This page is compatible with FastGPT v4.17.0.
What the data for this category looks like
The data for this category primarily comes from daily business documents and production equipment data from garment OEM/ODM and cutting and sewing workshops. Daily documents include process sheets, layout plans, cutting orders, first sample sealing records, process quality tables, and more. Fields cover cutting piece specifications, size ratios, fabric color numbers, process quality inspection standards, and more. Common units are pieces, meters, centimeters, and yards. Data updates align with each order, each style, and each production batch. Order size charts, CAD template parameters, and hanging system circulation records are updated synchronously.
Constraints for Document Parsing and Chunking
Documents for this category mostly consist of structured content such as process details and template parameters. A large volume of cross-document associated size and color number information exists. Chunking must retain field relevance. Multi-batch, high-frequency updated data requires chunking to adapt to dynamically changing order information. This prevents chunking from being out of sync with latest production requirements. The irreversible nature of cut pieces requires parsed chunks to accurately link to original production documents. Numeric fields such as fabric width and cutting piece quantity must avoid parsing deviations. Manual production processes lead to handwritten annotations on some paper documents. Parsing must support non-standard document formats.
Configuration Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Category documents mostly include process details and template parameters. This range fully covers the complete requirements of a single process and avoids split breaks |
chunkSplitter | Split by paragraph and semantic units | Adapts to the structured paragraph layout of process sheets and layout plans, and retains logical connections between fields |
customPdfParse | Enable CAD template-specific parsing mode | A large number of layout plans and template files in PDF format exported from CAD exist for this category. The dedicated mode accurately identifies cutting piece coordinates and size parameters |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large layout plan PDFs take a long time to parse. This duration covers the parsing needs of most bulk documents |
chunkSettingMode | Manual configuration | Adjust chunking parameters for different document types (process sheet / cutting order / layout plan). Manual mode adapts to flexible configuration needs |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Support batch uploading of production documents for multi-batch orders, covering the upload needs of large CAD files |
Three Common Configuration Mistakes
- An error occurs when processing PDFs after deploying Marker, with the response
{"detail":"Error message"}. The cause is that the CAD parsing dependency environment corresponding tocustomPdfParseis not configured. Layout plan PDFs cannot be parsed normally as a result. - Document source cannot be traced after knowledge base chunking. The cause is that document metadata binding configuration is not enabled. Chunking does not link identification fields such as the original file's order number and style number.
- Fields are scrambled after parsing multi-line XLSX format size tables. The cause is that the default chunking logic does not adapt to merged cells and cross-row data in tables. This causes the field splitting logic to fail.
How to Verify Correct Configuration
- A typical layout plan PDF exported from CAD can be uploaded. Parsed text blocks are checked for core information including cutting piece coordinates and width parameters.
- Process sheets from multiple distinct orders can be uploaded. Chunked text is verified to retain associated fields such as size ratios and process requirements for the corresponding order.
- Large files under 1000 MB can be tested. Upload and parsing processes are confirmed to have no timeout errors.
- Chunked text is checked for metadata including the original file's name and upload time. The traceability function is confirmed to be available.
The values provided are common starting points and should be measured against the reader's own samples.