance supporting materials containing images, the knowledge base can display the image links, but the text content in the images cannot be extracted during question-and-answer testing. Cause: The
customPdfParseconfiguration is not enabled, only the image links are identified without completing OCR text extraction. - Issue: After importing a complete performance evaluation report, the question-and-answer test cannot associate a single performance target with its corresponding supporting materials. Cause: The
chunkSizesetting is too small, splitting a single performance target and its corresponding supporting materials into different chunks, damaging the association logic of business information. - Issue: When parsing a single performance evaluation report with more than 50 pages, the system returns a parsing timeout error. Cause: The threshold set for
PARSE_FILE_TIMEOUT_SECONDSis too low, unable to complete the complete parsing process for large-volume documents.
How to confirm the configuration is correct
- Upload a single performance supporting material containing images, and check whether the parsed text content includes the text information within the images.
- Import a complete performance evaluation report, and verify that the chunking results retain the hierarchy and association relationship between performance targets and supporting materials.
- Upload a single document package under 100 MB, confirm that the parsing task is completed within the set timeout threshold without errors.
- During question-and-answer testing, input a question pointing to the association between performance targets and supporting materials, and confirm that the system can return matching chunked content.
The values listed above are common starting points, and should be evaluated against the reader's own sample datasets.
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.