What the data for this category looks like
Business data for SMT mounting and electronic assembly includes multiple types of documents and real-time collected information. Daily documents include BOM and coordinate files, line change lists, stencil and furnace temperature records, first article confirmation sheets, SPI/AOI reports, maintenance records, and traceability labels. Data sources include BOM, Gerber files, and coordinate data delivered by customers, as well as MES systems, solder paste and component batch information, placement machines, SPI/AOI testing equipment, and reflow oven temperature data. Data updates occur per work order, each line change, and real-time collection. Fields include part number, reference designator, substitute material, furnace temperature curve parameters, defect type, and other professional information.
What constraints these characteristics impose on model access and configuration
Data is updated in real time and aggregated by work order. This requires the model context window to adapt to the total length of multiple documents within a single work order, to avoid context overflow that causes call failures. Data mixes structured numerical information such as coordinates and furnace temperature curves, visual data such as SPI/AOI defect images, and text data such as maintenance records. This requires model access to support multimodal input and adapt to different types of business data formats. Professional fields such as part number, reference designator, and defect type must maintain terminology accuracy. This requires the model to control random deviation during generation. The data structure of different work orders varies greatly. This requires flexible adjustment of model output length to match the Q&A needs of different scenarios.
How to set the configurations
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total length of multiple documents such as BOM, line change list, and SPI report included in a single work order, avoiding context overflow |
maxTokens | 2000–3000 token | Controls the reply length of maintenance Q&A, ensuring complete information including defect analysis and maintenance steps |
vision | Enabled | Supports processing visual inspection data such as SPI/AOI defect images and stencil photos, matching the image analysis requirements in business |
temperature | 0.3–0.5 | Reduces random generation probability, ensuring the accuracy of professional terms such as part number and reference designator |
AIPROXY_API_ENDPOINT | Fill in according to the deployed large model service address | Connects to official service endpoints of models such as Zhipu AI and Qianwen 2, adapting to multi-channel model access |
AIPROXY_API_TOKEN | Key obtained from the corresponding model platform | Completes model identity verification, ensuring normal interface call permissions |
Three common mistakes
- Symptom: After adding Zhipu AI or Qianwen 2 models via the OneAPI channel, no corresponding model options appear when creating a knowledge base or application. Cause:
AIPROXY_API_ENDPOINTorAIPROXY_API_TOKENis configured incorrectly, and the interface address and key of the channel are not properly connected. - Symptom: A rerank error is prompted when accessing the reranking model offline, and the log contains an empty or abnormally formatted detail field. Cause: A reranking model adapted to the text features of the electronic assembly industry is not selected, or the model deployment path does not meet the loading requirements of FastGPT v4.17.0.
- Symptom: When testing the visual model, the uploaded SPI/AOI defect image displays normally, but the interaction button has no response, and the log shows interface call timeout. Cause: The
maxContextconfiguration exceeds the window upper limit supported by the model, causing a timeout when the model processes multimodal input.
How to confirm the configuration is complete
- Enter the model channel management page of FastGPT v4.17.0, confirm that added models such as Zhipu AI and Qianwen 2 are normally displayed in the optional model list.
- Upload the SPI defect image of a single SMT work order, trigger the visual model test, confirm that the interaction button can normally trigger interface calls without lag or unresponsive situations.
- Enter a maintenance question containing part number and reference designator, check the model reply content, confirm that there are no obvious errors in professional terms, and the reply length meets expectations.
- Check the interface call log, confirm that there are no format errors in the configuration of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN, and no 401 or 404 type error reports.
The values provided are common starting points and should be measured against the reader's own samples.