What this category of data looks like
Reference only; verify against the target FastGPT release and deployment documentation.
What constraints do these characteristics impose on the "document parsing and chunking" link
The embedded tables and images in steel smelting Party building documents increase parsing complexity. Parsing tools must fully extract assessment data in tables and activity descriptions corresponding to images to avoid information loss. A large portion of documents combine Party building policies with production indicators. When chunking, contextual connections must be retained, and policy points must not be separated from their corresponding production scenarios. Some documents have inconsistent formatting: for example, a batch of issued learning files may mix title levels, so flexible chunking rules must be supported. In addition, large Party building compilation documents have large individual file sizes, so upload and parsing time and capacity limits must be relaxed.
How to set the configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Party building documents include original policy text and production-related cases. Chunks that are too long will lose contextual connections, while chunks that are too short will damage semantic integrity |
chunkSplitter | Split by paragraph + title | Matches the chapter structure of Party building documents, and avoids splitting learning key points across titles |
customPdfParse | Enable table and image OCR enhancement mode | Steel smelting Party building documents often contain non-standard layout content such as workshop Party building activity photos and monthly learning assessment statistics tables. The enhancement mode can fully extract table text and image-related descriptions |
UPLOAD_FILE_MAX_SIZE | 500 MB | Meets the upload requirements for individual large Party lecture courseware or annual Party building compilation documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Prevents parsing timeouts when processing documents with multiple high-definition activity photos |
chunkSettingMode | Manual configuration | Adjust chunking rules for different types of Party building documents such as policy documents and meeting minutes |
Three common mistakes
- Phenomenon: Extracted document table content is missing, or text corresponding to images is empty; Cause: The OCR enhancement mode of
customPdfParseis not enabled, so associated text of embedded tables and images cannot be extracted. - Phenomenon: Content from across chapters is spliced together after chunking, and the chapter logic of Party building documents cannot be reflected; Cause:
chunkSplitteris not set to split by title, and the default character-based splitting destroys the document's hierarchical structure. - Phenomenon: Reference materials or citation fields at the end of the document are not included in the parsing result; Cause: The reference field extraction rule of
customPdfParseis not configured, and only the core main body content is extracted.
How to confirm the configuration is correct
- Upload a Party building test document that includes tables and images, and check whether the table text and image descriptions in the parsing result are fully extracted.
- Adjust the
chunkSizeparameter, and confirm that the length of each chunk falls within the preset 800–1200 character range via the chunk preview interface. - Upload a large Party building compilation document under 500 MB, and confirm that the parsing task completes within 600 seconds without timeout errors.
- Switch
chunkSettingModeto manual configuration, upload different types of Party building documents, and confirm that the chunking rules take effect as preset.
The values provided here are common starting points, and should be evaluated against the user'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.