What the data for this category looks like
Data for this category comes from material grades, customer technical specifications, temperature control instruments, furnace temperature uniformity testing equipment, hardness testers and metallographic equipment, and MES systems. Data is updated per furnace run and per batch, and includes documents such as heat treatment process cards, furnace loading diagrams, furnace temperature curves, hardness and metallographic reports, equipment inspection forms, and exception handling forms. Structured documents include fields like furnace number, batch number, temperature values, time intervals, and hardness values. Unstructured documents include visual furnace temperature curves and metallographic micrographs. Field units include degrees Celsius, hours, HV hardness units, and other standard units.
What constraints these characteristics impose on the "model access and configuration" workflow
The update frequency and document volume of single furnace run data require the model context length to adapt to the full associated data of a single heat treatment batch, preventing context overflow. Multiple types of documents include structured tables and unstructured visual content, which requires the configuration to support multimodal parsing capabilities. Most data sources are internal enterprise equipment and systems, which requires model access to adapt to internal enterprise authentication rules and network environments. It also requires ensuring the accuracy of professional term parsing, to avoid analysis deviations caused by field recognition errors.
How to set the configurations
| Configuration Item | Recommended Setting | Rationale for This Setting |
|---|---|---|
AIPROXY_API_ENDPOINT | Full API path of the large model service deployed within the enterprise | Adapts to internal enterprise model call requirements for heat treatment scenarios, avoids public network transmission latency |
AIPROXY_API_TOKEN | Exclusive authentication token assigned by the internal enterprise model | Ensures the security of internal model calls, complies with enterprise data compliance requirements |
vision | Enabled | Supports parsing unstructured visual documents such as furnace temperature curves and metallographic micrographs, adapts to parsing needs for multiple types of business data |
maxContext | 8000–12000 token | Covers the context length of all associated documents for a single furnace run, including process cards, furnace loading diagrams, and furnace temperature curves |
temperature | 0.2–0.4 | Reduces random generation probability, ensures the consistency and professionalism of defect analysis results |
quoteMaxToken | 1500–2000 token | Limits the length of recalled context fragments, avoids redundant information interfering with model analysis |
This document applies to FastGPT v4.17.0.
Three common configuration mistakes
- Occasional empty responses or errors occurring 10 seconds after a call gets stuck. Incorrect configuration of
AIPROXY_API_TOKENorAIPROXY_API_ENDPOINTcauses internal enterprise model call failures, and logs are not generated normally. - A 401 error is prompted when accessing the vector model. The authentication token is configured incorrectly and does not match the internal model's authentication rules, causing the request to be blocked.
- The deployed OneAPI service restarts indefinitely. The permissions of the configuration file mounted by the container are insufficient, or the internal model service pointed to by
AIPROXY_API_ENDPOINTis not started normally.
How to confirm the configuration is complete
- Call the test interface of the internal enterprise model to verify that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare configured correctly, and return normal model responses. - Upload the heat treatment process card and furnace temperature curve for a single furnace run to verify that the
visionconfiguration takes effect, and the model can correctly parse the visual document content. - Trigger a batch analysis task to verify that the
maxContextconfiguration adapts to the single furnace run data volume, and no context overflow error occurs. - Adjust the
temperatureparameter to verify the consistency of the analysis results, with no obvious random deviations.
The values provided are common starting points and should be measured against the reader's own samples.