What the data for this category looks like
Data sources for this category include wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM inspection equipment, and customer specifications. Updates occur per production batch, single-unit test result, and daily summary. Daily documents include packaging process cards, bonding and molding parameter sheets, incoming wafer maps, test program version files, yield analysis reports, and failure analysis reports. Data fields include batch identifiers, parameter thresholds, test bin codes, yield values, and units for metrics such as packaging temperature, pressure, test voltage, and current.
What constraints these characteristics impose on the model access and configuration link
This category has high-frequency, fine-grained data updates. Model access must support real-time pulling of the latest batch test data and associated parameters to avoid expired context. Multiple document types coexist. Configured models must parse structured parameter sheets, visual wafer maps, and text reports. Equipment and system sources vary widely. Access configuration must support multi-endpoint API adaptation. Batch association rules are strictly required. Configuration must clearly define context batch identifier field association rules to prevent data misalignment that reduces audit accuracy.
How to set the configuration
The following configuration items are set based on FastGPT v4.17.0.
| Configuration Item | Suggested Method | Rationale |
|---|---|---|
AIPROXY_API_ENDPOINT | Obtain the corresponding endpoint from the official interface documentation of the enterprise MES system or ATE test machine | This parameter specifies the target service address for model calls, must match the API specifications of the equipment or system, and adapt to the multi-source data access requirements of semiconductor packaging and testing |
AIPROXY_API_TOKEN | Obtain a dedicated access token from the enterprise internal API gateway | This parameter is used for identity authentication, must comply with the secure access requirements of equipment and systems, and prevent unauthorized access to production data |
maxContext | 8000–12000 token | This category has a large amount of batch data association information, needs to accommodate context content from multiple test reports and parameter sheets, and adapt to the audit requirements of multi-source data |
maxTokens | 2000–3000 token | The output requirements for single-batch failure analysis and yield analysis are clear, need to control the length of generated content to avoid redundancy, and comply with the format specifications of audit documents |
vision | Enabled | Need to parse visual data such as incoming wafer maps and X-ray inspection images, and adapt to the visual verification and failure analysis requirements of packaging and testing |
temperature | 0.2–0.4 | Parameter configuration and quality audit require rigor, avoid generating uncertain analysis results, and ensure the consistency of audit conclusions |
Three Common Mistakes
- Phenomenon: A 401 unauthorized status code is returned when calling the model, or the interface cannot pull test data normally. Cause: The valid values of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare not obtained correctly, and unofficial endpoints or expired access credentials are misused. - Phenomenon: The model cannot parse visual data such as incoming wafer maps and X-ray inspection images, and the output content only includes text descriptions. Cause: The
visionconfiguration item is not enabled, so the model does not turn on visual parsing capabilities. - Phenomenon: When using a third-party deployed large model, GPU usage remains low and single-core CPU usage reaches 100%. Cause: The access configuration does not bind the GPU device identifier, so the model runs in CPU mode by default and does not fully utilize hardware resources.
How to confirm configuration is complete
- Perform a model call test, upload packaging parameters and test data for a specified batch, verify that audit content can be normally pulled and generated, and confirm that the
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENconfigurations are valid. - Upload an incoming wafer map to the model input, verify that the wafer partition and identification information in the image can be correctly parsed, and confirm that the
visionconfiguration item is effective. - Adjust the
temperatureparameter to different ranges, observe the rigor of the analysis report generated by the model, and confirm that the parameter value meets the accuracy requirements of business audits. - Check the model running logs, confirm that the batch identifier loaded in the context matches the actual production batch, and verify that the
maxContextconfiguration adapts to the multi-source data association requirements.
The values provided here are common starting points and should be measured against the reader's own samples.