What the data for this category looks like
Shipbuilding business processing data primarily comes from manufacturing execution systems, classification society inspection archives, and order management systems. Data is updated according to production milestones such as block construction and sea trials. Individual business documents include fields including order number, block identification code, process name, inspection specifications, man-hour quota, and others. Field units cover millimeters, tons, hours, inspection batch numbers, and more. Document length varies widely, ranging from process notices with hundreds of characters to full-vessel inspection reports with tens of thousands of characters.
What constraints these characteristics impose on the multi-turn dialogue and prompt engineering link
The wide range of document lengths, numerous specialized fields, and specialized units in shipbuilding business data require multi-turn dialogue context management to filter historical information not associated with the current order, to avoid cross-order confusion. Long documents such as full-vessel inspection reports require limiting the effective length of context recall, to prevent exceeding the model's context limit. Unified explanations for specialized units such as millimeters and tons must be embedded in prompts, to avoid the model making misjudgments of process parameters. Data updated according to production milestones requires that dialogue contexts be bound to the current block or process identifier, ensuring that each round of interaction context only associates with the current business link.
How to set the configuration
Reference only; verify against the target FastGPT release and deployment documentation.
| Configuration Item | Recommended Approach | Rationale |
|---|---|---|
chatId | Bind order number + block identification code | Ensure multi-turn dialogue only associates with the current business node, avoid cross-order confusion |
messages | Limit context length to 8000–12000 characters | Adapt to context recall requirements for long shipbuilding documents, avoid truncating critical inspection parameters |
variables | Pass current process name, unit descriptions | Align with specialized fields and units of shipbuilding business, reduce parameter misjudgments |
| Context recall count | Top 3–5 entries | Filter redundant historical dialogue, only retain key process information for the current order |
| Model adaptation configuration | Enable general large model support | Adapt to processing requirements for full-vessel inspection reports with tens of thousands of characters |
Three common mistakes
- Phenomenon: Dialogue context is lost, and subsequent interactions cannot associate previously mentioned order block information. Cause:
chatIdis not correctly configured to bind specific business identifiers, causing the session to be reset by the system. - Phenomenon: The model selection dropdown only displays GPT series models, and GLM series models configured via the interface cannot be selected. Cause: General model adaptation is not enabled in the model management of the dialogue module, and only interface keys for a specified series are bound.
- Phenomenon: After uploading images and process documents, the dialogue cannot correctly parse mixed-format input content. Cause: Adaptation rules for multi-format input are not enabled in system configuration, and only text-type input is supported.
How to confirm the configuration is correct
- Initiate a first-round dialogue containing an order number and block number, then initiate a second-round process question associated with that block, and check if the dialogue can correctly associate context information.
- Enter the model management page, and check if the added general large models appear in the model selection dropdown of the dialogue module.
- Upload a shipbuilding process notice and full-vessel inspection report, initiate multi-round interactions, and check if the model can correctly identify specialized units and field parameters.
- After configuring the multi-turn dialogue workflow, run test cases and check that only the results of the last dialogue node are displayed, with no redundant output from intermediate steps.
The values provided here are common starting points and should be measured against appropriate samples.
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.