What data for this category looks like
This category of data originates from metrology management systems, LIMS, external calibration institution certificates, standard instruments, and environmental monitoring equipment. It covers documents including gauge ledgers, calibration plans, calibration certificates, interim verification records, original test records, reference material ledgers, and deviation handling forms. Organizations update this data based on calibration cycles, single test results, or abnormal situations. Documents contain fields such as instrument number, calibration value, measurement uncertainty, traceability chain number, and environmental parameters. Units include metrology-specific units such as millimeters, newtons, milligrams, and degrees Celsius.
What constraints these characteristics impose on the tool calling and plugins workflow
The traceability requirement of data means tool calling must be linked to precise instrument traceability information. Plugins must support binding ledger interfaces of specific metrology systems to ensure query results can be traced back to original data sources. Multiple types of heterogeneous documents, such as calibration certificates and original records, mean tool calling must support multi-format parsing and adaptation. This avoids field extraction failures caused by single parsing rules. Frequent abnormal updates mean tool calling must have real-time data pulling capabilities. A reasonable timeout threshold must be set. Metrology-specific fields and units mean plugins must include built-in term mapping rules. This ensures accurate unit conversion and term accuracy during Q&A, avoiding quality judgment errors caused by unit confusion.
How to configure settings
The following configurations are based on FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
| Workflow/API tool selection | auto | Allows the model to independently judge whether to trigger tool calling, adapting to scenarios where standard clause queries and tool calling are mixed in metrology Q&A |
| Workflow/API function-call configuration | enabled | Enables tool calling capability forcibly, ensuring questions involving traceability chain queries and calibration result verification can normally trigger corresponding plugins |
| Tool calling timeout | 300–600 seconds | Metrology data pulling may involve cross-LIMS systems and external calibration institution interfaces. Reserve sufficient time to complete multi-document retrieval and parsing |
| Plugin recall result limit | Top 3 results | Metrology documents have strong uniqueness. Too many recall results will interfere with the model's judgment of precise traceability information |
| Context window length | 8000–12000 token | Must accommodate the complete content of multiple calibration certificates and ledgers, avoiding truncation of key traceability fields |
Three common misconfigurations
- Symptom: When tool calling support is enabled for a locally deployed Ollama model, calls return no tool response or an empty result. Cause: The tool calling format of some open-source models does not match FastGPT requirements. Parameter transfer for the
functionCallprotocol is not implemented correctly. - Symptom: Using a universal account KEY as the BASE_URI for a privately deployed FastGPT instance results in failed plugin calls. Cause: The universal account KEY is bound to public service resources. It cannot access privately deployed metrology systems and LIMS interfaces. Permission mismatch occurs.
- Symptom: After configuring file upload in a workflow, calling the document parsing tool returns an error. Cause: The plugin’s file path permissions are not configured correctly, or local files are not uploaded to a storage directory accessible to FastGPT. This prevents the tool from reading the target file.
How to confirm the configuration is complete
- Submit a test question with a clear traceability query, such as "Please query the calibration certificate for the micrometer with number JQ-2024-001". Confirm that the tool triggers normally and returns corresponding ledger data.
- View the tool calling log to confirm that the effective status of the
toolChoiceparameter and the number of recall results meet the configuration requirements. - Upload a calibration certificate template file, call the document parsing tool, and confirm that core metrology fields such as instrument number, calibration value, and measurement uncertainty can be extracted.
- Test abnormal scenarios, such as requesting multi-document queries that exceed the context window. Confirm that the tool calling timeout configuration can normally trigger retries or return prompt information.
The values provided are common starting points and should be assessed against samples appropriate for the specific deployment context.