What data for this category looks like
Data for this category comes from registered process documents, raw material batch ledgers, DCS control systems, weighing systems, LIMS laboratory information management systems, chromatography data systems, and QMS quality management systems. Daily retained documents include batch production records, material balance reports, reaction and purification step records, cleaning validation records, impurity profile reports, and deviation and change control documents. Data updates trigger once per production batch and each process step. Fields include process parameters, batch numbers, impurity content, yield data, and others. Some fields are structured numerical values, others are unstructured text records.
What constraints these characteristics impose on model access and configuration
The multi-source heterogeneous data structure, batch-based update rhythm, and documents containing sensitive compliance content such as impurity profiles and deviation records create multiple constraints for model access and configuration. Mixed multi-source data formats require configurations to support parsing of both structured numerical values and unstructured text. Batch-updated data requires limiting the context range of single calls to avoid interference from redundant information. The professional nature of impurity profiles and compliance documents requires the model to have accurate domain terminology understanding. Tool call configurations must adapt to process parameter query needs related to equipment inspection and maintenance, while ensuring batch-sensitive information is not leaked during data calls.
How to set configurations
This document is written based on FastGPT v4.17.0.
| Configuration Item | Suggested Value | Rationale for This Value |
|---|---|---|
maxContext | 8000–12000 token | Covers complete batch production records, impurity profile reports and other document content for a single batch, to avoid loss of critical information due to context truncation |
maxTokens | 4000–6000 token | Adapts to single interaction requirements for equipment inspection and maintenance Q&A, balancing context retention and model call limits |
| Workflow/API tool selection | auto | Supports automatic triggering of tool calls such as process parameter queries and batch data retrieval, adapting to real-time data acquisition needs for equipment maintenance |
vision | Enabled | Adapts to the parsing requirements for image materials such as equipment appearance, pipeline interfaces, and faulty components in equipment inspection scenarios |
temperature | 0.2–0.4 | Reduces random generation probability to ensure accurate and reliable results for equipment maintenance and process parameter queries |
AIPROXY_API_TOKEN | Enter the valid token for the corresponding model | Ensures valid identity verification for model access, preventing requests from being intercepted or rejected |
Three common mistakes
- Phenomenon: An error indicating file parsing failure is prompted after uploading documents such as impurity profile reports and batch production records. Cause: The length characteristics of documents for this category are not adapted, and the maximum length for single file splitting is not limited, resulting in a single file exceeding the context range supported by the model.
- Phenomenon: When using a model deployed locally via Ollama, tool calls related to equipment inspection cannot be triggered after enabling tool calls. Cause: Some open-source models have tool call protocols that differ from FastGPT’s adaptation logic, and the
toolChoiceparameter is not correctly configured for adaptive mode. - Phenomenon: A 400 error is returned when accessing the model via OneAPI. Cause:
AIPROXY_API_ENDPOINTorAIPROXY_API_TOKENis not correctly filled, resulting in failed request parameter verification.
How to confirm configuration is complete
- Upload batch production records and impurity profile reports for a single batch. Verify that the model can accurately extract core fields including batch numbers and impurity content.
- Initiate queries related to equipment inspection. Confirm that the model can automatically trigger tool calls and retrieve corresponding process parameter data.
- Upload images of equipment pipelines and faulty components. Verify that the model can correctly identify markings and fault points in the images.
- View interface call logs. Confirm that the configuration formats of
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENmeet requirements.
The values provided are common starting points and should be measured against the user’s own samples.