Reference only; verify against the target FastGPT release and deployment documentation.
What the data for this category looks like
Data for this category comes from daily business documents of entities including central SOE procurement centers, tendering offices, business requirement departments, audit and discipline inspection departments, and tendering agencies. Documents include procurement plans, requirement specifications, tender documents, bid evaluation records, contracts, acceptance slips, supplier complaint materials, and more. Data sources cover enterprise procurement platforms, public resource trading platforms, supplier databases, contract management systems, and audit problem databases. The update frequency is daily. Most documents contain both structured fields and unstructured text. Structured fields involve procurement methods, budget basis, scoring methods, supplier qualification requirements, and similar items. Unstructured text mostly consists of policy clauses, review opinions, and compliance explanations.
What constraints do these characteristics impose on the "document parsing and chunking" link
The multi-source and high-frequency update characteristics of this category require the document parsing process to have stability and adaptability. Diverse document types cover procurement plans, tender documents, and more. Different document formats need matching parsing rules. The mixed document structure of structured fields and unstructured text requires retaining the association between fields and corresponding text during chunking. This avoids breaking the context of compliance basis. Audit-related documents have high compliance requirements. The parsing process must avoid missing or truncating key policy clauses. When uploading multiple procurement-related documents in batches, ensure that the parsing results of each document can be accurately distinguished for compliance review of corresponding projects.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
customPdfParse | Enabled | This category contains a large number of PDF-format tender documents and bid evaluation records. Enabling this option preserves the original layout and text structure, avoiding layout chaos of policy clauses |
chunkSize | 800–1200 characters | Policy clauses and review opinions in this category are mostly coherent compliance bases. This range preserves the complete context of a single compliance basis, avoiding truncation of key information during chunking |
chunkSplitter | Split by paragraph + title | Documents in this category mostly divide compliance content by chapters and clauses. Splitting by paragraph and title ensures chunking matches business logic, facilitating accurate recall for subsequent policy basis Q&A |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Parsing large tender documents or bid evaluation records takes a long time. This range covers the parsing needs of most documents, avoiding parsing failure due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Some batch-uploaded tender document collections in this category have large total volume. This upper limit meets conventional batch upload requirements |
chunkSettingMode | Manual configuration | Documents in this category have large structural differences. Manual configuration allows adjusting chunking rules for different document types, adapting to the recall needs of policy and regulatory basis |
Three Common Mistakes
- Phenomenon: When calling the file collection creation API, cannot specify the parameter to enable PDF parsing, returns a parameter error. Cause: Did not correctly use the configuration format of the
customPdfParseparameter, or did not correctly declare this parameter in the API request body. - Phenomenon: PDF parsing process suddenly reports an error with no clear log prompt. Cause: The recently updated parsing dependency version is incompatible with the existing deployment environment, or the uploaded PDF file is encrypted, damaged, or otherwise unparseable.
- Phenomenon: After uploading multiple procurement documents in batches, the parsing results cannot be distinguished by file, leading to confusion of policy bases of different projects during compliance review. Cause: Did not enable file metadata binding configuration, or did not retain the identification information of the corresponding file during chunking.
How to Confirm the Configuration is Correct
- Upload a typical PDF-format tender document, check if the parsed text retains the original chapter titles and paragraph structure, to verify whether the
customPdfParseconfiguration takes effect. - Adjust the value range of
chunkSize, upload a document containing long policy clauses, check if the chunked text completely retains the clause context, to verify whether the chunking rules adapt to business needs. - Call the batch upload API to upload multiple different types of procurement documents, check if the parsing results come with the metadata identification of the corresponding file, to verify whether the multi-file distinction configuration is correct.
- Upload a document collection with a volume close to
UPLOAD_FILE_MAX_SIZE, check if the parsing process completes normally, to verify whether the upload and parsing upper limit configuration is reasonable.
The values provided are common starting points and should be measured 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.