What Data for This Category Looks Like
Data for this category is generated by manufacturing enterprise change management teams. Sources include PLM, ERP, MES, supplier materials, validation tests, and customer-specific requirements. Common daily documents include ECN/ECR, change assessment reports, drawings and BOM version files, process validation reports, transition plans, inventory disposal forms, and customer approval documents. Data is updated irregularly with each change. Documents contain fields such as version numbers, material codes, process tolerances, and customer approval items. Some documents include embedded drawings and tables.
What Constraints Do These Characteristics Impose on Document Parsing and Chunking
Documents in this category have diverse types, including text, tables, and embedded drawings. Parsing workflows must retain the relational connections between all parts. Data sources are scattered, and document formats exported from different systems vary. Parsing rules must be adapted for multiple data source types. Documents used in bidding scenarios must fully present change impacts, version transition points, and customer approval information. Chunking must not break critical logical connections. Irregular updates require parsing workflows to quickly adapt to new document formats and field combinations.
Configuration Settings
The following configurations are developed for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | This category of documents contains key fields such as process parameters and material codes. Chunks that are too long will lose associated information, while chunks that are too short will damage the integrity of change logic |
chunkSplitter | Split by title + paragraph | Documents such as ECN/ECR have clear change titles and impact sections. Splitting by title preserves complete change logic units |
customPdfParse | Enabled | Documents in this category often include embedded drawings and BOM tables. Custom parsing can extract table content and image descriptions |
PARSE_FILE_TIMEOUT_SECONDS | 300–600 seconds | Parsing change documents that contain large BOM lists or multiple drawings takes a long time. This range covers most scenarios |
UPLOAD_FILE_MAX_SIZE | 1000 MB | Supports uploading large process validation reports or packaged change documents with multiple files, adapting to the scale of documents submitted by teams daily |
Three Common Configuration Errors
- After parsing an ECN document that includes embedded drawings, images are not extracted and cannot be passed to the large model. The cause is that the
customPdfParseconfiguration is not enabled. The default parsing only extracts plain text content. - When uploading a public Yuque link as a data source, parsing fails, and the interface prompts that page content cannot be obtained. The cause is that the link is not set to fully publicly accessible, or includes permission verification steps. FastGPT only supports public web links without permission restrictions.
- In the debug panel, HTTP tool parameters are correctly filled, but the target interface does not receive a request. The cause is that the configured
PARSE_FILE_TIMEOUT_SECONDSis less than the actual interface response time, or the request header does not include the authentication fields required for this category of documents.
How to Confirm the Configuration Is Correct
- Upload an ECN document that includes embedded drawings and BOM tables. Check if the parsing result includes image descriptions and complete BOM table content.
- Randomly select a chunked section from the parsed content. Confirm that it includes a complete change impact section and associated material code and version number information.
- Set
PARSE_FILE_TIMEOUT_SECONDSto 300 seconds. Upload a 500MB change document. Confirm that the parsing task does not time out and completes normally. - Call the HTTP tool and pass the parsed chunk parameters. Confirm that the target interface receives the request and completes the corresponding operation.
The values provided are common starting points and should be evaluated against samples relevant to the specific deployment.