What the data for this category looks like
This category of data includes documents such as packaging process cards, bonding and molding parameters, wafer incoming material maps, test program versions, yield reports, and failure analysis reports. Data sources include wafer test maps, MES systems, bonding and molding equipment, ATE test machines, X-ray/SAM testing equipment, and customer specification requirements. Data is updated per production batch, per chip test node, and daily. Fields include batch number, test bin value, bonding temperature, molding pressure, failure type, and others. There are no unified fixed units; values vary based on equipment parameters and test items.
What constraints this category imposes on model access and configuration
The multi-source heterogeneous data sources and high-frequency update rhythm of this category require model access configuration to support multi-document splicing and real-time data calls. Fields without unified units such as batch number and test bin value require the configuration to support the model in identifying equipment parameters with units and associating corresponding test records. High-frequency single-chip test and daily updated data require the response time of model calls to match production beats, avoiding delays that affect production line processes. At the same time, the association requirements between wafer maps and physical batches must be aligned. The configuration must ensure that knowledge base retrieval accurately matches full-link documents associated with batches. The configuration on this page is based on FastGPT v4.17.0.
How to set the configuration
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the context length requirements of multi-document splicing such as packaging process cards and failure analysis reports |
maxTokens | 4000 token | Matches the output length limit for single-chip test data and failure analysis |
quoteMaxToken | 2000 token | Ensures that retrieved document fragments such as wafer test maps and ATE test records do not exceed the context quota |
temperature | 0.2–0.4 | Reduces output randomness and ensures consistency of quality traceability analysis conclusions |
AIPROXY_API_ENDPOINT | Fill in according to the actual address of the deployment node | Adapts to proxy access configuration for enterprise internal networks |
AIPROXY_API_TOKEN | Fill in according to the authorization credential | Completes identity verification for the model channel |
Three common configuration mistakes
- Phenomenon: A proxy connection timeout error is returned when calling the model, with status code 502. Cause:
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare not configured correctly, or the corresponding port is not open on the proxy node. - Phenomenon: The model return content includes complete prompt thinking process fragments. Cause: The model's internal thinking output switch is not turned off as required, resulting in redundant content mixed into the return result.
- Phenomenon: Random failures occur when calling the model channel, with no fixed error message. Cause: Reasonable load balancing rules are not configured, causing a single node to be overloaded and unable to respond to requests.
How to confirm the configuration is complete
- After filling in
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN, initiate a test call to confirm that a normal quality traceability analysis result is returned and no errors occur. - Upload a packaging process card and the corresponding batch test data to verify that the number of document fragments retrieved by the model meets the configuration requirements.
- Set
temperatureto 0.3, then call the same test prompt twice to confirm that the difference in output results is within an acceptable range. - After configuring
maxContext, splice multiple long documents and initiate a call to confirm that no context overflow error occurs.
The values given are common starting points and should be measured against the reader's own samples.