This page’s configuration instructions are based on FastGPT v4.17.0.
What Data for This Category Looks Like
Typical data for this category originates from process integration, equipment, and yield departments in wafer fabs. It includes process flow cards, batch run records, equipment recipes, SPC charts, exception handling sheets, defect maps, and yield reports. Data sources include MES systems, equipment FDC modules, defect detection and metrology devices, wafer map files, process recipe libraries, and experimental datasets. Update frequencies align with individual wafer batch production cycles, real-time parameter pushes, and daily summary statistics. Fields include nanoscale process parameters, batch numbers, yield values, and defect spatial distribution coordinates, among others.
Constraints Imposed During Deployment and Upgrade
Multiple data sources and varying update frequencies require adapting access rules for multiple industrial data sources during deployment. Configure stable storage and cache resources to support real-time data synchronization. The large document volume and metadata complexity from extended process chains require limiting the knowledge base context window to avoid overload. Deployments in cleanroom environments must adhere to intranet access restrictions. Preset metadata extraction rules tied to wafer batch information to ensure Q&A results are accurately linked to corresponding production batches. Synchronize local configurations and image versions during upgrades to prevent industrial data synchronization interruptions caused by configuration mismatches.
How to Configure Settings
| Configuration Item | Recommended Value | Rationale |
|---|---|---|
DB_MAX_LINK | 30–50 connections | Adapt to concurrent query requests for industrial data, prevent query failures caused by connection pool exhaustion |
MONGODB_URI | Industrial-grade authenticated cluster connection string | Store wafer batch records, exception handling sheets, and yield reports, ensure data persistence and access security |
REDIS_URL | Factory intranet port 6379 address | Cache real-time updated SPC charts and defect map data, reduce latency for high-frequency queries |
FE_DOMAIN | Cleanroom intranet domain name | Adapt to factory intranet access restrictions, prevent unauthorized access from external networks |
FILE_TOKEN_KEY | Randomly generated 32-bit encryption key | Encrypt access tokens for process documents, prevent unauthorized personnel from accessing sensitive production data |
STORAGE_S3_ENDPOINT | Factory intranet S3 storage address, reserve 1000 GB of storage space | Store large-volume industrial documents such as wafer maps and defect maps, ensure data traceability |
Three Common Configuration Mistakes
- Symptom: All configured models and application entries are lost after a Windows device restart. Cause: The FastGPT data directory was not mounted to a persistent storage partition. System restarts clear configurations and data stored in temporary directories.
- Symptom: New environment variables do not take effect after a Docker Compose upgrade. Cause: Local deployment Compose files were not updated synchronously. Pulling the latest image directly leads to a mismatch between configurations and image versions.
- Symptom: Token consumption statistics for conversation responses are abnormally high. Cause: The maximum loading length of the knowledge base context was not limited. Each query loads an excessive amount of wafer process document data.
How to Verify Successful Configuration
- Access the intranet address corresponding to
FE_DOMAIN, confirm the FastGPT console loads normally, with no domain resolution or certificate errors. - Run the
redis-cli pingcommand to connect to the cache service corresponding toREDIS_URL, confirm a PONG response is returned. - Upload a wafer process flow card document to trigger the vector import process. Check the database corresponding to
MONGODB_URIfor metadata records containing batch numbers. - Submit a query targeting a specific process parameter, confirm the returned results include metadata fields from the corresponding document, with no abnormal truncation.
The values provided here are common starting points, and should be evaluated against the reader’s own samples.