What this category of data looks like
Refinery and chemical production document and material drafting data comes primarily from plant DCS real-time parameter databases, process operation specification documents, safety compliance manuals, and equipment operation and maintenance ledgers. Real-time parameter data updates once per second. Process specifications and compliance documents update every seven days. Document structures include fields such as process tag numbers, temperature, pressure, and flow rate. Common units are degrees Celsius, megapascals, and cubic meters per hour. Individual document lengths range from hundreds to tens of thousands of words. Some compliance documents include unique clause numbers and version identifiers.
What these characteristics impose on citation source and traceability
The high-frequency updates of real-time parameters require citation traceability to link to the latest document versions, to avoid using expired data. Multiple fields and professional units require citations to retain original fields and units, without alteration or simplification. Long documents combined with multiple data sources require precise matching of exclusive identifiers such as process tag numbers during recall, to prevent irrelevant documents from being retrieved. Compliance requirements mandate that traceability information include document versions and release dates to meet regulatory verification needs.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
quoteMaxToken | 800–1200 token | Refinery process documents have dense parameter segments, to avoid truncating critical tag numbers and units |
responseData | Enabled | Must retain exclusive fields such as original document process tag numbers and compliance clause numbers |
detail | Full | Must display the original document's release date and version number to meet compliance traceability requirements |
| Single recall limit | Top 3 entries | Refinery document citations require precise matching to avoid interference from irrelevant documents |
| Context window limit | Calibrated based on actual testing | Adjust based on the deployed large language model's computing power to prevent timeouts |
Reference only; verify against the target FastGPT release and deployment documentation.
Three common mistakes
- Phenomenon: Returned content does not retain fields and units from the original knowledge base text, and is altered by the large language model. Cause: The
responseDataconfiguration is not enabled, so only abstract content generated by the large language model is returned. - Phenomenon: Recall results include documents unrelated to refinery processes, such as administrative office files. Cause: The knowledge base is not filtered using refinery-specific document tags, leading to recall of irrelevant data sources.
- Phenomenon: Interface responses time out or content is truncated. Cause: The
quoteMaxTokensetting exceeds the model's supported range, or the number of documents recalled in a single operation is not limited.
How to confirm the configuration is correctly set
- Initiate a query that includes a specific process tag number, and check if the returned result includes the original document's fields and units.
- View the citation source module to confirm that the knowledge base document's version number and release date are displayed.
- Adjust the
quoteMaxTokenparameter to test response speed and content completeness across different values. - Switch the bound knowledge base collection to confirm that recall results only come from the specified refinery process document library.
The values provided here are common starting points and should be measured against the reader's own samples.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.