What Data for This Category Looks Like
Daily documents for this category include packaging process cards, bonding and molding parameter tables, wafer incoming material drawings, test program version files, yield ledgers, and failure analysis reports. Data sources cover wafer test maps, manufacturing execution systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specifications. Update frequency adjusts based on production batches, individual test results, and daily equipment operation data. Document fields include equipment operating parameters, batch identifiers, test classification numbers, failure location coordinates, and some parameters use specialized units such as degrees Celsius and pascals.
What Constraints These Characteristics Impose on Document Parsing and Chunking
Wafer incoming material drawings and failure analysis reports contain bitmaps and structured annotations, requiring the parsing link to support vector annotation extraction and cross-element context association. Frequently updated production batch data requires parsing to have low-latency batch processing capabilities to avoid timeouts that disrupt production workflows. Mixed-type documents (parameter tables, process text, inspection images) need chunking adapted to multiple structural features to avoid splitting associated fields. Specialized parameters and business identifiers must retain binding relationships; chunking must not truncate the complete logic of key parameters and their corresponding batches.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the field length of packaging parameter tables and process text, avoiding truncation of complete single-process logic |
chunkSplitter | Chunk by paragraph + semantics | Adapts to mixed document structures, preserving complete business logic for processes such as bonding and molding |
customPdfParse | Enable Marker parsing mode | Adapts to wafer incoming material drawings and failure analysis reports with vector annotations, improving structured extraction accuracy |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Covers parsing time for large batch inspection reports, avoiding disruption of production-related data processing due to timeouts |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports batch upload of single-batch wafer test maps and complete yield ledgers |
chunkSettingMode | Manual configuration mode | Allows adjustment of chunking rules for different document types (parameter tables, image reports) to adapt to business scenarios |
Three Common Misconfigurations
- Parsing tasks stay in a waiting state for extended periods, with the interface returning a timeout status code. This occurs when the
PARSE_FILE_TIMEOUT_SECONDSparameter is not adjusted; the default value is insufficient for processing single-batch large-format yield reports. - The file enhanced parsing function does not work correctly in version 4.9.0. Functionality is restored after switching to FastGPT v4.17.0 stable version or manually setting
customPdfParseto Marker mode. This stems from compatibility issues with custom parsing plugins in older versions. - File parsing tools cannot be invoked during conversations, with a tool call failure prompt shown on the conversation interface. This happens when file parsing permissions are not enabled in the knowledge base configuration, or the associated parsing service fails to start normally.
How to Verify Proper Configuration
- Upload a single packaging process card, review the parsed chunking results, and confirm that process paragraphs are not truncated.
- Upload a wafer incoming material drawing with vector annotations, check if the parsed results extract annotation coordinates and corresponding batch information.
- Submit a batch test report parsing task, verify that parsing time does not exceed the configured
PARSE_FILE_TIMEOUT_SECONDSthreshold. - Trigger a file parsing tool call from the conversation interface, confirm that the tool can be invoked normally and completes the parsing process.
The values provided are common starting points and should be measured against the reader's own samples.