What data does this category include?
This category’s business data covers the entire packaging workflow. Daily documents include packaging process cards, bonding and molding parameters, wafer incoming material drawings, test program versions, yield reports, and failure analysis reports. Data sources include wafer test maps, manufacturing execution systems (MES), bonding and molding equipment, ATE test machines, X-ray/SAM testing equipment, and customer specification requirements. Update cycles span each production batch, single-chip testing, and daily operation and maintenance statistics. Fields include equipment number, test bin code, failure type, parameter thresholds, and more. Units adjust dynamically based on document type.
What constraints do these characteristics impose on the tool calling and plugins workflow?
The data sources for this category are scattered, covering equipment, production systems, testing equipment, and customer requirement documents. Tool calling must support pulling and parsing multi-source heterogeneous data. Data update frequencies cover single-chip testing, batch production, and daily operation and maintenance. Tool calling must adapt to high-frequency updated data sources, and avoid using expired equipment parameters or test programs. Field structures vary significantly across different documents. For example, wafer incoming material drawings include coordinate parameters, while yield reports include statistical dimensions. Plugins must configure dedicated field mapping rules for different document types. Incorrect equipment parameters will directly lead to production failures. Tool calling must add a parameter validity check link to ensure that the called test program version and equipment threshold match the current batch.
How to configure the settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Supports flexible invocation of multi-source heterogeneous data, no manual tool type specification required |
| Workflow/API function-call configuration | Enabled | Enables structured parameter passing, supports validity checks for fields such as equipment number and test program version |
| Query Optimization | Calibrated via actual testing | Adapts to high-frequency updated production data, improves matching accuracy between queries and business documents |
| Maximum Context Window | 8000 token | Accommodates long-form content such as wafer incoming material drawings and failure analysis reports |
| Maximum Single Tool Call Limit | First 6 entries | Prevents context overflow caused by excessive data returned in a single call |
| Tool Call Retry Count | 3 times | Adapts to occasional fluctuations in device interfaces, ensures data pull success rate |
| Knowledge Base Retrieval Threshold | Calibrated via actual testing | Adapts to matching requirements for yield reports and testing data across different production batches |
Three common errors
- Phenomenon: Tool call configuration returns field format error, prompt states fields containing
$schemaare invalid. Cause: Incorrectly added a dollar-sign-prefixed schema field in tool input parameters, which violates FastGPT’s parameter validation rules. - Phenomenon: Unable to call third-party device interfaces normally, prompt shows connection failure or no permission. Cause: MCP access channel not configured correctly, device IP address and port mapping not bound, causing the plugin to fail to obtain real-time device data.
- Phenomenon: Knowledge base retrieval results do not meet expectations, or vector model switching fails. Cause: Vector model invocation parameters not set correctly in system configuration, causing the retrieval link to fail to trigger vector model invocation, or parameter configuration errors leading to model switching failure.
How to confirm successful configuration
This verification process applies to FastGPT v4.17.0.
- Navigate to the tool call configuration page, check the configuration status of
toolChoiceandfunctionCallto confirm alignment with preset values. - Initiate a simulated equipment inspection query, verify correct pulling of corresponding device parameters and historical maintenance records, confirming multi-source data call functionality operates normally.
- Submit a test program version matching query, check that the tool validates parameter validity, confirming structured invocation rules are active.
- View the knowledge base retrieval configuration, confirm the vector model switching path has been configured, adjust retrieval parameters to verify matching logic.
The values provided here are common starting points, and should be evaluated against the reader’s own samples.