What the data for this category looks like
Shipbuilding policy and regulatory basis data primarily comes from industry standards, construction specifications, and regulatory documents issued by the Ministry of Industry and Information Technology, the Ministry of Transport, and local maritime management agencies. Data updates trigger when new policies are released or existing versions are revised. No fixed update cycle exists. Most documents use structured, chaptered text. Fields include document number, effective date, applicable ship types, technical indicators, and compliance requirements. Units follow internationally standard shipbuilding industry conventions, such as gross tonnage, ship length in meters, and main engine power in kilowatts.
Constraints for model integration and configuration
Reference only; verify against the target FastGPT release and deployment documentation. Shipbuilding policy data has a high proportion of long text, dense professional terminology, and multiple parallel policy versions. This creates three constraints for model integration and configuration:
- Adapt models with large context windows to fully load the full content of a single policy document.
- Select embedding models that support industrial domain professional term encoding to ensure accurate matching of policy clauses.
- Configure a flexible data source synchronization mechanism to handle non-fixed policy update cycles, and avoid calling expired policy content.
How to set configurations
| Configuration Item | Recommended Range | Rationale |
|---|---|---|
maxContext | 8000–16000 token | Single shipbuilding policy documents are often thousands of characters long. Full context loading is required to accurately match policy clauses |
maxTokens | 2000–3000 token | Policy Q&A requires detailed compliance basis generation. Output length must cover complete clause references |
quoteMaxToken | 1000–1500 token | Full reference of original policy text fragments is required to avoid truncation of critical compliance requirements |
temperature | 0.1–0.3 | Policy Q&A requires strict matching with original text. Divergent generative content should be avoided |
AIPROXY_API_ENDPOINT | API address of a locally deployed VLLM (e.g., http://localhost:8000/v1) | Specify the correct service endpoint when connecting to a locally deployed large model |
AIPROXY_API_TOKEN | Access token generated by the local model service | Verify access permissions for the local model service |
Three common configuration errors
- Symptom: A 503 status code is returned when a local VLLM model is called, or a connection timeout prompt is displayed. Cause: The port path for
AIPROXY_API_ENDPOINTis not configured correctly, or the local model service is not running properly. - Symptom: Returned policy clauses are incomplete, or truncated. Cause: The
quoteMaxTokenvalue is too small, leading to truncation of referenced original policy text fragments. - Symptom: Uploaded policy documents fail to correctly match professional terms, or context loading is incomplete. Cause: An embedding model adapted for industrial domains is not selected, or the
maxContextvalue is insufficient to load full document content.
How to confirm correct configuration
- Call the health check interface of the local VLLM model to verify that the
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENconfigurations are correct. - Upload a shipbuilding policy document, initiate a compliance query, and test whether the returned context fully covers the document content to confirm that the
maxContextvalue meets requirements. - Initiate a policy query for a specific ship type, verify that the returned reference fragments are not truncated, and confirm that the
quoteMaxTokenvalue is reasonable. - Adjust the
temperatureparameter, verify that the generated answer strictly matches the original policy text with no divergent content.
The values provided above are common starting points. They should be evaluated against the specific samples and requirements of the target deployment.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.