What Data for This Category Looks Like
Core data for this category comes from manufacturers of single-use consumables, implanted/interventional devices, diagnostic equipment, and active medical devices. It includes production-related documents such as production process specifications, batch production records, clean zone records, sterilization batch records, inspection reports, software version records, adverse event materials, and traceability resources. Data sources include design history files, device and material batch information, MES systems, clean environment monitoring equipment, sterilization equipment, and inspection and calibration systems. Data updates run per batch or per product. Clean environment-related records are updated regularly per environmental monitoring plans. Document fields contain key traceability information including device unique identifiers, sterilization parameters, cleanliness levels, and batch numbers.
Constraints Imposed on Model Access and Configuration
The batch-based, highly traceable, and high-risk nature of this category's data creates clear constraints for model access and configuration. Single-batch production documents and associated records have a large total length. Sufficient context windows must be configured to carry all information completely. Data must be strictly linked to device unique identifiers and batch information. Model calls must retain full ability to reference key traceability fields. Work involving sterile assurance and patient safety risks requires model outputs to maintain rigor, and avoid arbitrary content. Local deployment scenarios require integration with internal model proxies. Correct proxy addresses and tokens must be configured to prevent internal production data from being transferred outside the environment.
How to Set Configurations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 16000–20000 token | Matches the total length of single-batch production documents and associated records, to avoid context truncation |
maxTokens | 8000–10000 token | Ensures exception reports can fully record abnormal phenomena, handling processes, and traceability information |
quoteMaxToken | 4000–6000 token | Retains complete references to key traceability fields including sterilization parameters, cleanliness levels, and device unique identifiers |
temperature | 0.1–0.3 | Reduces randomness in model outputs, to maintain rigor for process documents and exception reports |
| Workflow/API tool selection | auto | Automatically determines whether to call tools to link batch data and traceability information |
AIPROXY_API_ENDPOINT | http://192.168.x.x:3000/v1 | Integrates with locally deployed model proxies, to ensure internal production data is not transferred outside the environment |
Three Common Configuration Mistakes
- Issue: When configuring model channels for local deployment, the interface displays connection timeout or inability to access OneAPI. Cause: The actual local address and port for
AIPROXY_API_ENDPOINTwere not correctly filled in the docker compose environment variables. FastGPT cannot connect to the internal model proxy as a result. - Issue: Generated process documents or exception reports experience unexpected text truncation. Cause: The
maxTokensparameter was not adjusted to a value matching production document lengths. Content exceeding the model's output limit is automatically truncated, leading to lost key information. - Issue: Generated reports do not include key traceability information such as device unique identifiers and sterilization parameters. Cause:
quoteMaxTokenwas not configured to retain key fields in long text. The model loses some traceability data as a result.
How to Confirm Proper Configuration
- A user logs in to the FastGPT v4.17.0 admin backend, navigates to the model channel configuration page, and verifies that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENmatch the locally deployed OneAPI configuration. - A user uploads a single-batch batch production record document, initiates a process document generation test, and checks that the output content is complete with no unexpected text truncation.
- A user generates an exception report, and checks that the report includes key traceability fields such as device unique identifiers, sterilization parameters, and cleanliness levels.
- A user triggers a batch document processing task, confirms that there are no connection errors during model calls, and checks that output results comply with production document formatting and traceability requirements.
The values provided are common starting points, and should be assessed against the specific samples relevant to each deployment.