What the Data for This Category Looks Like
Daily documents for this category include mold design task sheets, DFM reports, part drawings, machining process cards, mold trial reports, mold modification notices, and others. Data sources include customer product models, mold flow or forming simulation files, CAD/CAM exported files, and machining and measurement data. Data updates are triggered by project milestones, with new documents and data generated after each mold trial or mold modification. The documents contain professional fields such as the number of mold cavities, material grade, hardness requirements, and machining tolerances, with units mostly using millimeters, HRC, megapascals, and other industry standard units.
What Constraints These Characteristics Impose on Document Parsing and Chunking
Mold documents mostly come from exports of CAD or simulation software, and contain complex tables and vector graphics. When chunking, the integrity of professional fields must be retained to avoid splitting key information such as tolerances and material parameters. The structure of different documents varies greatly; the chapter logic of a design task sheet and a mold trial report are completely different, so it is necessary to adapt to diverse chunking rules. Frequently updated project data requires the parsing process to support fast incremental synchronization, avoiding repeated parsing of full documents. There are many professional terms, and associated terms must not be split into different chunks during chunking.
How to Configure Settings
These configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Mold documents contain professional tolerances and material parameters. Chunks that are too long will lose contextual association, while chunks that are too short will split process steps |
chunkSplitter | Split by paragraph + heading | Mold documents have clear task chapters and mold trial chapters; splitting by heading preserves the logical structure of the document |
customPdfParse | Enabled | Mold PDFs are mostly exported from CAD, containing vector images and complex tables. Enabling custom parsing preserves table structure and text position |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Large mold design drawing PDFs have many pages and take a long time to parse; the default timeout may cause parsing failure |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Mold design source files include CAD models and simulation data; packaged files have large volume, so the upload limit must be adjusted |
chunkSettingMode | Manual mode | The document structure varies greatly across different projects; manual configuration allows adjusting chunking rules for different mold types such as injection molding and stamping |
Three Common Configuration Mistakes
- Symptom: Empty text or parsing failure is returned when configuring a custom document parsing workflow. Cause: The
customPdfParseparameter is not configured correctly, and the special format of mold CAD-exported PDFs is not adapted to. - Symptom: Key professional information is lost when a file is uploaded and directly passed to the large model. Cause: The
chunkSizeparameter is not adjusted; the default chunk length cannot accommodate complete paragraphs containing professional parameters. - Symptom: The PDF parsing workflow suddenly triggers an error. Cause:
PARSE_FILE_TIMEOUT_SECONDSis not set; parsing of large mold drawings takes longer than the system default threshold, causing termination.
How to Verify the Configuration Is Correct
- Upload a mold design task sheet PDF, and check whether the parsed text fully retains the format and content of professional fields such as material grade and number of mold cavities.
- Adjust
chunkSizeto 1000 characters, generate chunking results, and check whether adjacent chunks retain the contextual association of process steps. - Upload a single 800 MB mold simulation report PDF, and confirm that there are no timeout errors in the upload and parsing process.
- Enter the file collection configuration page of the corresponding knowledge base, and confirm that the
customPdfParseswitch is enabled.
The values provided are common starting points and should be measured against the reader's own samples.