Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
This category of data mainly comes from the national government procurement website, local government procurement websites, fiscal integration systems, purchaser budget systems, and supplier credit records, with a daily update frequency. Commonly processed documents include procurement requirements, procurement plans, bidding documents, tender documents, bid evaluation reports, winning bid announcements, procurement contracts, question and reply documents, etc. Data fields involve procurement subjects, project numbers, supplier qualifications, evaluation results, etc. Some documents are in structured table format, while others are long unstructured explanatory texts, with no unified fixed formatting rules.
What constraints do these characteristics impose on the "document parsing and chunking" link
Daily updated business data requires the parsing link to have stable concurrent processing capabilities to avoid delays affecting the real-time response of service consultation. Mixed processing of multiple types of documents requires adapting chunking logic for structured tables and unstructured explanatory texts, to prevent splitting from destroying table associations or the integrity of key clauses. Data formats vary significantly across different sources, so support for custom parsing rules is needed to adapt to proprietary formats exported by fiscal integration systems. Service consultation relies on accurate context association, so business logical associations across paragraphs must be retained during chunking to avoid splitting key process descriptions.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Government procurement documents often include clause numbers and associated explanations. This range retains the integrity of a single business logic and avoids splitting cross-page clauses |
chunkSplitter | Split by paragraph + heading | Adapts to document structures with hierarchical headings such as bidding documents and bid evaluation reports, ensuring chunks align with business chapters |
customPdfParse | Enabled | Adapts to encrypted or proprietary format PDF documents exported by fiscal systems, retaining original table and formatting structures |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large bidding documents takes a long time. This duration covers the parsing process for most conventional documents |
chunkSettingMode | Manual configuration | Allows adjusting chunking parameters for different document types, adapting to significantly different document formats such as procurement requirements and winning bid announcements |
Three common mistakes
- Duplicate document blocks appear in the knowledge base after parsing, and the index order does not match the custom chunking results. The cause is that the system automatic deduplication configuration is not turned off, or the chunking parameters do not adapt to duplicate clause numbers in the document, causing the system to mistakenly judge the content as duplicate.
- A timeout error is triggered when parsing large bidding documents, and the interface displays parsing failure. The cause is that the
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted, and the default duration is insufficient to cover the parsing process of large documents. - The parsing status cannot be obtained after creating a knowledge base via API. The cause is that the corresponding status query interface is not called correctly, or the query request is initiated before the parsing process is completed.
How to confirm the configuration is correct
- Upload a single typical bidding document, check the parsed chunking results, and confirm that clause numbers and business content are not split.
- Adjust the
chunkSizeparameter, re-upload the same document, compare changes in chunk length, and verify that the parameter takes effect. - Upload an encrypted fiscal system exported PDF, confirm that the original table structure is retained after parsing with no content loss.
- Call the API to submit a parsing task, wait for the preset duration, query the parsing status, and confirm that the status updates normally.
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.