What data in this category looks like
Data related to service inquiries and responses for rail transit equipment primarily comes from equipment operation and maintenance manuals, project bidding documents, government service approval templates, and official response standard documents. Data updates occur irregularly alongside equipment iterations, policy adjustments, or project progress, with no fixed cycle. Most documents follow a chapter-based structure, including standardized parameter tables, fault handling procedures, embedded assembly diagrams, fault code legend images, and other visual materials. Some spare parts lists are stored in Excel format, containing fields such as equipment model, responsible team, maintenance cycle, and more. Common parameter units include degrees Celsius, kilonewtons, and kilometers per hour.
Constraints Imposed on Document Parsing and Chunking
The multi-image embedded, chapter-based structure, and cross-page process characteristics of rail transit equipment documents create multiple constraints for the document parsing and chunking stage. First, embedded images in Excel files and assembly diagrams within documents must be fully extracted to avoid losing visual auxiliary information after parsing. Second, dense parameter tables and cross-page fault handling procedures require retaining field associations and contextual coherence during chunking, to avoid splitting complete operational steps. Finally, document length varies widely, from short several-page service guides to hundreds of-page operation and maintenance manuals, so adaptive chunking thresholds are needed to prevent overly short fragmented chunks or overly long redundant chunks.
Configuration Settings
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the length of parameter tables and operational steps in rail transit equipment documents, avoiding splitting of single complete processes |
chunkSplitter | Split by paragraph + heading | Matches the chapter-based structure of documents, preserving contextual integrity of fault handling and parameter descriptions |
customPdfParse | Enable embedded image extraction | Adapts to assembly diagrams and fault code legends embedded in documents, ensuring visual information can be passed to the large model |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Covers upload requirements for hundreds-page operation and maintenance manuals and large bidding documents |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses time consumption requirements for long document parsing, avoiding parsing interruptions due to timeout |
chunkSettingMode | Auto-adjust based on document type | Adapts to chunking differences across different document types such as service guides and operation and maintenance manuals |
Three Common Misconfigurations
- The image fields in parsed Excel documents are empty, and the large model cannot obtain visual annotations for the spare parts list. This occurs because the Excel image extraction switch in
customPdfParseis not enabled, so only table text is parsed while embedded images are ignored. - A 504 timeout error occurs when parsing large operation and maintenance manuals, resulting in incomplete returned results. This happens because the
PARSE_FILE_TIMEOUT_SECONDSvalue is set lower than the actual parsing time, failing to adapt to the processing rhythm of long documents. - The chunking result splits cross-page fault handling procedures into multiple independent chunks, preventing the large model from associating preceding and following steps. This is caused by using only character length for
chunkSplitter, without matching the document's heading and paragraph structure.
How to Verify Proper Configuration
- Upload an Excel spare parts list containing embedded images, and check whether the parsed result retains image previews or corresponding image text annotations to confirm that the
customPdfParseconfiguration is effective. - Upload a single operation and maintenance manual with more than 500 pages, wait for parsing to complete, and check the completeness of the returned results to confirm that the
PARSE_FILE_TIMEOUT_SECONDSsetting matches the actual processing time. - Select a cross-page fault handling procedure document, and check whether the chunking result retains the complete contextual flow of the procedure to confirm that the
chunkSplitterrule adapts to the document structure. - Upload a test document of the maximum allowed size, confirm that the upload and parsing processes are not interrupted, and verify the rationality of the
UPLOAD_FILE_MAX_SIZEconfiguration.
The values provided are common starting points and should be evaluated against relevant test 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.