What data for this category looks like
Daily documents for this category include production orders, VIN lists, assembly work instructions, tightening data, off-line inspection reports, rework records, and consistency release documents. Data sources cover ERP/MES systems, VIN and configuration systems, supplier batch libraries, tightening tools, EOL inspection lines, and quality traceability systems. Documents are updated either in real time per individual vehicle, or aggregated and archived per shift. The document structure uses VIN as the core associated field, and includes structured fields such as torque values, inspection thresholds, part numbers, and rework stations. Some documents contain charts and formula content.
Constraints on document parsing and chunking
The VIN-centric document structure requires parsing and chunking to preserve the contextual association of individual vehicles, and avoid splitting content associated across different VINs. Otherwise, subsequent configuration matching will fail. Real-time generated tightening data and off-line reports require the parsing link to have low-latency processing capability, to avoid timeouts that disrupt production tempo. The presence of structured fields such as torque values and inspection thresholds requires chunking to preserve the binding relationship between fields and their corresponding values, and paragraphs must not be split arbitrarily. Mixed-line production creates multi-configuration documents, requiring chunking to be done by station or inspection item to ensure subsequent retrieval and association accuracy. Batch-aggregated shift documents require the parsing link to support large file processing, to avoid task interruptions due to insufficient single-parsing capacity.
This configuration guide is based on FastGPT v4.17.0.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapt to the length of vehicle documents containing VIN, torque parameters, and inspection items, and avoid splitting the complete assembly node content of a single vehicle |
chunkSplitter | Separate by paragraph + title | Match the chapter structure of assembly work instructions, and preserve the binding relationship between station and inspection item titles and their corresponding content |
customPdfParse | Enable Marker parsing mode | Adapt to formulas, charts, and structured table content in documents, and improve parsing accuracy of tightening data and inspection reports |
UPLOAD_FILE_MAX_SIZE | 500 MB | Support batch upload of multi-vehicle inspection reports aggregated per shift, and avoid parsing failures due to oversized files |
PARSE_FILE_TIMEOUT_SECONDS | 300 seconds | Adapt to the parsing duration of real-time generated tightening data documents, and avoid interrupting production-related document processing processes due to timeouts |
chunkSettingMode | Manual configuration | For structured documents associated with VIN, customize chunking rules to preserve core field associations, and avoid automatic chunking disrupting business logic |
Three common configuration mistakes
- Phenomenon: Calling the document parsing tool returns "plugin not loaded" or the interface returns a 504 timeout status code. Cause: The associated dependencies of
customPdfParseare not correctly configured, or the deployed parsing image version does not match the FastGPT version. - Phenomenon: The VIN field and corresponding part configuration content in the parsed document are split into different chunks, resulting in failure to associate during subsequent retrieval. Cause: The
chunkSplitterrule is set to split by character, and the binding relationship between chapters and titles is not preserved. - Phenomenon: Parsing tasks terminate early when uploading large shift aggregation reports. Cause: The value of
PARSE_FILE_TIMEOUT_SECONDSis lower than the actual time required for document parsing, and the timeout parameter is not adjusted for batch documents.
How to confirm the configuration is correct
- Upload an off-line inspection report containing complete VIN and tightening torque data, and check whether the parsed chunks retain the complete inspection node content of a single vehicle, and do not split the associated paragraphs of VIN and corresponding configuration.
- View the parsing plugin logs in the FastGPT backend, confirm that the Marker parsing mode has been loaded normally, and there are no error messages about failed formula or chart parsing.
- Adjust the
chunkSizeparameter to 600 characters and 1500 characters, upload test documents separately, and compare whether the chunking results conform to the expected chapter splitting logic. - Upload a single shift aggregation report larger than 400 MB, verify that the configurations of
UPLOAD_FILE_MAX_SIZEandPARSE_FILE_TIMEOUT_SECONDStake effect, and there are no upload or timeout error messages.
The values provided are common starting points and should be measured against the reader's own samples.