What Data for This Category Looks Like
Data sources for this category include resin batch property sheets, injection molding machine operating parameters, mold temperature controller data, MES system records, first article inspection forms, material change and machine cleaning records, and mold history. Daily documents include injection molding process cards, raw material drying logs, and patrol inspection forms. Fields cover material temperature, holding pressure, cooling duration, raw material moisture content, and similar metrics. Common units are degrees Celsius, megapascals, and seconds. Data updates are triggered by production batches, mold or material changes, and shift patrols, with no fixed schedule. Every production action triggers an update to the corresponding document.
Constraints on the Model Access and Configuration Workflow
The data for this category is multi-source and scattered, with updates triggered in real time by production batches, mold or material changes, and shift patrols. This requires model access to support real-time pulling of documents from different production nodes. Individual process documents are lengthy, so the system must support long-text chunking logic. Document fields cover multiple engineering units, so the model must accurately identify and associate corresponding parameters. Document fields vary significantly across different production stages, so flexible field mapping rules must be configured to avoid parameter matching errors. Process Q&A requires cross-verification across multiple documents, so the model context capacity must support loading multiple documents simultaneously.
Configuration Settings
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Individual injection molding process documents are mostly thousands of characters long. This range accommodates the context of multiple process cards and patrol records, preventing loss of critical information during Q&A |
maxTokens | 4000 token | Matches the length of process parameter explanations and problem analysis returned in a single Q&A, preventing result truncation due to exceeding model output limits |
| Workflow/API tool selection | auto | Supports automatic calling of built-in tools to handle long-text chunking and multi-source data association, adapting to the multi-step analysis requirements of process Q&A |
AIPROXY_API_ENDPOINT | Fill in the deployed model gateway address | Adapts to the internally deployed model service address, ensuring requests are properly routed to the target model |
AIPROXY_API_TOKEN | Fill in the key assigned by the gateway | Completes request identity verification, ensuring the security and legitimacy of model access |
temperature | 0.2–0.4 | Maintains accuracy of process parameter responses, preventing generation of non-standard parameters that do not comply with injection molding specifications |
Three Common Configuration Errors
- Phenomenon: After calling the HTTP API to obtain process documents, the returned long text is truncated or a context overflow error is triggered. Cause: Long-text automatic chunking logic is not configured, and the full document is directly passed to the model, exceeding the upper limit set by
maxContext. - Phenomenon: After configuring
toolChoiceas false, the built-in text chunking and multi-segment sending functions cannot be triggered. Cause: The automatic tool selection switch is incorrectly turned off, preventing calls to the system's built-in text processing tools. - Phenomenon: A locally deployed ChatGLM2 model accessed via a container cannot be selected when creating a knowledge base. Cause: The model's access address and key are not correctly entered in the system configuration, or the model version is incompatible with FastGPT v4.17.0.
How to Verify Successful Configuration
- Send a test request with a typical injection molding process card text, check whether the model can correctly return process parameter explanations, and verify that context loading works properly.
- Set
toolChoiceto auto, send a test request containing long text, check whether the system automatically triggers the text chunking tool, and verify that the tool calling logic is active. - Verify the configured values of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKEN, send a model connectivity test, and check whether the request returns a normal response. - Adjust the value of
temperature, send multiple test requests, and check whether the consistency of responses meets the accuracy requirements for process parameters.
The values provided here are common starting points, and should be validated against the relevant test samples.