What data for this category looks like
Data for this category comes from design documents, standard instrument environmental monitoring data, calibration software outputs, laboratory information systems, and customer range requirements. Daily documents include assembly and commissioning sheets, calibration procedures, original calibration records, error curves, metrology certificates, and software version records. Each calibration of a single device updates its exclusive data. Documents contain fixed fields: tested part model, calibration date, standard instrument number, error calculation value, and metrology traceability chain information. Some documents include measured curves with units and certificate metadata.
What constraints these characteristics impose on knowledge base retrieval and recall
This category of data has a high update frequency. Each single device is bound to exclusive calibration records. Retrieval must precisely match unique fields such as device identifiers. Documents contain mixed formats: text, curve metadata, certificate fields, and more. Associated recall for multiple field types must be supported. Metrology traceability chains are strongly linked to error values. Retrieval results must hit associated fields simultaneously to meet cross-review requirements. Customer range requirements act as dynamic inputs. Real-time association with calibration range rules in the knowledge base is required to avoid recalling documents outside customer needs.
Configuration settings
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
similarity | 0.75–0.85 | Matches strongly linked fields of calibration procedures and error records, filters low-correlation general documents |
limit | Top 8 entries | The number of calibration-related documents for a single device is limited. Returns enough results to cover associated records and procedures |
searchMode | fullTextRecall | The query examples emphasize identifiers or specialist terms; test this mode with representative data, not as a product default. |
usingReRank | Enabled | Rearranges recalled results by error value and traceability chain priority. Prioritizes returning key review basis |
datasetSearchUsingExtensionQuery | Enabled | Automatically expands customer range requirements into associated calibration range fields. Improves retrieval accuracy |
collectionIds | Specified by calibration document collection | Only retrieves exclusive collections such as assembly and commissioning sheets and calibration records. Excludes irrelevant documents |
Three common configuration errors
- Retrieval results return a large number of irrelevant general device documents. The phenomenon is that calibration records of non-target devices appear in search results. The cause is failure to specify
collectionIdsto limit the exclusive document collection. - Knowledge base splitting fails to correctly recognize actual line breaks. The phenomenon is that text blocks are incorrectly split at the "\n" character. The cause is failure to configure splitting according to actual line break characters.
- Retrieval times out or returns empty results. The phenomenon is that the interface returns a timeout status code or no matching documents. The cause is that the
limitvalue is too large, and unreasonable configuration of the retrieval range leads to context overrun.
How to confirm the configuration is correct
- Enter the unique number of a single device. Check whether the retrieval results only include calibration records and associated documents of that device.
- Test a text fragment containing actual line breaks. Confirm that split text blocks are not interrupted by the "\n" string.
- Verify that multiple accounts can access the same knowledge base collection. Confirm that collaborative sharing configuration is in effect.
- Call the retrieval interface. Check that the number of returned results matches the
limitvalue. - Check that the knowledge base ID returned by the retrieval interface matches the target calibration document collection ID. Confirm that
collectionIdshas been correctly specified.
The values provided are common starting points and should be measured against the reader's own samples.