What the data for this category looks like
Daily documents for this category typically include process sheets, layout plans, cutting orders, process quality tables, and packaging and size matching lists. Data comes from order size charts, CAD patterns, hanging systems, needle detectors, and customer sample seals. Documents are updated by order, style, and batch. Each document includes fields such as piece specifications, fabric direction, process requirements, and exception records. Common units are centimeters, piece counts, and batch numbers. Piece data is irreversible, with no room for repeated reuse.
What constraints these characteristics impose on document parsing and chunking
A large number of styles and strong seasonality lead to small per-batch document volume but large total stock. Chunking must isolate data by order batch to avoid cross-batch data confusion. Piece data is irreversible, so original field associations must be preserved during parsing. Key information cannot be arbitrarily merged or truncated. High-frequency updates require chunking to support incremental parsing, processing only newly added or modified documents. Diverse document formats — including CAD drawings, process tables, and exception report text — require adaptation to different parsing rules. This ensures complete extraction of core fields such as piece specifications and fabric direction.
How to set configurations
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the length of single-process descriptions in garment process documents, avoiding splitting that breaks process associations |
chunkSplitter | By paragraph + custom delimiter | Supports splitting by process steps and exception record types, preserving document logical structure |
customPdfParse | Enabled | Adapts to PDF process sheets exported from CAD, retaining field extraction from vector image text layers |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Reserves sufficient parsing time for large layout plan PDFs, avoiding timeout interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates batch uploads of multiple process documents per batch, adapting to bulk order scenarios |
chunkSettingMode | Manual configuration | Adjusts chunking rules for different document types (cutting orders/exception reports) to meet specialized requirements |
Three common misconfigurations
- Symptom: Index errors occur after the total number of chunks exceeds 3000. Cause: Documents are not split by order batch, resulting in too many cross-batch chunks in a single knowledge base, exceeding the system's default chunk limit.
- Symptom: CSV file chunking errors occur after upgrading the version. Cause: The new parsing rule's validation of CSV field delimiters is not adapted, and the delimiter configuration for existing files is not updated synchronously.
- Symptom: Uploaded layout plans cannot extract piece counts. Cause: The
customPdfParseconfiguration is not enabled, resulting in failure to correctly extract core fields from the vector text layer of CAD-exported PDF documents.
How to confirm configurations are correct
- Upload a single process sheet PDF, view the chunk preview, and confirm that core fields such as piece specifications and fabric direction are not split or truncated.
- Upload batch packaged documents, check that the uploaded file size does not exceed the
UPLOAD_FILE_MAX_SIZEsetting, and there are no upload failure prompts. - Trigger an incremental parsing task, confirm that only newly added or modified documents are processed, and there are no duplicate chunk records.
- Test an exception report CSV file, confirm that complete associations of fields such as exception type and process number are retained after chunking.
The values provided are common starting points and should be measured against the reader's own samples.