This page is written based on the configuration rules for FastGPT v4.17.0.
What the data for this category looks like
Core data sources for this category include wafer test maps, MES systems, bonding and molding equipment, ATE testers, X-ray/SAM testing equipment, and customer specification documents. Data update frequencies cover per production batch, per chip test cycle, and daily device synchronization.
Daily documents include packaging process cards, bonding and molding parameter tables, incoming wafer maps, test program version lists, yield reports, and failure analysis reports. Fields include batch numbers, process temperature and pressure parameters, test bin codes, equipment numbers, yield values, and more. Some documents include visual wafer distribution maps and failure point markers.
Constraints for Document Parsing and Chunking
Multi-source heterogeneous data sources and visual document formats require parsing to adapt to both structured parameter tables and unstructured content such as wafer distribution maps and failure analysis reports. Separate text and image parsing rules must be configured.
High-frequency batch-level data updates require chunk granularity to match the time span of production batches. This prevents mixing of data across batches.
Professional fields include process units such as degrees Celsius and megapascals. Chunking must retain context binding between parameters and units. This prevents parameter and unit mismatches during retrieval.
Bid documents must align with customer specifications and internal process parameters. Chunking must retain cross-document association indexes. This ensures complete parameter chains for corresponding batches can be quickly matched during retrieval.
Configuration Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Adapts to the single-paragraph parameter length of packaging process cards and yield reports, avoids breaking the association between process steps and test results |
chunkSplitter | By paragraph + semantic boundary | Adapts to structured paragraphs of bonding parameters and failure analysis reports, avoids splitting professional term combinations |
customPdfParse | Enabled | Adapts to image content in scanned incoming wafer maps and failure analysis reports, calls built-in OCR parsing |
PARSE_FILE_TIMEOUT_SECONDS | 120 seconds | Covers parsing time for large yield reports and multi-batch packaging process cards, prevents timeout errors |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Accommodates bulk uploads of multiple device logs and test reports for a single batch |
chunkSettingMode | Manual configuration | Adjusts chunking rules for different document types (parameter tables, visual images, reports) to meet multi-document integration needs for bid documents |
Three Common Mistakes
- After uploading scanned incoming wafer maps or failure analysis reports, the parsing result contains no text content. The cause is that the image parsing configuration for
customPdfParseis not enabled, and only plain text PDF content is recognized. - When uploading a bulk document containing multiple device logs, the platform returns a timeout error. The cause is that the duration set for
PARSE_FILE_TIMEOUT_SECONDSis insufficient to cover the overall parsing time of multiple documents. - Duplicate matches or complete missing of a segment of process parameters occurs during bid document retrieval. The cause is that the
chunkSizesetting does not match the professional paragraph length of the document, leading to incorrect splitting or merging of parameters and their context.
How to Confirm the Configuration Is Correct
- Upload a scanned incoming wafer map, check whether the parsing result includes the wafer distribution markers and text annotations in the image.
- Upload a yield report of around 100 MB, wait for parsing to complete and check for timeout errors.
- Extract a segment of bonding parameter paragraph, manually adjust
chunkSizeand compare the chunking results, confirm that parameters and units are not split. - Check the platform's file upload limit settings, confirm that
UPLOAD_FILE_MAX_SIZEmatches the daily uploaded document size.
The values provided here are common starting points, and should be evaluated against the samples used in actual deployment.