What the data for this category looks like
This category serves injection molding, stamping, and die-casting mold enterprises, as well as mold centers within manufacturing enterprises. Daily generated documents include mold design task sheets, DFM reports, part drawings, machining process cards, mold trial reports, mold modification notices, and mold history records. Data sources include customer product models, mold flow or forming simulation results, CAD/CAM output files, machining and measurement data, and mold trial sample records. Data updates are triggered by project milestones. New record documents are generated after each mold trial or modification. Documents contain professional fields such as demolding angle, shrinkage rate, and machining tolerance, with units mostly including angles, millimeters, and percentages.
What constraints these characteristics impose on document parsing and chunking
Mold design and manufacturing documents have diverse types. They include both structured report documents and annotated CAD drawing export files. Data is updated with each mold trial or modification, which places high requirements on the relevance of parsed chunks. If the association between "mold trial defects" and "modification plans" is split during chunking, subsequent ledger summary will fail to match corresponding relationships. Additionally, some documents contain large CAD attachments and multi-page mold trial records, leading to long parsing times. If chunking rules do not adapt to the document structure, content truncation or loss of key parameters may occur. Furthermore, the units and meanings of professional fields must be retained during chunking to avoid confusing parameters from different mold projects.
How to set the configurations
The following configurations are formulated based on the document parsing and chunking feature of FastGPT v4.17.0:
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the conventional length of process segments and defect records in mold documents, avoiding splitting of key parameter associations |
chunkSplitter | Separate by paragraph + title | Matches the structured title hierarchy of mold documents, preserving the content integrity of modules such as "mold trial defects" and "modification plans" |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Addresses parsing time for large CAD export documents and multi-page mold trial reports, avoiding mid-process interruptions |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates large documents such as complete mold design task sheets and multiple batches of mold trial records |
customPdfParse | Enable text-first mode | Preserves tables and professional annotated text in DFM reports and process cards, avoiding loss of key parameters from image parsing |
chunkSettingMode | Manual configuration | Adapts chunking rules for different types of mold documents, avoiding damage to structural associations of professional documents by generic modes |
Three common errors
- Markdown table rendering fails and displayed content is truncated when outputting models in Markdown table format. The cause is failure to adjust
chunkSizeto fit table width, resulting in splitting of complete row or column structures of the table during chunking. - An invalid image file error is returned when uploading a docx file with images. The cause is failure to enable the image compatibility processing rules of
customPdfParse, or failure to configure the local image path mapping for document parsing. - A slow operation timeout is logged during document parsing. The cause is that
PARSE_FILE_TIMEOUT_SECONDSis set too small, failing to complete the complete parsing process for large CAD attachments or multi-page mold trial reports.
How to confirm the configuration is correctly set
- Upload a single mold design task sheet, check if the parsed chunks retain complete content under titles such as "design parameters" and "processing requirements", with no split breaks.
- Upload a docx file containing a multi-page mold trial report, confirm there are no timeout errors in the parsing log, and the output ledger content can fully match the corresponding relationship between mold trial defects and modification plans.
- Upload a machining process card document containing images, confirm that dimension annotations and process description texts associated with the images in the parsing results are not lost or disordered.
- Adjust
chunkSizeto the range of 700 to 1300 characters, verify that the split content does not separate key mold parameters and corresponding records.
The values provided are common starting points and should be measured against the reader's own samples.