What this category of data looks like
Heat treatment scenario data sources include material grade standards, customer technical requirements, furnace temperature control instruments, furnace temperature uniformity test records, hardness tester and metallographic equipment test reports, and workshop MES systems. Data update frequency is per heat load, per batch. Daily documents include heat treatment process cards, furnace loading layout drawings, furnace temperature curve charts, hardness and metallographic analysis reports, equipment inspection forms, and exception handling sheets. Fields include furnace temperature (℃), holding time (hours), cooling rate (℃/min), hardness value (HV), metallographic structure type, and others. Some documents include process-related image materials.
What constraints these characteristics impose on the "model access and configuration" link
Heat treatment scenarios have high data update frequency. Single heat load data must be included in context in real time. Configuration must adapt to high-frequency document indexing and updates. Documents contain structured process parameters and unstructured images and curves. Models must support multimodal content recognition and structured information extraction. Professional fields use clear engineering units. Q&A must strictly match predefined parameters. The model's free generation space must be limited. Some data relates to part service safety. Q&A results must be traceable back to original documents. Sufficient context reference length must be configured.
How to set the configuration
This configuration applies to FastGPT v4.17.0.
| Configuration Item | Recommended Value Range | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Heat treatment scenarios need to accommodate the full content of multiple heat load process cards, furnace temperature curves and other documents to avoid truncation of critical context |
quoteMaxToken | 2000–3000 token | Need to fully reference the core parameters of a single process document to ensure Q&A results are traceable to original data |
temperature | 0.1–0.3 | Process specification and standard Q&A must strictly match predefined parameters, reducing deviations from free model generation |
vision | Enabled | Heat treatment documents include furnace loading layout drawings, metallographic micrographs and other image materials, requiring models to identify process information in images |
AIPROXY_API_ENDPOINT | Exclusive API address of the deployment node | Enables targeted model calls, avoiding interference from global configurations |
AIPROXY_API_TOKEN | Access key for the corresponding model | Ensures permission security for API calls and prevents unauthorized access |
Three common mistakes
- Phenomenon: After saving the model configuration, chat automatically switches to an unspecified large model. Cause: The
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENconfigured at the model level are not locked at the application level, causing the global default model configuration to be used instead. - Phenomenon: Processing takes too long when uploading image documents such as metallographic reports and furnace temperature curves. Cause: No local preprocessing acceleration node is configured, or unnecessary multimodal preprocessing processes are enabled, increasing document parsing load.
- Phenomenon: Q&A results have deviations that do not match actual process parameters. Cause: The
temperaturevalue is set too high, causing the model to generate content that deviates from predefined standards and fails to accurately match heat treatment process requirements.
How to confirm the configuration is complete
- Upload a standard heat treatment process card and corresponding furnace temperature curve, initiate a query asking "Confirm the heating and holding system for this heat load", and check whether the returned result fully references the key parameters in the documents.
- View the model call logs in the FastGPT application, and confirm that the called API endpoint exactly matches the configured
AIPROXY_API_ENDPOINT. - Upload a metallographic micrograph, initiate a query asking "Describe the metallographic structure type of this sample", and check whether the model can accurately identify the structural features in the image.
- Adjust the
temperaturevalue to 0.2, initiate three identical process Q&A queries, and check whether the consistency of the returned results meets expectations.
The values provided are common starting points and should be measured against the reader's own samples.