What this category's data looks like
Daily documents for this category include inspection specifications, sampling plans, incoming material reports, supplier audit reports, exception liaison sheets, 8D reports, concession acceptance approval forms, and more. Data sources include purchase orders, drawings and specifications, supplier quality certificates, inspection equipment data, historical quality records, and supplier portal information. Update frequency aligns with each incoming material batch, monthly or quarterly supplier evaluations, and irregular audits. Documents include fixed fields such as batch number, supplier ID, non-conformance description, approval status, and others. Some drawing documents have annotation information involving units such as parts, batches, and days.
What constraints do these characteristics impose on the "document parsing and chunking" link?
Documents in this category include multi-dimensional associated fields such as batch number and supplier ID. Chunking must retain context association information to avoid inability to correspond to inspection or approval processes after splitting. Different document types have significant structural differences: inspection specifications use chapter-based structures, and 8D reports use step-by-step processes. Different chunking logic must be adapted to retain business integrity. Data sources cover multi-format files, including PDF-format quality certificates, CSV-format historical quality data, and Excel-format purchase orders. Support for multi-format parsing and adaptation is required. Large documents summarizing single batches may appear in bulk incoming material scenarios. Parsing and chunking processing for large files must be supported. Update frequency is high, so adaptation is required for frequently updated incoming material data.
Configuration Settings
This configuration is formulated based on the logic of FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
chunkSize | 800–1200 characters | Documents in this category include short fields such as batch number and non-conformance item. Excessively large chunks will cause associated information to be scattered |
chunkSplitter | Separate by title + paragraph | Documents in this category have structured layouts such as chapter-based and step-by-step formats. Separating by title can retain business logic integrity |
customPdfParse | Enabled | Some documents in this category include special layouts such as drawing annotations and stamped pages. Custom parsing can retain core information |
UPLOAD_FILE_MAX_SIZE | 500 MB | Bulk incoming material summary inspection reports may reach large sizes, adapting to bulk upload scenarios |
PARSE_FILE_TIMEOUT_SECONDS | 600 seconds | Parsing large supplier audit reports requires long processing time to avoid mid-process interruptions |
chunkSettingMode | Manual configuration | Documents in this category have significant structural differences. Manual configuration can adapt to different types of business documents |
Three Common Mistakes
- Phenomenon: After upgrading the version, when uploading CSV-format historical quality data, the system prompts an error about field format mismatch. Cause: The default chunking strategy of the new version has been adjusted, and it does not adapt to the format of non-standard fields such as batch number and supplier ID in CSV files of this category.
- Phenomenon: After enabling
customPdfParse, annotation information still cannot be extracted from drawing documents. Cause: No custom parsing rules have been uploaded, or the rules do not cover the special layout structure of drawing documents. - Phenomenon: Uploaded files are not directly passed to the large model node, but are parsed into text by the system before being passed in. Cause: The file pass-through configuration of the custom node has not been enabled, or the file processing node has not been correctly bound.
How to Confirm the Configuration is Correct
- Upload a standard incoming material inspection report PDF, and check whether the parsed text retains core fields such as batch number and inspection items.
- Adjust the
chunkSizeparameter, and check whether the length of the chunking result falls within the preset 800–1200 character range. - Upload a supplier audit report of approximately 100 MB, and confirm that the parsing process does not trigger a timeout error.
- Upload CSV-format historical quality data, and confirm that associated fields such as supplier ID and non-conformance description are not lost after chunking.
The values provided are common starting points and should be measured against the reader's own samples.