Reference only; verify against the target FastGPT release and deployment documentation.
What this category of data looks like
Document and material data for urban water utilities mainly comes from pipeline network monitoring systems, water quality testing laboratories, repair service platforms, and water plant operation logs. Data update rhythms fall into two categories: real-time data includes user repair records and pipeline network pressure fluctuation logs; batch data includes daily water quality test reports and monthly water supply scheduling plans. Most documents are structured tables and semi-structured text, with core fields including sampling point numbers, test indicator values and corresponding units (such as mg/L, MPa), repair addresses, and processing workflow nodes. Some documents also include on-site photos of equipment taken during inspections.
Constraints on Model Integration and Configuration
Document data for urban water utilities includes three forms: structured tables, semi-structured text, and on-site photos. It also includes exclusive test indicators such as COD and ammonia nitrogen, as well as dedicated units such as mg/L and MPa. These characteristics require model integration configuration to support precise mapping of structured fields, to avoid unit identification errors. Real-time updated repair and pipeline network data requires the context window to adapt to short-cycle data access, to avoid redundant old data interfering with document generation. Some documents include on-site equipment photos, so configuration must enable visual recognition capabilities to extract parameter information from images. Differences in data formats across multiple sources require access configuration to be compatible with field rules of different data sources, to avoid deviations in document content caused by field misalignment.
How to Configure Settings
| Configuration Item | Recommended Setting | Rationale |
|---|---|---|
maxContext | 8000–12000 token | Water utility documents often include multiple test reports and scheduling logs, requiring sufficient context information to ensure complete generated content |
temperature | 0.2–0.4 | Water utility documents require strict accuracy and should not be overly divergent. This range balances content accuracy and flexibility |
vision | Enabled | Some water utility documents include on-site equipment photos, requiring extraction of equipment numbers and parameter information from images |
AIPROXY_API_ENDPOINT | Fill in the dedicated access address of the corresponding model resource, such as the endpoint of an Azure OpenAI resource | Adapt to model call requirements in water utility scenarios, ensuring matching request paths and protocols |
maxTokens | 1500–2000 token | The typical generation length for a single water utility document falls within this range, avoiding overly long or short generated content |
quoteMaxToken | 4000 token | Sufficient reference data must be retained for document corroboration, ensuring generated content has reliable basis |
Three Common Configuration Errors
- A 400 status code is returned when calling the model, with the prompt "invalid api version" or "unsupported endpoint". The cause is that the dedicated access address of Azure OpenAI has not been configured in
AIPROXY_API_ENDPOINT, and the API path and version rules of Azure have not been adapted. - Unit confusion appears in generated water utility documents, for example, "mg/L" is identified as "g/L". The cause is that the model's term prompt has not been configured for water utility-specific test indicators and units, or the
temperaturevalue is set too high, causing the model to diverge. - Core fields are missing from generated documents, for example, sampling point numbers or test times are missing. The cause is that the
maxContextvalue is too small to accommodate complete data source field information.
How to Verify Proper Configuration
- Import structured data from a standard water utility water quality test report, trigger document generation, and verify that the generated content includes correct sampling point numbers, test indicators and corresponding units.
- Upload an on-site photo of a water utility pipeline network equipment, enable the
visionconfiguration, and verify that the model can accurately extract equipment numbers and parameter information from the photo. - After configuring
AIPROXY_API_ENDPOINTandAIPROXY_API_TOKENfor Azure OpenAI, initiate a test call, and verify that the returned result has no format errors and matches the expected model response. - Adjust the
temperaturevalue to the 0.2–0.4 range, generate multiple test documents, and verify that the content remains strict with no overly divergent statements.
The values provided are common starting points and should be evaluated against the samples used in the respective deployment.
Question material comes from public community discussions. Configuration values are common starting points and should be measured against your own samples. Verified on 2026-09-28.