What data looks like for this category
Typical production documents for this category include recipe sheets, water treatment records, blending records, sterilization or aseptic filling records, packaging material inspections, and finished product inspections. Data sources cover water source and raw material testing, formula systems, online sugar and acidity monitoring equipment, filling equipment, visual inspection equipment, and LIMS systems. Separate archived documents are generated for each production batch to update data. Online detection data is collected continuously. Test results for each production stage are summarized and organized per shift. Documents are stored linked to their associated production batches. Fields include raw material identifiers, test values, equipment operating parameters, test conclusions, batch numbers, and other relevant content.
What constraints these characteristics impose on model integration and configuration
Data for this category comes from multiple sources and updates frequently. It requires support for real-time synchronization and associated retrieval of multiple production document types. Each batch document includes detailed parameters across multiple production stages. The model context must cover enough associated data to enable accurate Q&A. High-speed liquid filling and aseptic control stages have strict data accuracy requirements. Answers to process parameter verification questions must exactly match technical standards. Content that deviates from specifications must not be generated. Production data involves product quality and safety. Model integration must prevent data from leaving the company’s internal network. The model calling process must comply with the company’s security specifications.
How to set the configurations
This guide is written for FastGPT v4.17.0.
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | This category’s single-batch production documents include test data across multiple production stages. At least three consecutive production batches’ associated data must be covered to avoid context overflow |
maxTokens | 4000 token | Most process Q&A are parameter verification questions. Output content does not need to be lengthy, which controls generation costs and response speed |
quoteMaxToken | 2000 token | Complete fragments of recipe sheets and sterilization records must be retained as reference basis to ensure answer traceability |
temperature | 0.1–0.3 | Process parameter Q&A must strictly match standard values to avoid content that deviates from specifications |
AIPROXY_API_ENDPOINT | Internal network address of the company’s own large model service | Must connect to the company’s internally deployed large model to ensure production data does not leave the company’s internal network |
AIPROXY_API_TOKEN | Company-customized access key | Restrict unauthorized access and secure process data and production information |
Three common errors
- Scenario: After deploying via docker-compose, a connection failure error appears when configuring
AIPROXY_API_ENDPOINTas the local large model address. Cause: The docker container did not map the corresponding port, or the local large model address was not added to the container’s internal network access whitelist. - Scenario: Reference content returned during Q&A is incomplete and cannot support process parameter verification. Cause:
quoteMaxTokenis set too small, truncating key parameter fragments from recipe sheets or sterilization records. - Scenario: Generated process recommendations do not match the company’s technical standards. Cause:
temperatureis set outside the reasonable range, leading the model to generate content that deviates from specifications.
How to confirm the configuration is correct
- Enter a recipe verification question for a specific production batch, and verify that the referenced documents in the returned result include the batch’s blending records and raw material test data.
- View FastGPT model call logs to confirm that
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENare configured correctly, with no connection failure or permission errors. - Trigger Q&A for multi-batch associated questions, and check whether the number of retrieved documents matches the batch coverage requirements of the current business.
- Enter a verification question with parameters outside the standard range, and confirm that the returned result prompts that the parameters do not comply with process specifications, with no incorrectly generated recommendation content.
The values provided here are common starting points and should be evaluated against the reader’s own samples.