Data Overview for This Category
Data for this category originates primarily from customer-provided Gerber files, drill files, and BOMs, plus internal factory CAM system data, production equipment parameters, bath test results, AOI inspection reports, electrical testing reports, slice analysis reports, and other sources. Data updates cover every production batch, each shift, and continuous collection scenarios. Documents include multiple structured and semi-structured files such as engineering production instructions, drilling lamination parameters, and chemical analysis records. Fields cover production and quality inspection related content including hole diameter, copper thickness, inspection points, and anomaly types.
Constraints on Model Access and Configuration
The data for this category is scattered and available in diverse formats, including structured parameters, semi-structured inspection reports, and original files. This requires model access configuration to support multiple input types. Data updates occur in batches and in real time. Model call response times must match production cycles to avoid delays that disrupt workflow progress. Documents contain a large number of PCB-specific technical terms and segmented fields. Sufficient context length is needed to carry professional content and historical audit records. Anomaly issues tie to multiple production parameter links. Configuration must support model call logic that associates multi-source data to ensure audit accuracy.
Configuration Parameter Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Carries PCB technical documents, historical audit records, and multi-source inspection data to avoid truncating critical professional content |
maxTokens | 2000–3000 token | Matches the output length required for auditing single-batch PCB inspection reports, meeting content needs for anomaly annotation and root cause analysis |
quoteMaxToken | 1500 token | Limits the length of referenced source documents, focuses on current batch production and inspection data, and avoids interference from irrelevant content |
temperature | 0.1–0.3 | Maintains the rigor of audit logic, reduces random deviations in technical terms, and meets the determinism requirements of quality inspection scenarios |
| Workflow/API tool selection | auto | Automatically triggers tool calls to associate multi-source data such as drilling parameters and bath test results, improving the accuracy of anomaly root cause localization |
AIPROXY_API_ENDPOINT | Fill in according to the deployed model service address | Follow FastGPT v4.17.0 model access specifications, and adapt to the secure transmission requirements of factory intranets |
AIPROXY_API_TOKEN | Fill in according to the access key of the model service | Ensures valid identity verification for model calls, and avoids call failures caused by insufficient permissions |
Three Common Misconfigurations
- Phenomenon: After configuring on a third-party model proxy platform, FastGPT page calls return a 404 error or "model not found" message. Cause:
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare not filled correctly, or the model name on the proxy platform does not match the model identifier configured in FastGPT. - Phenomenon: The output result of a previous AI model in a workflow cannot be updated to an empty value, and the variable retains historical content. Cause: The trigger conditions for variable updates are not configured correctly, or the default cache value of the variable is not cleared, resulting in old data not being overwritten.
- Phenomenon: A locally deployed Ollama model does not work in the thinking settings, and calls return a model loading failure error. Cause:
AIPROXY_API_ENDPOINTis not pointed to the local Ollama service address, or correct model access permissions are not configured.
How to Verify Correct Configuration
- Test a single PCB drilling parameter document with a model call, check if the model output includes correct technical fields such as hole diameter and layer number, and confirm that context and output length configurations do not truncate critical content.
- Submit an AOI report with anomaly markers, observe if the model automatically associates parameters from the corresponding production link, and confirm that the tool call configuration is active.
- View the model call log, confirm that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENmatch the deployed model service, and that there are no network or permission error messages. - Test updating a workflow variable to an empty value, confirm that the variable content is cleared, and verify that the variable configuration logic works correctly.
The values provided are common starting points and should be measured against the reader's own samples.