What the data for this category looks like
Core documents for petroleum refining process specification and technical standard Q&A include operating procedures, production plans, material balances, distillation outlet quality reports, interlock ledgers, shift handover records, and change management documents. Data sources cover DCS/APC, LIMS, tank farm metering, crude oil evaluation, equipment management systems, and energy systems. Update frequencies include real-time, per shift, and daily. Fields include process temperature, pressure, flow rate, crude oil distillation range, product quality indicators, and other items. Corresponding engineering units are attached.
What constraints these characteristics impose on model integration and configuration
Heterogeneous multi-source data has large update frequency differences. It requires models to adapt to real-time, per-shift, and daily update rhythms during integration. This avoids calling expired process parameters. Documents contain complex logic for multi-unit linkage. A sufficient context window is needed to hold recalled multi-segment document fragments. This ensures complete process constraints are covered during Q&A. Refining data involves production safety and compliance requirements. Model calls must protect data from leakage. Internal deployed model endpoints should be prioritized for configuration. Process parameters have extremely high rigor requirements. Model-generated content must strictly match existing specifications. Arbitrary parameter adjustments must be avoided.
How to set the configuration
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Petroleum refining process documents have complex multi-unit linkage logic. Sufficient context is needed to hold recalled document fragments and user questions |
temperature | 0.1–0.3 | Q&A for process specifications must strictly match existing standards. It avoids generating non-standard process parameters and ensures production operation safety |
AIPROXY_API_ENDPOINT | Internal deployed large model API address of the enterprise | Refining data involves production safety and compliance requirements. Using internal endpoints ensures data is not leaked |
AIPROXY_API_TOKEN | Exclusive access token for enterprise internal models | It restricts unauthorized access and ensures model call security |
quoteMaxToken | 1500–2000 token | Quoted process document fragments must fully cover key operation requirements. It avoids truncation of core content |
maxTokens | 2000–3000 token | Q&A replies must include specific process steps and parameters. It should not be too long to affect readability |
Three common errors
- Phenomenon: Clicking the model test prompts a request error. Curl testing the interface returns 401 or connection timeout. Cause:
AIPROXY_API_ENDPOINTorAIPROXY_API_TOKENis not configured correctly. Parameters may contain extra spaces or line breaks. - Phenomenon: Content retrieved from the knowledge base is not organized and output by the large model. The reply only returns empty values or original document fragments. Cause:
maxContextis set too small to accommodate recalled document fragments and user questions. This prevents the model from completing context integration. - Phenomenon: Generated process recommendation parameters do not match the specifications. Obvious deviations appear. Cause:
temperatureis set too high, causing model-generated content to deviate from existing standards. Or an internal model endpoint is not enabled, resulting in calls to external non-compliant models.
How to confirm the configuration is complete
- On the model configuration page of FastGPT v4.17.0, click the test button. A normal Q&A result containing process parameters should be returned, with no error prompts.
- Upload a distillation outlet quality report document. Initiate a Q&A asking "What are the qualified indicator ranges for the current distillation outlet". Confirm that the recalled document fragments are complete and match the uploaded content.
- Adjust the
temperatureparameter to 0.5. Observe changes in the rigor of the reply. Confirm that the configuration parameters take effect. - Initiate a process question involving multi-unit linkage. Confirm that the reply includes the associated logic of multiple document segments, with no context truncation issues.
The values provided are common starting points and should be measured against the reader's own samples.