This configuration is formulated based on FastGPT v4.17.0.
What the data for this category looks like
Daily data for the general mechanical equipment assembly category includes assembly process cards, multi-level BOMs, torque records, gap adjustment records, commissioning reports, and factory inspection sheets. Data comes primarily from PLM/ERP systems, parts inspection databases, assembly tool collection terminals, test benches, MES systems, and customer technical agreements.
Data updates follow the rhythm of single equipment or single order. Each assembly batch has exclusive assembly parameters, inspection records, and delivery documents. Fields include material codes, torque values (unit N·m), gap values (unit mm), and test load parameters.
This category typically uses order-based assembly for multiple varieties. Final performance depends on component fit and on-site debugging. Its data characteristics differ significantly from single-part machining categories.
What constraints do these characteristics impose on model access and configuration
Multi-level BOMs and scattered assembly inspection data for this category require model access to support associated recall across multiple source data. Isolated single-record recall must be avoided.
Exclusive assembly parameters and inspection records for a single device require configuration to support isolating data context by order. Confusion between assembly data from different batches must be prevented.
Frequently updated real-time assembly data requires the model call context window to adapt to single-batch data volume. Call efficiency must be maintained to match on-site analysis real-time needs.
As a multi-variety order-based assembly category, personalized data characteristics require configuration to support flexible data source binding. This adapts to exclusive assembly documents for different orders.
How to set the configuration
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total character volume of multi-level BOMs and assembly records for a single device, covering context information of the complete assembly chain |
maxTokens | 4000 token | Limits the output length of single-round defect analysis, avoiding redundant content generation that affects troubleshooting efficiency |
quoteMaxToken | 1500–2000 token | Limits the total length of recalled assembly document fragments, ensuring only records strongly related to the current quality issue are retained |
temperature | 0.2–0.4 | Reduces randomness of generated results, ensures rigor of quality traceability conclusions, and meets deterministic requirements of engineering scenarios |
AIPROXY_API_ENDPOINT | Fill in the actual interface address according to the deployment environment | Connects to local or cloud-based large model services, ensuring connectivity for data calls |
AIPROXY_API_TOKEN | Match the access key of the deployed model | Verifies model call permissions, prevents unauthorized access from interfering with analysis processes |
Three common mistakes
- Issue: Locally deployed large model calls return connection timeout or permission errors. Assembly-related data cannot be obtained normally. Cause:
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare not configured correctly, or local model access permissions are not opened to the FastGPT service. - Issue: Large model generated results do not include URL links for specified assembly documents. Direct access to view original records is not possible. Cause: Model document reference format is not configured, or
quoteMaxTokenis set too small, truncating document metadata. - Issue: Knowledge base model is configured, but a prompt still indicates it is not configured. BOM and assembly records cannot be recalled. Cause: Data source and knowledge base binding relationship is not correctly associated, or
maxContextdoes not adapt to BOM hierarchical volume, leading to recall failure.
How to confirm the configuration is complete
- Manually trigger a quality traceability task for a specific order. Check if returned results include core assembly data such as the order’s corresponding BOM and torque records.
- Check FastGPT model call logs. Confirm
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENhave no format errors or missing characters. - Adjust the
temperatureparameter. Compare two analysis results to confirm generated content randomness meets engineering scenario rigor requirements. - Verify multi-order data isolation. Confirm assembly data from different batches is not confused or called in the same analysis.
The values provided are common starting points and should be measured against the reader's own samples.