What the data for this category looks like
Daily documents for this category include process sheets, marker plans, cutting orders, first sample sealing records, process quality forms, packaging size allocation sheets, and more. Data comes from order and size charts, CAD pattern systems, fabric and accessory inspection records, hanging production systems, needle inspection equipment, and customer sample sealing documents. Data updates are adjusted in real time with each garment order and each production batch. Single documents often contain fields such as pattern parameters, process requirements, size ratios, and fabric specifications. Units include centimeters, gram weight, piece count, batch numbers, and more. Document structure varies significantly based on order style.
What constraints these characteristics impose on document parsing and chunking
Documents for this category vary widely based on order style, and contain mixed-format content such as CAD drawings and process lists. This places higher requirements on parsing format adaptability. Real-time updated production data requires the parsing link to support fast response, to avoid slowing down the ledger summary process. The multi-field mixed document structure requires retaining field relevance during chunking, to avoid losing key associated information such as size and batch numbers after splitting. The irreversible business characteristic of fabric cutting pieces requires parsed chunk data to accurately correspond to production batches, to prevent batch deviations caused by data misalignment.
How to set configurations
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the mixed length of pattern parameters and process descriptions in documents of this category, avoids losing complete logic of single process after splitting |
chunkSplitter | Semantic chunking | Matches the multi-field mixed structure of garment production documents, retains the association between process requirements and size ratios |
customPdfParse | Enable structured parsing | Adapts to non-pure text PDF content such as CAD patterns and marker plans, extracts reusable pattern parameters and piece count information |
PARSE_FILE_TIMEOUT_SECONDS | 180–240 seconds | Addresses parsing time consumption for single large-format marker plans or multi-page process sheets, avoids interrupting the parsing process due to timeout |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Adapts to upload requirements for large-format marker plans or multi-page process sheets, avoids failure to parse due to oversized files |
chunkSettingMode | Manual configuration | Adapts to document differences across different order styles, flexibly adjusts chunking rules to match various production document structures |
Three common errors
- Calling the file parsing API shows a successful call but no returned results: The cause is incorrect configuration of the
PARSE_FILE_TIMEOUT_SECONDSparameter, the result return logic is not triggered after timeout. - Parsing logs prompt split-related errors: The cause is that
chunkSplitteris configured for fixed-length chunking, format breakage occurs when splitting vector content related to CAD drawings. - No data is input to the document parsing module: The cause is that the structured parsing switch for
customPdfParseis not enabled, non-pure text documents such as CAD patterns and marker plans cannot be read correctly.
How to confirm the configuration is set correctly
- Upload a single process sheet PDF, view the parsed chunk results, confirm that each chunk contains complete process description or size ratio information.
- Call the file parsing API of FastGPT v4.17.0, wait for the preset duration of
PARSE_FILE_TIMEOUT_SECONDS, then check if the returned chunk data is complete. - Upload a document containing a CAD marker plan, confirm that structured fields such as fabric specifications and piece count are extracted in the parsing results.
- Adjust the
chunkSizeparameter, then upload the same type of test document, verify whether the chunk length matches the preset value range.
The values provided here are common starting points, and should be evaluated against samples relevant to the intended deployment.