What this category of data looks like
Daily documents for semiconductor packaging and testing include 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 specifications. Data is updated per wafer batch, per individual test result, and on a daily summary basis. Document fields cover batch numbers, parameter thresholds, bin codes, failure types, and professional units such as voltage, current, and packaging thickness.
What constraints these characteristics impose on the model access and configuration link
This category of data is multi-source, heterogeneous, and updated at high frequency. The model access link must adapt to the output formats of different devices and systems to avoid parsing failures caused by format incompatibility. The documents contain a large number of professional process parameters and failure codes. Configuration must reserve space for custom field mapping to ensure the model can accurately recognize professional terms. The fine-grained update rhythm (per batch, per test, daily) requires controlling the model context length to avoid redundant data occupying too many token quotas. Additionally, the need for associated analysis of process documents and exception reports requires the configuration link to support multi-document associated calls, and model call parameters must be set appropriately.
Configuration Recommendations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Adapts to the total length of single-batch process documents and exception reports for this category, avoiding context overflow |
maxTokens | 2000–3000 token | Controls the length of single-round drafting output, matching the conventional writing requirements of process documents |
quoteMaxToken | 1500–2000 characters | Retains sufficient source document fragments to align wafer test maps and test results, avoiding context overload |
temperature | 0.3–0.5 | Ensures the accuracy of process parameters and failure descriptions, reducing the probability of random output |
AIPROXY_API_ENDPOINT | Exclusive access address provided by the vendor | Adapts to data format conversion for multi-source devices and systems, solving cross-data-source access issues |
AIPROXY_API_TOKEN | Bound exclusive access token | Ensures the security and permission controllability of model access |
Three Common Configuration Mistakes
- Phenomenon: An interface pop-up displays the error message "No available channel for model qwq:latest under current group default". Cause: The access permission for the corresponding model is not bound in the model channel configuration, or
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare configured incorrectly, preventing retrieval of a valid channel list. - Phenomenon: Professional parameter fields are missing from the process document fragments returned by model calls. Cause:
quoteMaxTokenis set too small, failing to retain sufficient source document fragments for extracting packaging parameters and test bin codes. - Phenomenon: Local models deployed via ollama cannot be called normally, and port 500 is occupied. Cause: Other services occupying this port have not been closed, or the access address of the local model is not correctly specified in FastGPT configuration, resulting in connection failure.
How to Confirm Successful Configuration
- Enter the model channel management page of FastGPT v4.17.0, verify that the configurations of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENmatch the information provided by the vendor. - Initiate a test call using a packaging process card and yield report as input, check whether the returned process document fragments contain the unique professional fields and parameters of this category.
- View the model call logs to confirm that the
maxContextandmaxTokensparameters do not trigger context overflow or output truncation errors. - For locally deployed models, check port occupancy, confirm that the access address and port configuration match the deployment environment.
The values provided here are common starting points and should be evaluated against corresponding samples.