What data for this category looks like
This category of data includes documents such as tooling requirement documents, positioning benchmark schemes, fixture drawings, gauge operation manuals, acceptance reports, calibration records, and change histories. Data sources include product digital models and GD&T, process routes, coordinate measuring machine detection data, on-site trial feedback, and metrology systems. Data updates are performed according to project milestones, product change milestones, and regular calibration cycles. Single-project data covers fields such as positioning and clamping schemes, interference verification records, and repeatability and reproducibility reports. There is no fixed uniform single-item data volume.
What constraints do these characteristics impose on model access and configuration
The multi-source and heterogeneous document types of this category require model access to adapt to input parsing for different formats, and separately handle mixed input of structured content such as GD&T tolerance parameters and coordinate measuring machine numerical detection data and unstructured documents. According to the update rhythm of projects, product changes, and regular calibrations, the model context window must support loading full historical data of a single project on demand to avoid interference from cross-project data. The mixed data structure of metrology systems and on-site feedback requires distinguishing the processing logic of structured numerical fields and unstructured text during configuration, while adapting to the update verification rules for periodic calibration data. Fields dedicated to single projects such as positioning and clamping schemes and interference verification records require presetting extraction and matching rules for corresponding fields during model access to ensure the accuracy of ledger summary fields.
How to set configurations
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
AIPROXY_API_ENDPOINT | Local Ollama-deployed DeepSeek model access address, for example http://192.168.1.100:11434/v1 | Adapts to the API access format of locally deployed models, and meets the requirements of FastGPT v4.17.0 for connecting local models |
AIPROXY_API_TOKEN | Access key generated during local model deployment, fill in null when no authentication is required | Meets the identity verification requirements of local models, and adapts to permission configurations for different deployment environments |
maxContext | 8000–12000 token | Adapts to the total context length of single-project multi-documents in this category, avoiding truncation of critical GD&T parameters and calibration records |
temperature | 0.2–0.4 | Ensures the accuracy and format consistency of ledger summary fields, reducing random output deviations in unstructured text |
maxTokens | 2000–3000 token | Adapts to the output length of a single report generation, covering the summary requirements of content such as fixture drawing descriptions and acceptance reports |
vision | Enabled | Adapts to the parsing requirements of image documents such as fixture drawings and gauge operation manuals included in this category |
Three common mistakes
- Symptom: After adding a locally Ollama-deployed DeepSeek model, the test interface returns a 500 error. Cause: The path suffix of
AIPROXY_API_ENDPOINTis not configured correctly, or the port address is filled incorrectly, causing the API request to fail to match the model service. - Symptom: The generated ledger summary result is missing coordinate measuring machine detection data or GD&T parameter fields. Cause: The
temperaturevalue is set too high, causing the model to randomly omit structured fields during extraction. - Symptom: Context truncation occurs when summarizing full data for a single project, and critical change history content is lost. Cause: The
maxContextvalue does not adapt to the total document length of the single project, and the set value is too small, resulting in insufficient context window.
How to confirm successful configuration
- Access the configured
AIPROXY_API_ENDPOINTaddress to confirm that the basic interface information of the model can be returned, with no network or path errors. - Upload a single tooling requirement document and coordinate measuring machine detection data to trigger the ledger summary task, and check whether core business fields are completely extracted.
- Adjust the
temperaturevalue to0.1and0.5, compare the consistency of the two summary results, and confirm that the parameter configuration takes effect. - Generate a single complete production report, check whether the output content covers all required document modules, with no abnormal truncation or overflow.
The values given are common starting points and should be measured against the reader's own samples.